diff --git a/README.md b/README.md index 418b7b6c..4c77d962 100644 --- a/README.md +++ b/README.md @@ -1,6 +1,8 @@ PhD Thesis --- -My PhD Thesis as submitted on 26th October 2018. +This repository contains my PhD thesis as submitted on 26th October 2018. +Code and data for its replication is available [here](https://github.com/mxochicale/phd-thesis-code-data). + ## Citation If you use or adapt any of the files in this repository, @@ -16,17 +18,23 @@ use the following BibTeX to cite [my PhD thesis](https://github.com/mxochicale/p address = {Birmingham, United Kingdom}, Title = {Nonlinear Analyses to Quantify Movement Variability in Human-Humanoid Interaction}, type = {{PhD} dissertation as submitted}, + note = {As submitted, awaiting viva and further revision}, + doi = {10.5281/zenodo.1473140}, + url = {https://doi.org/10.5281/zenodo.1473140} } ``` ->> **NB** I will be polishing and updating this repository -while I am wait for my viva (tentatively January 2019 or February 2019). -# Issues -If you see any errors or have any questions -please create an [issue](https://github.com/mxochicale/phd-thesis/issues) +# Notes +> I will be polishing and updating this repository while I am awaiting viva +(tentatively for viva might be in January 2019 or February 2019). + + +# Contact +If you have specific questions about the content of this thesis, you can contact [Miguel Xochicale](mailto:perez.xochicale@gmail.com?subject="[PhD thesis]"). +If your question might be relevant to other people, please instead [open an issue](https://github.com/mxochicale/phd-thesis/issues). diff --git a/abstract/abstract.tex b/abstract/abstract.tex index 8358a79a..4447c4bc 100644 --- a/abstract/abstract.tex +++ b/abstract/abstract.tex @@ -5,118 +5,29 @@ %%%% 200 words %%%% \begin{abstract} -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%% One or two sentences proving a basic introduction to the field, -%%%%% comprehensible to a scientist in any discipline. Nonlinear analyses investigate the dynamics of observed time-ordered data. Such dynamics, for this thesis, are complex systems of sensorimotor variables of movement variability (MV) in the context of human-humanoid interaction. -%31 words -%movement variability (MV) and also in the mechanical -%limitations of a humanoid robot (robot with the general structure of a human). -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%% Two to three sentences of more detailed background, -%%%%% comprehensible to scientist in related disciplines. Hence, this dissertation not only explores questions such as what to quantify in MV?, or which nonlinear tools are appropriate to quantify MV?, but also how nonlinear analyses are affected with real-world time series data (e.g. nonstationary, data length limitations, sampling rate changes or noisiness). -%46 words -%We conducted two experiment in human-humanoid interaction where participants -%imitate simple horizontal and vertical arm movements of a humanoid robot. -%to test the -%Given that traditional methods in time-domain and frequency domain fail to -%detect tiny modulations in frequency or phase of time series, -%we consider a methodology from nonlinear dynamics called uniform reconstructed state space -%to quantify movement variability which essentially the dynamics of -%an unknown system can be reconstructed using one dimensional time series. -%As pointed out by Bradley et al. uniform reconstructed state space, -%if done right, can guarantee to be topologically identical to the true dynamics -%and determine dynamics invariants such as fractal dimension, Kolmogorov-Sinai -%entropy or Lyaponov exponents. -%These algorithms, however, require time series measured with costly sensors -%that provide well sampled data with little noise. -%Such requirement is generally a common problem when doing precise -%characterisation of time series using dynamic invariants, -%to which Bradley et al. proposed additional tools of -%nonlinear time series analysis for practitioners such as surrogate data, -%permutation entropy, recurrence plots and -%network characteristics for time series -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%% One sentence clearly stating the general problem being addressed by this -%%%%%% particular study. Particularly, I review nonlinear tools such as methods to determine embedding parameters, Reconstructed State Spaces (RSSs), Recurrence Plots (RPs) and Recurrence Quantification Analyses (RQA). -%25 words -%(window size length) and -%preprocessing techniques (e.g. smoothing and normalisation) -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%% One sentence summarising the main result (with the words "here we show" -%%%%%% or their equivalent) -%such as window size length, participants, sensors and levels of smoothness -%(see weaknesses and strengths of RQA in Chapters \ref{chapter5} and -%\ref{chapter6}). -%So, here we show the characterisation for time series to understand -%human movement variability in the -%context of human-humanoid imitation activities and demonstrate the potential -%of nonlinear techniques to quantify human movement variability. -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%% Two or three sentences explaining what the main results reveals in direct -%%%%%% comparison to what was thought to be the case previously, or how the main -%%%%%% results adds to previous knowledge. To my knowledge, I can conclude that, no scientific work has been reported regarding nonlinear analyses (e.g. RSSs with UTDE, RPs and RQAs) to quantify movement variability in the context of human-humanoid interaction. -%34 words -%thorough experimentation and exploration to test -%the weaknesses and robustness of such tools, Also, we created 3D surfaces of RQA values considering the variation of embedded parameters and recurrence thresholds to show that 3D surfaces of RQA ENTR might be a better approach to provide understanding on the dynamics of different characteristic of real-world time series data. -%44 words -%Specifically, we explore the reconstruction of state spaces, its recurrence plots -%and metrics of recurrence quantification analysis for -%20 participants performing repetitions of simple vertical and horizontal arm -%movements in normal and faster speed. -%We also explore the differences between wearable inertial sensors attached -%to the person and to the humanoid robot and between different axes of inertial sensors. -%With that in mind, our contribution to knowledge is in regard to the -%reliability of data from cheap wearable inertial sensors -%to analyse human movement variability in the context of human-humanoid imitation activities -%using methodologies of nonlinear dynamics. -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%% One or two sentences to put the results into a more general context. -%Such understanding and measurement of movement variability using -%cheap wearable inertial sensors lead us to have a more intuitive selection of parameters -%to reconstruct the state spaces and to create meaningful interpretations -%of the recurrence plots and the results of the metrics with recurrence quantification -%analysis. Additionally, having a better understanding of -%nonlinear dynamics tools with the use of cheap inertial sensors -%can enhance the development of better diagnostic tools for various pathologies -%which can be applied in areas of rehabilitation, entertainment or sport science. -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%%%% Two or three sentences to provide a broader perspective, -%%%%%% readily comprehensible to a scientist in any discipline, -%%%%%% may be included in the first paragraph -%%%%%% if the editor considers that the accessibility of the paper is -%%%%%% significantly enhanced by their inclusion. Under this circumstances, -%%%%%% the length of the paragraph can be up to 300 words I can foresee many areas of applications where humanoids robots can be pre-programmed with nonlinear analyses algorithms to evaluate the improvement of movement performances, to quantify and provide feedback of skill learning or to quantify movement adaptations and pathologies. -%38 words - - - \end{abstract} diff --git a/acknowledgement/acknowledgement.tex b/acknowledgement/acknowledgement.tex index 0ed7e111..b6fc3b23 100644 --- a/acknowledgement/acknowledgement.tex +++ b/acknowledgement/acknowledgement.tex @@ -1,7 +1,6 @@ % ************************** Thesis Acknowledgements ************************** \begin{acknowledgements} - I would like to acknowledge to the Mexican National Council of Science and Technology (CONACyT) that funded my curiosity-driven research which allow me to quench this endless thirst for new knowledge during my diff --git a/appendixA/appendix.tex b/appendixA/appendix.tex index d252fbd0..fa884618 100644 --- a/appendixA/appendix.tex +++ b/appendixA/appendix.tex @@ -85,37 +85,12 @@ \section{20 sample length vector.} \end{equation} %%********************************[EQUATION]************************************ -%With that in mind, the values of $\boldsymbol{X}^5_3$ in each row are the embedded -%values from $ \{ \boldsymbol{x}_n \} $. For instance, $\boldsymbol{X}[13]$ is built -%from the following values of $ \{ \boldsymbol{x}_n \} $: -%$ \{ x_1, x_4,x_{7},x_{10},x_{13} \} $. -% - -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -% E)eeeeee X) xx T)tttttt R)rrrrr A)aa -% E) X) xx T) R) rr A) aa -% E)eeeee X)xx T) R) rrr A) aa -% E) X)xx T) R) rr A)aaaaaa -% E) X) xx T) R) rr A) aa -% E)eeeeee X) xx T) R) rr A) aa -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -% It is important to note that when thinking about embedding, the down-sampled -% values of $\boldsymbol{v}_n$ are embedded into $\boldsymbol{V}[]$. -% the rows of the matrix can be viewed as down-sampled vectors by $\tau$ (three in this case). - - - \section{Time series for horizontal movement of a triaxial accelerometer.} In this example, it is considered a time series of a triaxial accelerometer (Figure~\ref{fig:acc}(C)), captured from repetitions of a horizontal trajectory (Figure~\ref{fig:acc}(A)) performed by user (Figure~\ref{fig:acc}(B)). -%The trajectory is indicated by the points \textbf{a} and \textbf{b} which also -%indicate a click sound in order to constraint the movement of the user -%so as to be consistent and synchronised (Figure~\ref{fig:acc}(A)). -%The click sound is 60 beats per minute which means that there is a click sound -%every second, so for this experiment the beat is only reproduced for 20 seconds. From Figure~\ref{fig:acc}(C)) is evidently that the $A_y(n)$ is the most affected axis of the accelerometer due to the movement's characteristics in the horizontal trajectory. @@ -213,43 +188,3 @@ \section{Time series for horizontal movement of a triaxial accelerometer.} - -% -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -% \centering -% \includegraphics[width=1.0\textwidth]{figures/appendices/appendixA/sketchs_for_trajectories/figures/trajectory1a/aAhorizontal00} -% \caption{ -% (A). Triaxial accelerometer (in red) is moved repetitively across a line -% of 251 mm from point \textbf{a} to \textbf{b} and then from -% \textbf{b} to \textbf{a}. -% The points \textbf{a} and \textbf{b} indicate when a click sound is produced. -% (B). Person's hand moving the sensor horizontally across the line. -% } -% \label{fig:acc} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------- -% -% -% -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -% \centering -% \includegraphics[width=1.0\textwidth]{figures/appendices/appendixA/outcomes/figure_second_experiment/fig_2E_v03} -% \caption{Time series for the triaxial accelerometer ($A_x(n)$, $A_y(n)$, $A_z(n)$) for -% ten repetitive horizontal movements across a line. The top time series only -% shows $A_y$ axis which corresponds to one cycle of the horizontal movement. -% The black arrows with sensors represent the movement's direction of the -% accelerometer with respect to the produced time series. -% } -% \label{fig:tsacc} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------- -% -% -% - - - - - diff --git a/appendixB/appendix.tex b/appendixB/appendix.tex index 8fdb63c8..beca5403 100644 --- a/appendixB/appendix.tex +++ b/appendixB/appendix.tex @@ -10,14 +10,8 @@ % \graphicspath{{appendixB/figs/vector/}{appendixB/figs/}} %\fi - \graphicspath{{figs/appendixB/PDF/}} -% -%\appendix -%\hypertarget{aA}{ \section*{Appendix B. Inertial Sensors}} \label{appendix:b} -% - \section{NeMEMsi IMU sensors} \label{appendix:imus} For this work, data were collected using NeMEMsi sensors \cite{Comotti2014} @@ -108,13 +102,6 @@ \subsection{Organising Data in Multidimensional Arrays} time using using \texttt{finddelay()} and \texttt{alignsignals()} in \MATLAB. -%The use of \texttt{alignsignals()} is useful when the data is relatively clean -%that means that when data was noisy the alignment were not even close -%when two signals were quite similar. Therefore, I decided to -%program my own \texttt{alignsignalsMX()} to use the synchronised -%data but with different length. -%Scripts in \R were used for postprocessing the data. - \subsection{Data Synchornisation} To find the delay between two two sensors that were attached to the same place of the body parts, a function called \texttt{finddelayMX()} was created. diff --git a/appendixC/README.md b/appendixC/README.md index 2ba1378a..d17c751b 100644 --- a/appendixC/README.md +++ b/appendixC/README.md @@ -2,9 +2,14 @@ #todo -* [ ] fill info for AC + + +# sorted + +* [x] fill info for Appendix C Wed 12 Sep 12:21:21 BST 2018 +sorted: Sun 28 Oct 20:58:41 GMT 2018 diff --git a/appendixD/README.md b/appendixD/README.md index 262a98ea..82463905 100644 --- a/appendixD/README.md +++ b/appendixD/README.md @@ -1,10 +1,23 @@ #todo +* [ ] Organise and balance the description of the results. + +Sun 28 Oct 21:02:55 GMT 2018 -* [ ] add figures for timeseries, RSSs, RPs, and RQA +* [ ] Refer each of the extra information in this appendix + with the description given in the chapters + +Sun 28 Oct 21:02:58 GMT 2018 + + + +# sorted + +* [x] add figures for timeseries, RSSs, RPs, and RQA Fri 14 Sep 12:01:30 BST 2018 +sorted: 26 October 2018 diff --git a/appendixE/README.md b/appendixE/README.md index 262a98ea..be7c7782 100644 --- a/appendixE/README.md +++ b/appendixE/README.md @@ -2,8 +2,21 @@ #todo -* [ ] add figures for timeseries, RSSs, RPs, and RQA + +* [ ] Organise and balance the description of the results. + +Sun 28 Oct 21:02:55 GMT 2018 + +* [ ] Refer each of the extra information in this appendix + with the description given in the chapters + +Sun 28 Oct 21:02:58 GMT 2018 + + +# sorted +* [x] add figures for timeseries, RSSs, RPs, and RQA Fri 14 Sep 12:01:30 BST 2018 +sorted: 26 October 2018 diff --git a/appendixE/appendix.tex b/appendixE/appendix.tex index 08c1d3ce..809680c6 100644 --- a/appendixE/appendix.tex +++ b/appendixE/appendix.tex @@ -491,18 +491,7 @@ \section{RPs} \label{appendix:e:rps} \newpage \section{RQAs} \label{appendix:e:rpas} \subsection{REC values} -%It can be seen in Figs~\ref{fig:rec_aH} and \ref{fig:rec_aV} -%that REC values, representing the \% of black dots in the RPs, -%are more spread for HN than HF movements with time -%series coming from HS01 sensor. -%In contrast, REC values appear to be constant and present little variation -%for both HN and HF movements with time series from the sensor attached -%to the humanoid robot RS01. -%With regard to the increase of smoothness of time series -%(sg0zmuvGyroZ, sg1zmuvGyroZ and sg2zmuvGyroZ), REC values present little -%variation as the smoothness is increasing for time series from HS01 and -%REC values more similar as the smoothness is increasing for data from RS01. -% + REC values, representing the \% of black dots in the RPs, are shown in Figs. \ref{fig:rec_aH} and \ref{fig:rec_aV}. diff --git a/chapter1/README.md b/chapter1/README.md index a8015fd7..5ec11eb5 100644 --- a/chapter1/README.md +++ b/chapter1/README.md @@ -1,15 +1,14 @@ -* [ ] Update IMAGE for Structure of the thesis +* [x] Update IMAGE for Structure of the thesis added: Fri 7 Sep 13:49:41 BST 2018 +sorted: 24 October 2018 - - -* [ ] restate the research questions with the style of \cite{pincus1991}: +* [ ] research questions can be restated using the following similar \citep{pincus1991}: ``` The purpose of this paper is to give a preliminary mathematical diff --git a/chapter1/chapter.tex b/chapter1/chapter.tex index 6381b156..b6470958 100644 --- a/chapter1/chapter.tex +++ b/chapter1/chapter.tex @@ -77,21 +77,6 @@ \section{Background} techniques and the interpretation of the results for RSSs, RPs and metrics of RQA. -%Also \cite{bradley2015} reviewed the use of Recurrence Plots (RP) and -%Recurrence Quantification Analysis (RQA), both of which show the -%recurrences of a given $n$-dimensional system -%in a two-dimensional map of black/white dots, which help -%to have a more intuitive meaning of the time series. -%Probably, the most important results of \cite{iwanski1998} is that -%RQA are quantitatively and qualitatively independent of embedding -%dimension which was also experimentally verified. -%However, the estimation of embedding parameters and finding the right -%parameters to perform RQA is still an open problem. -% - - - - \section{Movement Variability} @@ -118,8 +103,6 @@ \section{Movement Variability} \cite{wagner2012} show how movement variability based on statistical analysis varies with skill for three levels of throwing techniques (low-skilled, skilled, and high-skilled). -%\cite{seifert2011} modelled movement variability using hierarchical clustering -%analysis for competitive and recreational swimmers. Therefore, \cite{bartlett2007} concluded that movement variability is ubiquitous across sports (javelin throwing, basketball shooting or running). Another interesting example is that movement variability can be considered @@ -144,7 +127,6 @@ \section{Movement Variability} variability either as a functional role in human movement for "coordination change and flexibility to adapt" in different environments \citep[p. 238]{bartlett2007} or -%movement variability is considered as an exploration and exploitation of body parts in the "perceptual-motor workspace" \citep{wu2014, herzfeld2014}. @@ -171,18 +153,9 @@ \subsection{Modelling Human Movement Variability} as system of differential equations for the neuro-musculoskeletal control system where motion variations can occur because of "perturbations of initial states of the skeletal system", -%for instance where right knee initial angles can produce another type -%of movement", perturbations of "muscular or neural subsystems ", -%where, for instance, initial conditions for certain muscles in -%the right leg can increase the "initial muscle activity" or "external torques and forces acting on the skeletal system" -%where for instance gust of wind might perturb the skeletal system. \citep[p. 13]{hatze1986}. -%For which any injury, pain sensation effect of fatigue are considered for -%"variation of muscular parameters, perturbation of sensory neurons", -%and finally any changes int the motor program will create -%"variation of the motor program." According to \cite{hatze1986} motion variability can be caused by (i) direct consequences of adaptive learning process, and (ii) random fluctuations which are the result of stochastic processes @@ -195,13 +168,6 @@ \subsection{Modelling Human Movement Variability} Hence, \cite{hatze1986} proposed the use of entropy as a global quantifier for motion variability and concluded that any movement deviation of a body joint may be the result of deterministic and stochastic causes. -%\cite{hatze1986} It is also important to note that -%experimental work of measuring -%running cycles, Hatzel1986 can quantify the variability between four -%cycles of running where the initial cycle has the largest (60m) -%then it decreased and stay stable until (1600m) and then again -%increased at the final phase. - @@ -332,13 +298,6 @@ \subsection{Modelling Human Movement Variability} measures of magnitude that limited the evaluation of the whole trajectories as structures of movement variability in human body activities. -%\cite{stergiou2011} -%pointed out that nonlinear measurement tools revealed -%that is not the magnitude -%that is important but the structure of movement variability. -%that helps -%to understand human-perceptual-motor functioning which led us to -%the next section of measurements of variability. Therefore, for this thesis, it is important to note that even with the proposed models for movement variability \citep{hatze1986, preatoni2010, preatoni2013, muller2004, seifert2011} @@ -347,16 +306,6 @@ \subsection{Modelling Human Movement Variability} tools when using real-world data that is commonly noisy, deterministic, stochastic or nonstationary \citep{newell1998}. A further review of nonlinear analyses is presented in Chapter \ref{chapter2}. -%\citealt[p. ?]{newell1998} stated that movement variability -%can be considered as -%"an emergent property of determinism, stochastic, and even singular processes -%in an evolving nonstationary dynamical system". -%as explained by \cite{newell1993} -%where variability is part of the exploitation of the workspace. - - - - \subsection{Movement Variability in Human-Humanoid Interaction} Movement variability in the context of human-humanoid interaction has been @@ -371,13 +320,9 @@ \subsection{Movement Variability in Human-Humanoid Interaction} Hence, \cite{gorer2013} with only graphical visualisation of joint angles trajectories extracted from the pose estimation of a kinect sensor, -%was presented for the movement performance in order to stated that only one subject out of eight fail to imitate the gestures correctly. Additionally, -%to the limitations of mapping human movements to a humanoid robot -%due to the differences in their degrees of freedom which were compensated -%with auditory feedback, \cite{gorer2013} surveyed participants using a 5-point Likert scale about the positive and negative effect, flow, immersion and challenge of the human-robot @@ -387,16 +332,6 @@ \subsection{Movement Variability in Human-Humanoid Interaction} However, the small sample size and somewhat naive analysis of data in the study makes it difficult to generalise these findings. -%\cite{gorer2013} pointed out that three out of the four DOF for -%NAO arm are for the shoulder -%and one for the elbow with its a limitation for the production -%of direct mapping -%from human arm join angles. -%Skeleton join angles collected using an a Kinect (RGB-D camera) -%mounted on Nao's head. -%It was reported that -%from the eight participants of which only one did not imitated -%the gesture correctly. Another example is the work of \cite{guneysu2014} who conducted experiments with children for upper @@ -431,20 +366,6 @@ \subsection{Movement Variability in Human-Humanoid Interaction} similarity error and recall measures with the ICC metric are not completely reliable since they did not model well complex movements. -%\cite{guneysu2014} used a moving average filter with a window size of -%50 frames to smooth the data and eliminate unnecessary fluctuations -%of local minima and maxima. -%Then peak detection of smoothened and unsmoothed data were performed to find -%alignment of motion directions. -%Evaluation of five therapist using Intraclass correlation coefficient. -%Following the work of Ranatunga et al. which use DTW for the evaluation -%of similarly of movements performance between robot and person, -%\cite{guneysu2014} implement DTW with a window size of three frames. -%\cite{guneysu2014} evaluated physiotherapist's evaluation using intraclass -%correlation coefficient based on pooled variance within subjects and -%variance of the trait between subjects, however the original formula -%proposed in [http://www.john-uebersax.com/stat/icc.htm] -%were modified as true values were not known. Recently, \cite{guneysu2015} presented an improvement of their previous research where less complex movements, from four physiotherapists performing five actions, were analysed: opening a door with a key, @@ -460,10 +381,6 @@ \subsection{Movement Variability in Human-Humanoid Interaction} time series from each of the participants which performed the movements at different frequencies and therefore with different data length (see Fig. 10 in \cite{guneysu2015} for further details). -%Additionally, different the structures of time series were presented from -%each of the physiotherapist, where for instance, Therapist 2 performed the -%activity at half amplitude compared to the others. -%reported quaternions data collected with two inertial sensors Movement variability in the context of human-humanoid interaction has also been @@ -483,10 +400,6 @@ \subsection{Movement Variability in Human-Humanoid Interaction} the trajectory of dance with a self-propelled robot were the closest to the trajectory of a dancer. Additionally, \cite{tsuchida2013} only applied traditional statistics (e.g. ANOVA) to characterise dance movements. -%considered a choreography of three elements -%of interseciton (front to back), approaching, -%and parallel translations. - @@ -523,33 +436,8 @@ \subsection{Movement Variability in Human-Humanoid Interaction} accordance with human aesthetics) \citep{peng2015}, it is important to note that one of the methodologies to create robotic dance motions is the use of chaotic dynamics -% With that in mind, creating robotics dances requires a methodology which consider initial conditions to generate movements that are neither deterministic nor stochastic. -%\cite{peng2015} -%imentined that imitation in human dance motions -%to do imitation, one can do it by using task model -%where the learning from oberation technique is occupied -%however -%due to the mecahincal and hardwrader software limitation of dance -%robots -%the robots is only able to imitate -%"the essential parts in human dance motions" -%since persons that create the dance patterns have different -%antropphmorphic carahcteristicts. - - -%%\subsection{Movement imitation} -%\cite{ijspeert2002} presented various scenarios for applications of humanoid robotics in -%rehabilitation where, for example, a robot can supervise rehabilitation exercises -%in stroke patients from demonstrations of professional therapist, -%demonstrating the motion with a robot, evaluating the performance of the patient and suggesting -%and demonstrating corrections. - -%Similarly, \cite{ijspeert2002} proposed the system of nonlinear -%differential equations which form a control policy for imitation of reaching -%movements, comparing trajectories of two-dimensional single-stroke patterns, -%and learning tennis swings with a humanoid robot. Although, movement variability in the context of human-humanoid interaction has not been directly investigated in recent years, it can be noted @@ -617,16 +505,9 @@ \section{Research questions} and different characteristics of time series (window length size, smoothness and structure)? -%\item How sensitive or robust are RQA metrics when quantifying MV? \item Additionally, what are the weaknesses and strengths of RQA metrics when quantifying MV? -%\item How much smoothing of the raw signal is appropriate in order to -% capture the nature of the variability? -%\item Is it fine to smooth raw time series for the quantification of MV? - -%\item What effect does smoothing the raw time series data have -% on the quantification of MV using these techniques? \item How the smoothing of raw time series affects the nonlinear analyses when quantifying MV? diff --git a/chapter2/README.md b/chapter2/README.md index 0d8d08a6..88c04ff4 100644 --- a/chapter2/README.md +++ b/chapter2/README.md @@ -1,4 +1,4 @@ - +% * find the page for ``` diff --git a/chapter2/chapter.tex b/chapter2/chapter.tex index 736a3b9e..8d5cfb34 100644 --- a/chapter2/chapter.tex +++ b/chapter2/chapter.tex @@ -32,13 +32,6 @@ \section{Introduction} nonlinear tools with real-world data will be reviewed. -%\cite{stergiou2006} -% mentioned that the reduction or increase of chaotic -%temporal representations is related to a decline of -%"healthy flexibility associated with behavioral rigidity and inability -%to adapt to stress placed in the human body." -%%Before going any further with nonlinear analysis, we have to -% \section{Fundamentals of time-series analysis} Biosignals from living systems can typically be nonstationary, nonlinear, @@ -88,9 +81,6 @@ \subsection{Deterministic chaotic time series} stochastic origin". Therefore, it can be concluded that time series for human body movement are neither independent nor stochastic but deterministic chaotic \citep{stergiou2006, harbourne2009, stergiou2011}. -%\subsubsection{Lorenz systems. A deterministic chaos system} -%replicate 3.3 of -%\citep{klonowski2007} \section{Quantifying Movement Variability with Nonlinear Dynamics} @@ -110,22 +100,12 @@ \subsection{Introduction} out that conventional statistics (e.g. standard deviation, coefficient of variation, intra-class correlation coefficient) only quantify the overall variability. -%\cite{preatoni2010, preatoni2013} pointed out that subtle changes in the -%neuro-muscular-skeletal system are caused by influences of environmental -%changes, training procedures or latent pathologies. Hence, measuring such -%variables with conventional statistics (e.g. standard deviation, -%coefficient of variation, intra-class correlation coefficient) is only -%for overall variability. Also, \cite{stergiou2011} stated that statistical tools (e.g. mean, standard deviation and range) are a measure of centrality, meaning such metrics are compared around a central point. Similarly, \cite{coffey2011} pointed out that the use of means and standard deviations led to reduction of data and information is therefore discarded. -%Additionally, \cite[p. 24]{goldberger2002b} -%stated that "no single statistical measure can be used to assess the complexity of -%physiologic systems" which is an illustration of the limitations of -% Additionally, one can apply frequency-domain tools to quantify movement @@ -142,30 +122,12 @@ \subsection{Introduction} \cite{klonowski2002, klonowski2007, klonowski2009} stated that frequency-domain tools require stationary data, otherwise using other type of data might create misleading results. -%One example is the decompositon of FFT into sine function -%that for instace fail to repsent a 12 hz signal with a modulated amplotude -%into a two freuqnecy of 11 and 13 hz and the main frequency of 12Hz dissaperis. -%%%%MAKE A STRONGER STATEMENT FOR THE FOLLOWING PARAGRAPH -%In contrast, \cite{preatoni2013} mentioned that Fourier basis approach -%may be appropriate for periodic signals while wavelet analysis may be for -%noisy data which contains informative spikes. - -%Recently, \cite{preatoni2013} investigated that movement variability is -%considered as a compensation of noise in the neuro-musculo-skeletal system -%and the exploration of different strategies of movements to find the most -%appropriate pattern for the actual task. -%Such compensation of noise and adaptation of movements cannot be -%quantified entirely with the use of conventional approaches for which -%non only the use of entropy measures (SampEn and ApEn) but -%Lyapunov exponent \cite{abarbanel1993, smith2010}. -% Therefore, applying either statistical tools or frequency-domain tools to quantify movement variability might create misleading results, specially when dealing with signals that are deterministic chaotic \citep{amato1992, dingwell2000, dingwell2007, miller2006}, considering -%With that in mind, \cite{preatoni2010, preatoni2013} stated that the subtle changes in the neuro-muscular-skeletal system are caused by influences of environmental changes, training or latent pathologies \citep{preatoni2010, preatoni2013} @@ -174,7 +136,6 @@ \subsection{Introduction} Hence, \cite{stergiou2011, preatoni2010, caballero2014} highlighted that movement variability can be better described and quantified with different nonlinear dynamics tools such as: -%correlation dimension largest Lyapunov exponent \citep{bruijn2009, donker2007, kurz2010b, yang2011}, fractal analysis \citep{delignleres2003}, @@ -192,112 +153,9 @@ \subsection{Introduction} hausdorff200} and Recurrence Quantification Analysis (RQA) \citep{zbilut1992, trulla1996, marwan2008}. -%(for applications of the tools, see \cite{caballero2014}). -%\cite{caballero2014} reviewed different entropy measures and its application -%in human movevent variability. -%For example, -%(Smith,Teulier, Sansom, Stergiou and Ulrich, 2011), -%mental fatigue (Liu,Zhang and Zheng, 2010), -%or changes in intracranial pressure -%(Hornero, Aboy, Abásolo, McNames and Goldstein, 2005). -%the problems with ApEn is the dpendency wht tiem series length for which, -%in 2000, Richman and Moorman proposed Sample Entropy which has been applied -%to quantify postural control (Menayo, Encarnación, Gea and Marcos, 2014), -%or -%"to find differences between schizophrenia and depression" -%(Hauge, Berle, Oedegaard, Holsten and Fasmer, 2011). -%Then in 2007, Chen et al. develop Fuzzy Entropy which has less -%depency to tdata lenght and offer more robutnsess to noise. -%FuzzyEn has been used to qunatify muscle fatique -%(Xie, Guo and Zheng, 2010) -%to qunatify the problems in satanding balance tasks -%(Barbado et al.,2012). -%Multiscale Entropy Costa, Goldberger and Peng (2002) -%Permutation Entropy Vakharia et al. (2014), -%Bandt and Pompe (2002). - - -%RQA is applied for postural fluctations or -%heart rate varialibyt that measure the regularity of time series -%\cite{caballero2014}, -% -% - -%\subsection{Measures of Variability} -%%Measuring movement variability represent also a challenge where for instance -%%traditional approaches in statistics or frequency domain tend to fail when -%%measuring different types and sources of variability. -% -%\cite{hatze1986} proposed a measure of dispersion to quantify the deviation -%of motion from a certain reference using the Fourier series. In this approach, -%deviations are from angular coordinates (radians) and linear coordinates -%(meters) which made them an unacceptable fusion of variables. -% -%Hence, \cite{hatze1986} proposed the use of transentropy as a global quantifier -%for motion variability which is able to measure dispersion considering that any -%movement deviation on a body join may be the result of deterministic and -%stochastic causes. -%Also, \cite{hatze1986} pointed out that transentropy, as a mesuremente of -%motion variabilty, is fundamental to compute other metrics such as -%average transentropy, weighted global transentropy or time transentropy. -%%\cite{hatze1986} It is also important to note that experimental work of measuring -%%running cycles, Hatzel1986 can quantify the variability between four -%%cycles of running where the initial cycle has the largest (60m) -%%then it decreased and stay stable until (1600m) and then again -%%increased at the final phase. -% -%\cite{vaillancourt2001} pointed out that it is rare for frequency and amplitude -%to differ in postural tremor of patients with Parkinson's disease -%but differences in time-dependent structures are apparent, and associated with -%a change of regularity of postural tremor. -%Therefore, \cite{vaillancourt2001} considered appreciate entropy (ApEn) -%to quantify such regularity in time-dependent structures. -%%Entropy metrics (Approximate Entropy ApEn, Sample Entropy SamEn) quantify the -%%regularity of time series either for kinematic or kinetic measure and therefore -%%the increase of regularly means that there is a decrease in the complexity of -%%the system that produce the time series therefore such decrease in complexity -%%is associated with pathological conditions \cite{vaillancourt2001}. -% -% -%\cite{preatoni2010} pointed out that subtle changes in the -%neuro-musculo-skeletal system are caused by influences of environmental -%changes, training procedures or latent pathologies. Measuring such -%variables with conventional statistics (e.g. standard deviation, -%coefficient of variation, intra-class correlation coefficient) is -%only for overall variability. Therefore, using nonlinear dynamics tools -%such as sample entropy (SampEn) and approximate entropy (ApEn) can help -%to analyse the deterministic and stochastic origin of movement variability. -%%using SampEn with original sources of ... and surrogate data where time series maintain -%%large-scale structures like periodicity, mean, variance and spectrum and -%%eliminate small-scale structures like chaotic, linear/nonlinear-determinism. -%%It is hence confirmed experimentally that movement variability is not noise -%%but the information concerning with regards to the nuero-musculo-skeletal system. -%%That is because SamEn after surrogation had an increase from 16\% to 59\%, -%%suggesting that the time series is not the outcome of a random process \cite{preatoni2010}. -%Recently, \cite{preatoni2013} investigated that movement variability is -%considered as a compensation of noise in the neuro-musculo-skeletal system -%and the exploration of different strategies of movements to find the most -%appropriate pattern for the actual task. -%Such compensation of noise and adaptation of movements cannot be -%quantified entirely with the use of conventional approaches for which -%non only the use of entropy measures (SampEn and ApEn) but -%Lyapunov exponent \cite{abarbanel1993, smith2010}. -%%reviewed methods of nonlinear dynamics such as -%%entropy measures as one of the alternative tools compared to the traditional -%%ones to investigate the nature of movement variability in elite athletes. -%%Research on quantifying pathologies with nonlinear dynamics has been done, -%%however, very little work were reported concerning movement variability in sport -%%science due to limited availability of data. -% -% -% - -%\subsection{What to measure in movement variability (MV)?, -%how to measure MV? and which tools are appropriate to measure MV?} - Having so many nonlinear tools to measure movement variability (MV) led \citealt[p. 67]{caballero2014} to raise the following question: "Is there a best tool to measure variability?" which leads us to ask @@ -329,22 +187,6 @@ \subsection{What to quantify in MV?} \label{what_to_measure_with_MV} chaotic systems, while lesser complexity of movement is characterised either as noisy systems or periodic systems (having either low amounts of predictability or hight amounts of predictability) \citep{stergiou2006}. -%FUSE PREVIOUS PARAGRAPH WITH THE FOLLOWING -%Similarly, in the neurobiology field, find the problem of assosiating random -%molecules of gas or regular organisation of molecues of cristals with low complexity -%but at the same time associated them with a level or regularity, -%for which \cite{tononi1996, tononi1998} proposed a statistical mesuare based on the deviation -%form the independence (mutual information) to to capture the regularities amongt subtes of systems. -%Hence, \cite{stergiou2006} proposed a model that relates -%health and motor learning based on the work of \cite{tononi1996, tononi1998}. -%In the model for -%\cite{stergiou2006} stated that greater amounts of complexity are related to -%rich behavioral states which are therefore associated with chaotic structures. -%In constrast, lesser amounts of complexity are associated with both -%random or periodic which are also related to the amount -%of predictability. Therefore, random and noisy systems are associated -%with low predictability, while periodic, repeatable and rigid behaviors -%are associated with high amounts of predictability. Therefore, with the works of \cite{vaillancourt2002, vaillancourt2003} and \cite{stergiou2006}, @@ -360,11 +202,6 @@ \subsection{Which nonlinear tools are appropriate to quantify MV?} where complexity and predictability variables of a system can characterise and quantify movement variability, it is important to find, to understand and to apply the right tools that measure such variables. -%With that in mind, \cite[p. 67]{caballero2014} raised an important question -%regarding the quantification of movement variability: "Is there a best tool -%to measure variability?". To the best of our knowledge, the answer is no. -%However, let us dig in further into the literature and provide -%state-of-the-art references that support our answer. Originally, \cite{pincus1991, pincus1995} proposed Approximate Entropy (ApEn) to quantify regularity of time series. @@ -379,30 +216,15 @@ \subsection{Which nonlinear tools are appropriate to quantify MV?} Multiscale Entropy (MSE) which computes SampEn of consecutive coarse-grained time series of the original time series defined by the scale factor, $\tau$. -%where the length of each time series is divided With MSE algorithm, \citep{costa2002} noted that pathology dynamics for time series of heartbeat intervals -%e.g. "increase of regularity and decrease of variability or -%increase of variability due to loss of correlation properties", are associated with reduction of complexity. Therefore, \citealt[p. 3]{costa2002} concluded that physiological complexity is associated with the adaptive capacity of the organism, disease states and aging which "may be defined by a sustained breakdown of long-term correlations and loss of information". -%Although, for large scales healthy vs pathology signals can be -%distinguishable, the pathological signals overall and become -%indistinguishable. Essentially, entropy measures (AppEn and SampEn), quantify regularity and complexity of time series \citep{preatoni2013}. -%For instance, Approximate Entropy (AppEn) values are in a range -%between 0 and 2. -%AppEn values closer to 0 are associated with time series of greater -%periodicity and regularity (e.g. sine wave) and -%AppEn values near to 2 are associated with irregularity -%(e.g. random time series) \citep{miller2006}. -%However, \cite{caballero2014} stated that is not clear how -%\cite{goldberger2002b} and \cite{vaillancourt2002} applied entropy metrics -%to analyse movement complexity. However, \cite{goldberger1996} mentioned that the increase of irregularity in time series is not synonymous of increase with physiological complexity. Similarly, an increase of ApEn or SampEn, "implying increase of irregularity @@ -420,29 +242,12 @@ \subsection{Which nonlinear tools are appropriate to quantify MV?} Hence, \cite{goldberger2002b, vaillancourt2002, costa2002} concluded that ApEn and SampEn do not necessary show the right representation of what they intend to measure. -%Additionally, \cite[p. 24]{goldberger2002b} -%stated that "no single statistical measure can be used to assess the complexity of -%physiologic systems" which is an illustration of the limitations of -%using single statiscts \citep{caballero2014}. - -%However,that Fractals are irregular but -%"not all irregular structures or erratic time series are fractal \cite{goldberger1996}. -%Hence, fractal features can also be used to assess complexity -%of movement variability \cite{holden2005, vanorden2003} Therefore, considering the previous cons of ApEn, SampEn and MSE, Detrended Fluctuation Analysis (DFA), which is based on analysing fractal features, can quantify long-term correlations of time series \citep{peng1995}. -%Considering that ApEn is a regularity statistic, -%the increase of ApEn when destroying fractal and nonlinear properties -%of heartbeat time series by a randomised prodecure -%might be related to a breakdown in long-range correlations -%"or due to subtle perturbations in nonlinear control". -%of long auto-correlation for nonstationary time series -%and "avoid spurious detection of apparent long-range correlations -%that are an artifac of nonstationary time series" DFA is calculated as the root mean square fluctuation of an integrated and detrended time series and it is represented by a scaling exponent, $\alpha$, which is an indicator for roughness of time series, @@ -471,43 +276,11 @@ \subsection{Which nonlinear tools are appropriate to quantify MV?} -%"movement trajectories evolve in a more confined region -%through their phase-space \cite[p. 89]{wijnants2009}. -%", also SampEn drops as practice is increasing -%which indicate lower-dimensional organisation of coordinate structures -%\cite[p. 89]{wijnants2009}. - - - - -% -%\cite{higuchi1988} introduced a method to -%compute the fractal dimensionality for non-periodic and irregular time series -%and test its robustness against other methods. -% -%Recently, \cite{klonowski2007} -%using higuchi method the complexity of a time series can be used in different -%scenarios of depth of anesthesia, bright light therapy, postural analysis, etc. -% -%Also, another of the advantes of Higuchi's method pointed out by -%\cite{klonowski2002, klonowski2007, klonowski2009} -%is that fractal can be computed only in time domain wotuhgotuih -%comptuing a satate spacewhich si offent computataiton -%and requries expreitse. -%Huguchi method is robust to noise signasl \cite{klonowski2002} -% -%However, \cite{klonowski2002, klonowski2007, klonowski2009} -%it is highlighthed that fractal dimension computed by higuchi's mehtod -%is different than the fractal dimension computed in teh state space representtation. -%%MORE DETAILS! Another tool to measure variability is the largest Lyapunov exponent (LyE) which is used to "quantify the rate at which nearby trajectories converge or diverge" \citep[p. 85]{stergiou2016b}. -%\cite{caballero2014} stated that local dynamic stability is defined -%as a metric of sensitivity to small perturpations with are generally -%measured with LyE. For instance, "LyE from a stable system with little to no divergence will be zero (e.g. sine wave)" and "LyE for an unstable system that has highest amount of divergence will be positive and relative hight in value @@ -515,11 +288,6 @@ \subsection{Which nonlinear tools are appropriate to quantify MV?} LyE is in between the two of the previous extremes (LyE$\approx0.1$) \cite[p. 2874]{miller2006}. However, LyE requires to be validated using surrogation \citep{dingwell2000, miller2006}. -%because of the time series is deterministic chaotic. -%We refer the reader to \cite{wolf1985} for more details about LyE -%and to \cite{theiler1992} for surrogation. - -%\subsection{Is there a best tools to measure variability?} Measuring human movement variability requires a combination of the @@ -542,9 +310,6 @@ \subsection{Which nonlinear tools are appropriate to quantify MV?} open question, finding the right tool to measure movement variability for an specific problem, and knowing its strengths and weakness of such tool is one of the research questions for this thesis. -%Therefore, the contribution to knowledge of this thesis is about the -%reliability of the Recurrence Quantification Analysis (RQA) metrics using -%different conditions of the time series. \section{Nonlinear analyses with real-world data} \label{nonlieaRealdata} @@ -558,20 +323,12 @@ \section{Nonlinear analyses with real-world data} \label{nonlieaRealdata} argued that there are weaknesses of different nonlinear tools regarding the characteristics of the time series such as nonstationarity, length data size, noise, sampling rate. -%which will be explained below. However, in the work of \cite{huffaker2017}, \cite{preatoni2013} and \cite{caballero2014} no further exploration of the metrics of nonlinear analyses with real-world data is presented. -%%INCORPORATE -%The lower dimension signals from biological signals are generally time series -%of one-dimension in $\mathbb{R}$ which commonly have -%high nonlinearity, complexity, and non-stationarity \citep{gomezgarcia2014}. -% - - \subsection{Nonsationarity} Nonstationarity of time series signals might create spurious increase or decrease in the metrics of nonlinear tools. @@ -601,22 +358,14 @@ \subsection{Nonsationarity} of EMD methods is still an open problem. For instance, an extension of EMD called Multivariate Empirical Mode Decomposition (MEMD) has been proposed to analyse multiple time series \citep{rehman2010, mandic2013}. -%and many applications have been presented using EMD See \citep{wu-hu2006, costa2007, daubechies2011, bonnet2014, mert2018} for applications of EMD. -%\cite{wu-hu2006} EMD in cariorespiratory sysncronisation -%\cite{costa2007} use EMD to detrend data of postural complexity in the elderly -%\cite{daubechies2011} proposed a different method with combine wavelet analysis and reallocation method to perform EMD -%\cite{bonnet2014} EMD for integreation of Human Walking Finally, one can use nonlinear tools that are unaffected by nonstationarity of time series such as Detrended Fluctuation Analys (DFA) \citep{hausdorff1995} -%, because removes local trends, \cite{hausdorff1995} %\cite{chen2002} and Recurrence Quantification Analysis (RQA) \citep{zbilut1992, trulla1996, marwan2008}. -%RQA metrics suffer from senstivitiy to the embedding parameters and -%threshodl recurrence to which can also create unrealiable reustls. However, \cite{bryce2012} reported negatives of DFA such as the introduction of uncontrolled bias, computational expensiveness and highlight that DFA cannot provide a generic protection against the nonstationarities @@ -624,24 +373,6 @@ \subsection{Nonsationarity} - - -%Derivative -%differencing is a technique to remove trends [chatfield1989] -%"the measured physical varialbel is a derivative of -%another fundamental variable" -%\cite{rangarajan2000} - - - -% -%\subsection{Nonlinear tools for nonstationary timeseries} -%Biosignals are tipically nonstationary \cite{klonowski2007, caballero2014, wijnants2009}. -% - - - - \subsection{Data length} Many of the nonlinear tools are sensitive to time series length \citep{caballero2014}. @@ -673,18 +404,6 @@ \subsection{Sampling rate} decrease of SampEn. Hence, \cite{rhea2011} concluded that signals at higher frequencies appear to be more regular due to the increase of data, therefore producing erroneous entropy results. -%\cite{rhea2011} highlighted the algorithms to compute entropy measures -%are different since ApEn and SampEn are approximations -%of the Kolmogorov-Sinai Entropy -%computing the likelihood that a template patter repeats in the time series -%while RQAEn is derived from Shannon entropy and is computing -%number of line segments of varying length in the RP. -% -%recommended to increase the collection time than -%the increase of sampling frequency to obtain large data points, -%since oversampling -%suggesting that SampEn is more robuts for shorter time series -%when colinarieties are not an issue. \cite{caballero2013} showed the robustness of SampEn and DFA tools when using different sampling rate frequencies, stating that frequencies near the dynamics of the activity create a more reliable analysis of the @@ -746,14 +465,6 @@ \subsection{Noise} nonlinear dynamics (see \citealt*{bradley2015} and references therein). -%\cite{preaotin2013} -% smoothging techinques -%if the time series are smooth and non-periodic, -%then B-splitnes may be approapriate \cite{coffey2011}. -%if the data is noisy with informative spikes then avoiding severe smoothign -%is necessary for which wavelet analysis may be appropriate. -% -%\cite{cencini2000} \section{Final remarks} diff --git a/chapter3/README.md b/chapter3/README.md index aca5f75c..b8014185 100644 --- a/chapter3/README.md +++ b/chapter3/README.md @@ -44,14 +44,14 @@ sorted: Mon 10 Sep 13:58:11 BST 2018 -* [ ] introduction +* [x] introduction %A LINK IS MISSING HERE! added: Mon 10 Sep 10:10:43 BST 2018 - +last week of october 2018 diff --git a/chapter3/chapter.tex b/chapter3/chapter.tex index e78883b0..dd1464aa 100644 --- a/chapter3/chapter.tex +++ b/chapter3/chapter.tex @@ -29,7 +29,6 @@ \section{Introduction} Henceforth, the method of state space reconstruction using a scalar time series can preserve dynamic invariants such as correlation dimension, fractal dimension, Lyaponov exponents or Kolmogorov-Sinai entropy -%and detrended fluctuation analysis \citep{bradley2015, Quintana-Duque2012, Quintana-Duque2013, Quintana-Duque2016, krakovska2015}. However, @@ -82,10 +81,6 @@ \section{State Space Reconstruction Theorem} \label{sec:rss} where $\Phi$ is a diffeomorphic map \citep{takens1981} whenever $\tau > 0$ and $m > 2d_{box}$ and $d_{box}$ is the box-counting dimension of $M$ \citep{garland2016}. -% "Given two manifolds $M$ and $N$, a differientiable map $f: M \rightarrow N$ -% is called diffeomorphic if it is one-to-one correspondence and its inverse -% $f^{-1}: N \rightarrow M$ is differientiable as well -%\cite{wiki:diffeomorphic}". Then, if $\Phi$ is an embedding of an attractor (i.e. evolving trajectories) in the reconstructed state space, a composition of functions represented with $F^t$ is induced on the reconstructed state space: @@ -165,19 +160,6 @@ \section{Uniform Time-Delay Embedding (UTDE)}\label{sec:utimedelayembedding} where $m$ is the embedding dimension, $\tau$ is the embedding delay and $ ^\intercal$ denotes the transpose. $m$ and $\tau$ are known as embedding parameters. -%Then after applying the transpose for the vectors of the delayed copies of the time series, -%$\boldsymbol{X}^{m}_{\tau}$, can be represented as -%%********************************[EQUATION]************************************ -%\begin{equation}\label{eq:etde2} -%\boldsymbol{X}_{\tau}^{m} = -%\begin{pmatrix} -% \boldsymbol{X}[ ( (m-1)\tau ) + 1 ] \\ -% \boldsymbol{X}[ ( (m-1)\tau ) + 2 ] \\ -% \vdots \\ -% \boldsymbol{X}[N-1] \\ -% \boldsymbol{X}[N] -%\end{pmatrix}. -%\end{equation} %%%********************************[EQUATION]************************************ The matrix dimension of $ \boldsymbol{X}_{\tau}^{m} $ is defined by $N-(m-1)\tau$ rows and $m$ columns and $N-(m-1)\tau$ defines the length of @@ -684,59 +666,8 @@ \subsection{Measures of RP based on diagonal lines} % % RATIO = diagP$DET / REC % - -%%%%%%%%%%%%%%%%% -%%(3er variable) -%$D_{max}$ is the longest diagonal line in the RP, defined as -%%%********************************[EQUATION]************************************ -%\begin{equation} -%D_{max}= \operatorname*{arg,\max}_{l} H_{D}(l). -%\end{equation} -%%%********************************[EQUATION]************************************ -%$D_{max}$ is an indicator of the divergence of trajectory segments. -%The smaller $D_{max}$ is, the more divergent the trajectories are -%\citep{marwan2007, marwan2015}. -%According to \cite{iwanski1998}, $D_{max}$ is also related to the inverse -%of the largest positive Lyapunov exponent, where for example periodic signals -%tend to have very long lines, as opposed to the chaotic time series where -%parallel lines are shorter. -%%#' \item \emph{Lmax}: Length of the longest diagonal line. -%%calculateDiagonalParameters = function(ntakens, numberRecurrencePoints, -%% lmin = 2, lDiagonalHistogram, -%% recurrence_rate_vector, maxDistanceMD) { -%% #pick the penultimate -%% Lmax = tail(which(lDiagonalHistogram > 0), 2)[1] -%% if (is.na(Lmax) || Lmax == ntakens) { -%% Lmax = 0 -%% } -% -% -%%%%%%%%%%%%%%% -%%(X variable) -%The average diagonal line length is defined as -%%%********************************[EQUATION]************************************ -%\begin{equation} -% \langle D \rangle = \frac{ \sum^{N}_{l=d_{min}} l H_D(l) } -% { \sum^{N}_{l=d_{min}} H_D(l)}, -%\end{equation} -%%%********************************[EQUATION]************************************ -%and it is the average time that two segments of the trajectory are close to -%each other. $\langle D \rangle$ can be interpreted as a measure for -%determinism (predictability) of the system \citep{marwan2007, marwan2015}. -%%#' \item \emph{Lmean}: Mean length of the diagonal lines. The main diagonal is -%%#' not taken into account. -%% -%% calculateDiagonalParameters = function(ntakens, numberRecurrencePoints, -%% lmin = 2, lDiagonalHistogram, -%% recurrence_rate_vector, maxDistanceMD) { -%% DET = num / numberRecurrencePoints -%% Lmean = num / sum(lDiagonalHistogram[lmin:ntakens]) -%% aux.index = lmin:(ntakens - 1) -%% LmeanWithoutMain = ( -%% sum((aux.index) * lDiagonalHistogram[aux.index]) / -%% sum(lDiagonalHistogram[aux.index]) -%% ) -% + + %%%%%%%%%%%%%%% @@ -765,91 +696,6 @@ \subsection{Measures of RP based on diagonal lines} % ENTR = -sum(pl[diff_0] * log(pl[diff_0])) - -%%(5th variable) -%Trend (TND) "is a linear regression coefficient over the recurrence point -%density of the diagonals parallel to the LOI" \citep[p. 16]{marwan2015} -%and is defined as -%%%********************************[EQUATION]************************************ -%\begin{equation} -%TND= \frac{ \sum^{\tilde{N} }_{i=1} (1- \tilde{N} /2 )( RR_i - \langle RR_i \rangle ) }{ \sum^{\tilde{N} }_{i=1} (i-\tilde{N} /2)^2 }. -%\end{equation} -%%%********************************[EQUATION]************************************ -%Trend value "provides information about the stationarity versus nonstationarity -%in the process" \citep[p. 16]{marwan2015}. TNT values near to zero represent -%quasi-stationary dynamics and TNT values far from zero represent nonstationary -%dynamics that reveal the "drift in the dynamics" \cite[p. 16]{marwan2015}. -%TNT measures how quickly the RP pales away from the main diagonal -%\citep{iwanski1998}. -%%#' \item \emph{TREND}: Trend of the number of recurrent points depending on the -%%#' distance to the main diagonal -%%calculateDiagonalParameters = function(ntakens, numberRecurrencePoints, -%% lmin = 2, lDiagonalHistogram, -%% recurrence_rate_vector, maxDistanceMD) { -%% -%% # use only recurrent points with a distance to the main diagonal < maxDistance -%% recurrence_rate_vector = recurrence_rate_vector[1:maxDistanceMD] -%% mrrv = mean(recurrence_rate_vector) -%% #auxiliar vector for the linear regresion: It is related to the general regression -%% #formula xi-mean(x) -%% auxiliarVector = (1:maxDistanceMD - (maxDistanceMD + 1) / 2) -%% auxiliarVector2 = auxiliarVector * auxiliarVector -%% # divide by two because we are having into account just one side of the main diag -%% num = sum(auxiliarVector * ((recurrence_rate_vector - mrrv) / 2)) -%% den = sum(auxiliarVector2) -%% TREND = num / den -% -% -%\subsection{Measures of RP based on vertical lines} -%The previous RQA metrics are based on length, number and distribution of -%diagonal lines. However, patterns for horizontal and vertical lines can also -%be quantified. The following are some examples. -% -%%%%%%%%%%%%%%%% -%%(6th variable) -%Laminarity (LAM) computes the percentage of recurrence points in vertical lines -%which is analogous to the DET variable \cite{marwan2015}. -%%#' \item \emph{LAM}: Percentage of recurrent points that form vertical lines. -%%calculateVerticalParameters = function(ntakens, numberRecurrencePoints, -%% vmin = 2, verticalLinesHistogram) { -%% #begin parameter computations -%% num = sum((vmin:ntakens) * verticalLinesHistogram[vmin:ntakens]) -%% LAM = num / numberRecurrencePoints -% -%%%%%%%%%%%%%%%%% -%%(7th variable) -%Trapping time (TT) variable computes the average length of vertical lines. -%"TT contains information about the amount and length of vertical -%structures in the RP" which represent "the mean time the system will" -%stay "at a specific time" \citep[p. 17]{marwan2015}. -% -%%%%%%%%%%%%%%%% -%%(8th variable) -%The maximal length of the vertical structures $V_{max}$ represents -%"the longest vertical line in the RP" which is analogous to $D_{max}$. -%According to Marwan et al. \cite[p. 17]{marwan2015} the dynamical -%interpretation of this variable is not clear but only as a relationship -%with "singular states in which the system is stuck in a holding pattern -%inscribing rectangles in the RP". -%%#' \item \emph{Vmax}: Longest vertical line. -%%#' \item \emph{Vmean}: Average length of the vertical lines. This parameter is -%%#' also referred to as the Trapping time. -%% -%%calculateVerticalParameters = function(ntakens, numberRecurrencePoints, -%% vmin = 2, verticalLinesHistogram) { -%% #begin parameter computations -%% num = sum((vmin:ntakens) * verticalLinesHistogram[vmin:ntakens]) -%% Vmean = num / sum(verticalLinesHistogram[vmin:ntakens]) -%% if (is.nan(Vmean)) { -%% Vmean = 0 -%% } -%% #pick the penultimate -%% histogramWithoutZeros = which(verticalLinesHistogram > 0) -%% if (length(histogramWithoutZeros) > 0) { -%% Vmax = tail(histogramWithoutZeros, 1) -%% } else { -%% Vmax = 0 -%% } \subsection{Some weaknesses and strengths of RP and RQA.} \label{sec:ws_rqa} diff --git a/chapter4/README.md b/chapter4/README.md index 2b5ff42e..5942d2a9 100644 --- a/chapter4/README.md +++ b/chapter4/README.md @@ -16,6 +16,8 @@ Appendix \ref{appendix:c}. ``` Wed 12 Sep 13:32:17 BST 2018 +sorted Last week of October 2018 + diff --git a/chapter4/chapter.tex b/chapter4/chapter.tex index 30b48d87..41c024ad 100644 --- a/chapter4/chapter.tex +++ b/chapter4/chapter.tex @@ -1,22 +1,19 @@ %******************************************************************************* -%****************************** Fourth Chapter *********************************%******************************************************************************* +%****************************** Fourth Chapter ********************************* +%******************************************************************************* \chapter{Experiments} \label{chapter4} % **************************** Define Graphics Path ************************** %\ifpdf -% \graphicspath{{chapter5/figs/raster/}{chapter5/figs/PDF/}{chapter5/figs/}} +% \graphicspath{{chapter4/figs/raster/}{chapter4/figs/PDF/}{chapter4/figs/}} %\else -% \graphicspath{{chapter5/figs/vector/}{chapter5/figs/}} +% \graphicspath{{chapter4/figs/vector/}{chapter4/figs/}} %\fi % \graphicspath{{figs/chapter4/PDF/}} -%%**************************** %Broad Purpose ********************************* -%\section*{Summary and broad purpose of the chapter} -%* How long (number of words)? - \section{Aims} We design two experiments for human-image interaction (HII) and diff --git a/chapter5/README.md b/chapter5/README.md index 39afe66c..687f41f5 100644 --- a/chapter5/README.md +++ b/chapter5/README.md @@ -5,6 +5,10 @@ # todo +* [ ] create other ranges of 3D RQA surfaces +Sun 28 Oct 23:05:28 GMT 2018 + + * [ ] add legends to RQA surfaces Tue 9 Oct 14:15:44 BST 2018 diff --git a/chapter5/chapter.tex b/chapter5/chapter.tex index c5c242c9..8d859a45 100644 --- a/chapter5/chapter.tex +++ b/chapter5/chapter.tex @@ -6,19 +6,15 @@ \chapter{Quantifying Human-Image Imitation Activities} \label{chapter5} %% **************************** Define Graphics Path ************************** %\ifpdf -% \graphicspath{{chapter6/figs/raster/}{chapter6/figs/PDF/}{chapter6/figs/}} +% \graphicspath{{chapter5/figs/raster/}{chapter5/figs/PDF/}{chapter5/figs/}} %\else -% \graphicspath{{chapter6/figs/vector/}{chapter6/figs/}} +% \graphicspath{{chapter5/figs/vector/}{chapter5/figs/}} %\fi % \graphicspath{{figs/chapter5/PDF/}} -%%**************************** %Broad Purpose ******************************** -%\section*{Summary and broad purpose of the chapter} -%* How long (number of words)? - \section{Introduction} In this chapter, we present the results of the experiments of @@ -34,9 +30,6 @@ \section{Time series} following an image while not hearing a beat (nb) and hearing a beat (wb). Also, three levels of smoothness of normalised time series are presented (sg0, sg1 and sg2). -%based on two different -%Savitzky-Golay filter lengths (29 and 159) with the same polynomial -%degree of 5 using \texttt{sgolay(p,n,m)} \citep{Rsignal}. The remained time series are presented in Appendix \ref{appendix:d:ts}. %%---------------------------------(FIGURE)------------------------------------- \begin{figure} @@ -563,52 +556,6 @@ \subsection*{ENTR values} -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -%\centering -%\includegraphics[width=0.6\textwidth]{RATIO} -% \caption{ -% {\bf Box plots for RATIO values.} -% RATIO values, representing dynamic transitions, -% 6 participants performing (A) horizontal arm movements -% and (B) vertical arm movements for sensors HS01, HS02 and -% three smoothed-normalised axis -% of GyroZ (sg0, sg1 and sg2). -% RATIO values were computed with -% embedding parameters $m=9$, $\tau=6$ and recurrence threshold -% $\epsilon=1$. -% R code to reproduce the figure is available from \cite{hwum2018}. -% } -% \label{fig:RATIO} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------- -%% -% - - -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -%\centering -%\includegraphics[width=0.6\textwidth]{ENTR} -% \caption{ -% {\bf Box plots for ENTR values.} -% ENTR values (representing the complexity of the structure of time -% series) for -% 6 participants performing (A) horizontal arm movements -% and (B) vertical arm movements for sensors HS01, HS02 and -% three smoothed-normalised axis -% of GyroZ (sg0, sg1 and sg2). -% ENTR values were computed with -% embedding parameters $m=9$, $\tau=6$ and recurrence threshold -% $\epsilon=1$. -% R code to reproduce the figure is available from \cite{hwum2018}. -% } -% \label{fig:ENTR} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------- -% -% - @@ -978,13 +925,6 @@ \subsection{Final remarks} such surfaces appear to be helpful to understand the dynamics of any kind of time series with little parametrization of nonlinear tools. For instance, -%Independently of the source of time series, surfaces for RATIO -%values are affected little by the changes of embedding values and recurrence -%thresholds with the exception for the peaks presented when embedding values -%increase for recurrence thresholds less than one. -%3D surfaces for DET values are affected slightly more than RATIO values. -%Nonetheless, REC values are affected by the window size, type of activity -%and level of smoothness. 3D surfaces of ENTR values, with only the selection of variation of range of parameters, show clearly differences in the shape of 3D surfaces irregardless of the source of the time series. Hence making 3D surfaces diff --git a/chapter6/README.md b/chapter6/README.md index 7a731cde..74778e42 100644 --- a/chapter6/README.md +++ b/chapter6/README.md @@ -1,5 +1,13 @@ +# todo + + + +* [ ] create other ranges of 3D RQA surfaces +Sun 28 Oct 23:05:28 GMT 2018 + + * [ ] add legends to RQA surfaces diff --git a/chapter6/chapter.tex b/chapter6/chapter.tex index ce68cfa8..16e890a0 100644 --- a/chapter6/chapter.tex +++ b/chapter6/chapter.tex @@ -6,9 +6,9 @@ \chapter{Quantifying Human-Humanoid Imitation Activities} \label{chapter6} %% **************************** Define Graphics Path ************************** %\ifpdf -% \graphicspath{{chapter7/figs/raster/}{chapter7/figs/PDF/}{chapter7/figs/}} +% \graphicspath{{chapter6/figs/raster/}{chapter6/figs/PDF/}{chapter6/figs/}} %\else -% \graphicspath{{chapter7/figs/vector/}{chapter7/figs/}} +% \graphicspath{{chapter6/figs/vector/}{chapter6/figs/}} %\fi \graphicspath{{figs/chapter6/PDF/}} @@ -158,31 +158,6 @@ \section{Minimum Embedding Parameters} -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -%\centering -%\includegraphics[width=1.0\textwidth]{AMI} -% \caption{ -% {\bf First minimum AMI values.} -% Box plots of first minimum AMI values are for -% Horizontal Normal (HN), Horizontal Faster (HF), -% Vertical Normal (VN) and Vertical Faster (VF) -% with sensor attached to participants (HS01) and -% sensor attached to robot (RS01). -% First inimum AMI values are for twenty participants -% ($p01$ to $p20$) with three smoothed signals -% (sg0zmuvGyroZ (sg0) , sg1zmuvGyroZ (sg1) and sg2zmuvGyroZ (sg2)) -% and window length of 10-sec (500 samples). -% R code to reproduce the figure is available -% from \cite{hwum2018}. -% } -% \label{fig:AMI} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------ -% - - - \newpage \subsection{Average minimum embedding parameters} Following the Section \ref{sec:overall_minMT} to compute the overall average @@ -198,7 +173,6 @@ \subsection{Average minimum embedding parameters} Recurrence Quantification Analysis (RQA) metrics for human-humanoid activities. -%\newpage \section{Reconstructed state spaces with UTDE} Considering Section \ref{sec:rsswithUTDE} and time series for participant $p01$ (Figs \ref{fig:tsH}, \ref{fig:tsV}) the reconstructed state spaces @@ -397,27 +371,8 @@ \subsection*{REC values} (see the incremental changes of mean values (rhombus)). See Figs~\ref{fig:rec_aH} and \ref{fig:rec_aV} in appendix \ref{appendix:e:ep} for more details about individual REC values for each participant. -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -%\centering -%\includegraphics[width=0.6\textwidth]{rec-bp} -% \caption{ -% {\bf Box plots for REC values.} -% REC values (representing \% of black dots in the RPs) for -% 20 participants performing HN, HF, VN and VF movements -% with sensors HS01, RS01 and three smoothed-normalised -% time series (sg0, sg1 and sg2). -% REC values were computed with -% embedding parameters $m=6$, $\tau=8$ and $\epsilon=1$. -% R code to reproduce the figure is available from \cite{hwum2018}. -% } -% \label{fig:REC} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------ -% -%\newpage \subsection*{DET values} Figs \ref{fig:RQABP}(B) illustrate DET values, representing predictability and organisation of the RPs, @@ -425,28 +380,7 @@ \subsection*{DET values} for type of movement, level of smoothness or type of sensor. See Figs~\ref{fig:det_aH} and \ref{fig:det_aV} in appendix \ref{appendix:e:ep} for more details about individual DET values for each participant. -% - -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -%\centering -%\includegraphics[width=0.6\textwidth]{det-bp} -% \caption{ -% {\bf Box plots for DET values.} -% DET values (representing predictability and organisation of the RPs) -% for 20 participants performing HN, HF, VN and VF movements -% with sensors HS01, RS01 and three smoothed-normalised -% time series (sg0, sg1 and sg2). -% DET values were computed with -% embedding parameters $m=6$, $\tau=8$ and $\epsilon=1$. -% R code to reproduce the figure is available from \cite{hwum2018}. -% } -% \label{fig:DET} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------ -% -%\newpage \subsection*{RATIO values} Figs \ref{fig:RQABP}(C) present RATIO values, representing dynamic transitions, for horizontal and vertical movements. @@ -459,27 +393,9 @@ \subsection*{RATIO values} See Figs~\ref{fig:ratio_aH} and \ref{fig:ratio_aV} in appendix \ref{appendix:e:ep} for more details about individual RATIO values for each participant. -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -%\centering -%\includegraphics[width=0.6\textwidth]{ratio-bp} -% \caption{ -% {\bf Box plots for RATIO values.} -% RATIO (representing dynamic transitions) for -% 20 participants performing HN, HF, VN and VF movements -% with sensors HS01, RS01 and three smoothed-normalised -% time series (sg0, sg1 and sg2). -% RATIO values were computed with -% embedding parameters $m=6$, $\tau=8$ and $\epsilon=1$. -% R code to reproduce the figure is available from \cite{hwum2018}. -% } -% \label{fig:RATIO} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------ -% -%\newpage + \subsection*{ENTR values} Fig. \ref{fig:RQABP}(D) show ENTR values, representing the complexity of the structure the time series, for both horizontal and vertical movements. @@ -492,25 +408,6 @@ \subsection*{ENTR values} See Figs~\ref{fig:entr_aH} and \ref{fig:entr_aV} in appendix \ref{appendix:e:ep} for more details about individual ENTR values for each participant. -%%%---------------------------------(FIGURE)------------------------------------- -%\begin{figure} -%\centering -%\includegraphics[width=0.6\textwidth]{entr-bp} -% \caption{ -% {\bf Box plots for ENTR values.} -% ENTR values (representing the complexity of the deterministic -% structure in time series) for -% 20 participants performing HN, HF, VN and VF movements -% with sensors HS01, RS01 and three smoothed-normalised -% time series (sg0, sg1 and sg2). -% ENTR values were computed with -% embedding parameters $m=6$, $\tau=8$ and $\epsilon=1$. -% R code to reproduce the figure is available from \cite{hwum2018}. -% } -% \label{fig:ENTR} -%\end{figure} -%%%---------------------------------(FIGURE)------------------------------------ -% %%---------------------------------(FIGURE)------------------------------------- @@ -536,37 +433,6 @@ \subsection*{ENTR values} - - - - - - - - - - - - -%% \subsection{Effects of different parameters in the computation of -%% different metrics of RQA.} -%Then we only select the axis AccY and GyroZ as being the axis which show -%better consistency in the patters for all the posibilities int he time series. -%That was doing visual inspection. It also worthwhile to note that those axis -%represent the majoy energy amothn other axis for both sensors. -%The selection of recurrecne threshold is 1 for HF activites, -%however this should be changed to -%for HN activies. -%With that we also can observe the effect of the seletion -%of recurrence threshold for different actibities is crucial -%to have meaninign values in the metrics of RQA. -%% added: Tue 19 Jun 2018 - -%\section{RQA metrics with different embedding parameters, -%recurrence thresholds, window lengths, levels of smoothness, and -%time series structures.} -% - \newpage \section{Weaknesses and strengths of RQA} Considering the Section \ref{sec:ws_rqa} regarding @@ -759,7 +625,6 @@ \subsection{Smoothness} -%\newpage \subsection{Participants} Figs ~\ref{fig:topo_participants} illustrate 3D surfaces of RQA metrics for three participants. @@ -791,8 +656,6 @@ \subsection{Participants} \newpage \subsection{Final remarks} -% You need to say whether the results are expected and -% whether different methods agree or contradict each other. Generally, it can be noted the changes for RQA metrics are evident with both the increase of embedding dimension parameters and the recurrence threshold for different structures, window size, @@ -819,14 +682,4 @@ \subsection{Final remarks} -%It is also important to highlight that the patterns in the 3D -%surfaces of the RQA metrics (REC, DET, RADIO and ENTR) -%(Fig \ref{fig:topo_rqas}) are certainly similar to its corresponded -%metrics for the different characteristics of the time series -%(Figs. \ref{fig:topo_sa_hs01}, \ref{fig:topo_sa_rs01}, -%\ref{fig:topo_windows}, \ref{fig:topo_smoothness}, -%\ref{fig:topo_participants}). - - - diff --git a/chapter7/README.md b/chapter7/README.md index ac064cb8..03cf762a 100644 --- a/chapter7/README.md +++ b/chapter7/README.md @@ -6,17 +6,20 @@ ``` With that in mind, we hypothesise that the differences in perception of -velocities are related to the background of each person, for example, -persons who have receive musical training in their infancy -are more aware of their body movement. %[add reference] +velocities are related to the background of each person, +personality traits or even to some kind of movement experience +(in music or sports) that make them more aware of their body +movements. %[add referenceS] ``` - Thu 20 Sep 06:54:11 BST 2018 + + + * [ ] To have a better understanding of RP, one requires to read the marvan2008: ``` diff --git a/chapter7/chapter.tex b/chapter7/chapter.tex index ff3b8a65..56eb7d75 100644 --- a/chapter7/chapter.tex +++ b/chapter7/chapter.tex @@ -44,144 +44,6 @@ \section{Conclusions} -%Although our experiment is limited to twenty healthy right-handed -%participants of a range of mean 19.8 SD=1.39 age, -%we can conclude that thought -%Considering the process to compute RQA metrics -%(embedding parameters, RSSs, RPs), -%can quantify human movement variability -%considering different conditions of the time series such as -%window size length, levels of smoothness, axis, sensors and activities. -% -% -%%Such understanding and measurement of movement variability using -%%cheap wearable inertial sensors lead us to have a more intuitive -%%selection of parameters -%%to reconstruct the state spaces and to create meaningful interpretations -%%of the recurrence plots and the results of the metrics with -%%recurrence quantification -%%analysis. -% -%Similarly, such differences in time series created differences in each of the -%RQA metrics, for instance, RATIO and ENTR are helpful to distinguish -%differences in any of the categories of the time series (sensor, activity, -%level of smoothness and number of participant), however for certain time series -%(data from the sensor attached to the robot) seemed to have little variations -%between each of the participants. The latter phenomena was in a way -%evidently as robot degrees of freedom did not allow it to move with a -%wide range of variability. -% -%%In this thesis, an experiment is performed in the context of human-humanoid -%%imitation activity to test nonlinear dynamics methods to quantify -%%human movement variability. -%%The presented results that illustrate the potential of nonlinear dynamics tools -%%by providing a balanced review of positive and negatives aspects of each -%%technique to quantitatively and qualitatively measure movement variability. -%% -%With regards to the visual inspection and understanding of the patters -%for reconstructed state spaces and recurrence plots, -%it can also be concluded that the performance of such tools is subjective -%since biased personal data interpretation might be provided. -%Hence, without any bias, RQA metrics (REC, DET, RATIO and ENTR) help us -%to show such differences of movement variability for difference categories -%of the time series (participants, movements, axis type, window length -%or levels of smoothness). -%Furthermore, it was noticed that each of the metrics of RQA show -%the differences but particularly the metrics of RATIO and ENTR are -%helpful to distinguish the differences in each of the categories -%of the time series. -% -%Although RPs and metrics for RQA are independent of embedding -%dimension \cite{iwanski1998}, it was found that recurrence threshold -%values can modify the results for both RPs and RQA. This thesis has -%carried out experiments on different activities in which the sensor's -%axis was representative of difference of structures in the time series. -% -% -% -% -%%We therefore collecte data series using inertial sensors and we smooth -%%the signals to see its graphical effects and also in the metrics. -%%With that in mind, embedding values to create the reconstructe state spaces -%%and theferfore its recurrence plots. By doing that, we saw that visually -%%each of the particpants in each of the conditions for movemetns and -%%levels of smootheness of the signlas are evicently differently, therefore -%%the challenge for us were to have a quantitavely method for each of the -%%difference in the patterns for RP for which we took the adnavce of the use -%%of Recurrene Qantification Analysis which show important results -%%that help us to quantify the variality between particpants and between -%%movments. We esentially noticed that each of the metrics of RQA -%%(REC, DET, RATIO and ENTR) can shown differences but specially RATIO -%%and ENTR are helpful to distinguis the differences in each of the movments. -%%%Tue 19 Jun 12:21:56 BST 2018 -% -% -%%With that in mind we did the same for the selection -%%of the right threshold for our particular problem where we first select -%%a threhold of 1 for all the signals however, these value is not -%%appropriate for both activites, as for example the HN and HF -%%show sometimes white RP which should not be the case if -%%we select a right emedding threshodl. -%%However, we have found the following problems -%%% ENT -%%For RQA the window effect is crucial for each of the metrics -%%so for example, ENT values for HN are higher and HF are lower -%%meaming that these are less comples than HN values, we believe -%%that happens because of the lnght windows. Althought, -%%these values apper to be more complex for HF, -%%the values in HN have less repetutions of tmovments -%%which is the reason to have those differences in ENT values. -%%%#added Mon 18 Jun 12:32:55 BST 2018 -%%For, RATIO values apper to have more variations across particpants -%%for which we believe that RATIO values represent a bit better -%%the variatbiltuy of imitation activitivies and also the -%%movment variaiblaituy that is created in this experiemnt. -%%%added: Mon 18 Jun 14:06:19 BST 2018 -%%and for ENTR values show little variation across partipants and these -%%are generally higher for HN than HF movmentes in each of the -%%smothed signals. We believe that ENTR values are affected -%%by the windown lenght so as to say that HN values -%%represent more complex structures than those coming from -%%HF values. -%%%added: Mon 18 Jun 14:17:27 BST 2018 -% -% -%%which is similar to the raw normalised data. Also the raw normalised -%%data can show more details information of for the movement, however -%%these are not lost as the time series is smoothed. -%%We can also conclude that finding the right balance between -%%smoothness and the raw data to capture movement variability is -%%a still a problem that has many avenues for exploration. -%% -%%other variables affect the data from the sensors such as -%%temperature of the sensor processing, variation of the sampling -%%rate which esentially are involved in the quqlity of the data. -%% -%%We only create two to four levels of smoothness using the -%%savtiktzy-golay filter with same degree for different filter lenght -%%for which we were able to see the differences in each of the -%%nonlinear dyamics tools. -%%We also propose four window size lenght to see the effects -%%in the nonlinear dynamics tools. -%%%add more conclusions about the reuslts of the changes in rss, rp and rqa -%%%as these parameters chagne. -%%\subsection{Perception of speed} -% -% -% -% -%%to understand and quantify movement variability using nonlinear dynamics tools. -%%With regard to the humanoid movements, I realise that -%%it is imporant to program the robot with using the same speed -%%and also considering movements that the robot can perform, -%%so for example when the speed is a bit higher the -%%robot's movements tend to be jerky. -%%Also create uniform speeds of each of the movments, -%%I realise the the horizontal and vertical movements in -%%normal and faster speed were different which might -%%affect the perception of people movements. -% - \subsection*{What are the effects @@ -200,32 +62,10 @@ \subsection*{What are the effects a new approach that is independent of either the type of time series or the selection of parameters. - -%With regard to RQA metrics, we observed that DET values varies little -%independently of the time series, and REC and RATIO values varied a bit -%more but not as much as ENTR values for which ENTR metrics are able -%to capture any change make in the time series. -% -%With that in mind, we can say that all begins with the selection of -%embedding parameters ... -%was our first challenge where we computed embedding parameters for -%each time series and then computed a sample mean over all time series -%in order to get two embedding parameters to compute all RSSs with -%its corresponded type of movement. -%Then we found that the quantification of variability with regard to the -%shape of the trajectories in RSSs requires more investigation since -%our original proposed method base on euclidean metric failed to -%quantify those trajectories. Specially, for trajectories -%which were not well unfolded. With that in mind, we proceed to take -%advantage of four RQA metrics (REC, DET, RATIO and ENTR) in order -%to avoid any subjective interpretations or personal bias with -%regard to the evolution of the trajectories in RSSs. \subsection*{What are the weaknesses and strengths of RQA metrics when quantifying MV?} -%We realise that using RQA metrics with fixed parameters is a partial -%view of the dynamics of the time series i After getting results from our experiments, we can state that the weaknesses of RQA are (i) the requirement of an expert(s) for @@ -246,103 +86,6 @@ \subsection*{What are the weaknesses and strengths of RQA metrics (ii) require little set up of parametrisation. -%The weaknesses and strengths of RQA metrics are -%related to the capacity of RQA metrics to provide understanding -%on the dynamics of real-world time series data. -%%It can be noted that not only the type of activity, window size length and -%%structure of the time series affects the values of RQA metrics but also -%%certain RQA metrics are better and more robust to describe the dynamics -%%of a determined type of movement. -%%related to the selection -%%of embedding parameters, recurrence thresholds and the -%% (predicability, organisation of the RPs, dynamics transitions, -%%or complexity and determinism) with is explained with more detail in the -%%following sections. -% - -%\subsubsection*{RQA metrics with fixed parameters} -%Considering that RQA metrics were computed with fixed embedding -%parameters (e.g. $m=6$ and $\tau=8$) and recurrence thresholds -%(e.g. $\epsilon=1$), we found the following for human-humanoid activities. -%REC values, representing the \% of black points -%in the RPs, were more affected with an increase of velocity -%for normal arm movements (HN and VN) than for faster velocity -%arm movements (HF and VF) with the sensor attached to the -%participants (HS01). -%Also, REC values for RS01 appear to be more -%constant than those from HS01 -%(see Fig. \ref{fig:RQABP}(A) in Chapter \ref{chapter6}). -%DET values, representing predictability and organisation in -%the RPs, present little variation in the any of the time series and -%little can be said but the effect of the increase of smoothness of -%time series which made DET values appear to be more similar and constant -%(see Fig. \ref{fig:RQABP}(B) in Chapter \ref{chapter6}). -%In contrast, RATIO values, which represent dynamic -%transitions, were more variable for arm movements performed at faster -%velocity (HF and VF) than normal velocity (HN and VN) -%with the sensor attached to the participants (HS01). -%For time series from the sensor attached to the robot (RS01), -%RATIO values for horizontal arm movements (HN, HF) -%appear to vary more than values coming from vertical arm movements (VN, VF) -%(see Fig. \ref{fig:RQABP}(C) in Chapter \ref{chapter6}). -%With that in mind, it can be said that RATIO values can represent -%better movement variability than the use of REC or DET metrics, -%particularly with their dynamics transitions of imitation activities -%in each of the conditions for time series. -%ENTR values for HN arm movements -%were higher than values for HF arm movements and ENTR values varied more -%for sensor attached to participants (HS01) -%than ENTR values for sensors of the robot (RS01) -%(see Fig. \ref{fig:RQABP}(D) in Chapter \ref{chapter6}). -% -%From Chapters \ref{chapter2} and \ref{chapter3}, it is known that the -%higher the ENTR metric is the more complex the dynamics of the movements are. -%However, ENTR values for normal velocity (HN, VN) appear to be a bit higher -%than ENTR values for faster velocity (HF, VF) -%(see Fig. \ref{fig:RQABP}(D) in Chapter \ref{chapter6}). -% -% -%For ENTR values of human-image imitation, -%there is a slight increase of ENTR values for -%movements with beat (HNwb, HFwb, VNwb, VFwb) -%against movements wiht no beat (HNnb, HFnb, VNnb, VFnb) -%(see the sample means (grey rhombus) in Figs -%\ref{fig:BPRQAH}(D) and \ref{fig:BPRQAV}(D) in Chapter \ref{chapter5}). -% -%With that in mind, we hypothesise this happens because of the structure -%the time series appear more complex for HN than HF arm movements -%(presenting less stability at normal velocity). -% -%We also explored the effect of smoothness of raw-normalised time series -%for RQA metrics where, for instance, REC and DET values -%appear to be constant. Hence, REC and DET values were little -%affected by the smoothness of time series. -%However, the effect of smoothness can be well noticed for both -%RATIO and ENTR values where a slightly but notable decrease in the amplitude -%of the values in any of the time series conditions is presented. -% -% -%%(see Section \ref{ch6:rqas} in Chapter \ref{chapter6}) -%%(see Section \ref{ch5:rqas} in Chapter \ref{chapter5}) -% -% -%\subsubsection*{RQA metrics with different parameters} -%We realised that varying embedded parameters and recurrence thresholds -%to create 3D surfaces of RQA might be a better approach to understand -%the dynamics of different characteristic of time series -%such as window size length, participants, sensors and levels of smoothness. -% -% -%In general, it can be noted that the patterns in 3D surfaces of RQA -%are sensible to the increase of embedding parameters and recurrence threshold, -%meaning that stability of RQA metrics is dependant on changes of -%embedding parameters and recurrence thresholds. -% -%%*HII -%%*HRI -% -% - \subsection*{How the smoothing of raw time series affects the nonlinear analyses when quantifying MV?} @@ -382,29 +125,8 @@ \subsection*{Inertial sensors} of displacement with respect time such as jounce, snap, crackle and pop \citep{eager2016}. -%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -%%%Thu 10 May 13:10:45 BST 2018 -%Considering the work of \cite{shoaib2016}, -%futher experimets can be permoed with the combination of linear acceleartion -%n, which is obtained by removing acceleration due to -%gravity from the accelerometer, with accelerometer and gyroscope. -%"the acceleration due to gravity is useful for differentiating static postures -%such as sitting and standing" but it is sensitive to changes in -%sensor orientation and body position. -% -%[Florentino-Liano, B.; O’Mahony, N.; Artés-Rodríguez, A. -%Human activity recognition using inertial sensors -%with invariance to sensor orientation. -%In Proceedings of the 2012 3rd IEEE International Workshop on -%Cognitive Information Processing (CIP), -%Baiona, Spain, 28–30 May 2012; pp. 1–6.] - - - \subsection*{Nonlinear analyses} -%While working with different nonlinear analyses I bumped into -%interesting areas that will be part of my future lines of research. \subsubsection*{Optimal embedding parameters} When using the method of False Nearest Neighbour \citep{Cao1997}, where @@ -435,18 +157,11 @@ \subsubsection*{Other methodologies for state space reconstruction.} (i) the nonuniform time-delay embedding methodology where the consecutive delayed copies of $\{ \boldsymbol{x}_n \} $ are not equidistant -%. Such method has been proved to create better representations -%of the dynamics of the state space to analyse quasiperiodic -%and multiple time-scale time series, and \citep{pecora2007, uzal2011, Quintana-Duque2012, Quintana-Duque2013, Quintana-Duque2016}, and (ii) uniform 2 time-delay embedding method which takes advantage of finding an embedding window instead of the traditional method of finding the embedding parameters separately \citep{gomezgarcia2014}. -%In general, uniform 2 time-delay embedding method computes $m$ with -%False Nearest Neighbour (FNN) algorithm and $\tau$ is computed as -%$\tau= d_w / (m-1)$, where $d_w$ is given by the minimisation of the -%Minimum Description Length \citep{small2004}. \subsubsection*{RQA parameters} @@ -511,10 +226,7 @@ \subsection*{Variability in perception of velocity} velocities are related to the background of each person, personality traits or even to some kind of movement experience (in music or sports) that make them more aware of their body -movements. %[add reference] -%It would also be interesting to ask participants to move in three -%different velocities without any constrain in order to capture -%the natural movements of slow, normal and faster velocity arm movements. +movements. \subsection*{A richer dataset of real-world time series} It should also be highlighted that the experiments for this thesis are @@ -540,49 +252,4 @@ \subsection*{Applications} for movement variability. -%It is also important to note that we considered the use of normalised raw -%time series from the inertial sensors, however, performing the -%reconstructed state space -%require a dimensionality recudtion using PCA where another noramlisation -%of data is performed for such dimensioanlity reduction. -%In constrast, computing RPs and RQAs require the creating of an UTDE matrix, -%however, there is no extra normalisation than the normalised -%raw data of the input -%of RPs and RQAs. -%added: Fri 20 Jul 00:48:14 BST 2018 - -% -%FUTURE WORK FOR RQA -%Similarly, Iwansky et al. \cite{iwanski1998} pointed out that -%two dissimilar RPs: -%one from the R\"{o}ssler system and the other one from -%a sine-wave signal of varying -%period have got equal values of REC (2.1\%) and near-equal values of -%DET (42.9\%, 45.8\%, respectively). Where we believe other RQA's can be more -%realiable for certain source of the time series. -%added: Fri 20 Jul 01:12:50 BST 2018 -% - - - - -%HUMAN_HUMANOID IMTAITON APPLICAITONS -%The work of \cite{guneysu2014} raised an important point of not considering -%latency of motions for velocity or symmetry of motion which can be used as -%indicators of attention deficit, boredom, or lack of perception. -%Tue 31 Jul 19:22:29 BST 2018 - - - -% Investigate \section{Group Activity in Human-Humanoid Imitation} -%added Thu 2 Aug 21:56:20 BST 2018 - - -%future work with more participants in a more -%controlled experiment -%With that in mind, we conclude that -%quantification of human-humanoid imitation activities is possible for -%participants of different ages, state of health and anthropomorphic features. -% -% diff --git a/declaration/declaration.tex b/declaration/declaration.tex index b27582c4..8def8575 100644 --- a/declaration/declaration.tex +++ b/declaration/declaration.tex @@ -2,17 +2,5 @@ \begin{declaration} -I hereby declare that except where specific reference is made to the work of -others, the contents of this dissertation are original and have not been -submitted in whole or in part for consideration for any other degree or -qualification in this, or any other university. This dissertation is my own -work and contains nothing which is the outcome of work done in collaboration -with others, except as specified in the text and Acknowledgements. This -dissertation contains fewer than 50,000 words including appendices, -bibliography, footnotes, tables and equations. - -% and has fewer than 150 figures. -% Author and date will be inserted automatically from thesis.tex \author \degreedate - \end{declaration} diff --git a/dedication/dedication.tex b/dedication/dedication.tex index 790dad35..d8a589ae 100644 --- a/dedication/dedication.tex +++ b/dedication/dedication.tex @@ -4,15 +4,8 @@ \begin{dedication} -%\begin{align} \[ \min_{G} \max_{X} F(G,X) \] -%\end{align} -%Thu 25 Oct 09:00:16 BST 2018 -%I would like to dedicate this thesis to my loving parents \dots -%whom with their love bring me to this world full of beautifulness -%but yet with many incomprehensible injustice -%Thu 30 Aug 16:15:21 BST 2018 \end{dedication} diff --git a/dependencies/README.md b/dependencies/README.md index 0c0d2bb6..2fae5d84 100644 --- a/dependencies/README.md +++ b/dependencies/README.md @@ -1,17 +1,16 @@ Package requirements --- -This thesis has been compiled in machine with Ubuntu 16.04 x64 using -the following packages: +This thesis has been compiled in a machine with +Ubuntu 16.04 x64 using the following packages: # TeX Live 2018 - Follow these instructions for LaTeX installation: [https://github.com/mxochicale/latex/tree/master/installation](https://github.com/mxochicale/latex/tree/master/installation) # inkspace 0.92.3 -for vector images +for processing vector files ``` sudo add-apt-repository --yes ppa:inkscape.dev/stable diff --git a/draft-revisions/draft02/corrections/v2.0-v2.75/corrections.pdf b/draft-revisions/draft02/corrections/v2.0-v2.75/corrections.pdf index c2260309..e7286601 100644 Binary files a/draft-revisions/draft02/corrections/v2.0-v2.75/corrections.pdf and b/draft-revisions/draft02/corrections/v2.0-v2.75/corrections.pdf differ diff --git a/draft-revisions/draft02/corrections/v2.0-v2.75/corrections.tex b/draft-revisions/draft02/corrections/v2.0-v2.75/corrections.tex index 200fc3e3..1e4b23aa 100644 --- a/draft-revisions/draft02/corrections/v2.0-v2.75/corrections.tex +++ b/draft-revisions/draft02/corrections/v2.0-v2.75/corrections.tex @@ -1,4 +1,4 @@ -\title{Corrections for thesis from draft v2.0 to draft v2.75} +\title{Corrections for thesis from draft v2.0 to draft v2.75 (pre-submission)} \author{Miguel Xochicale} \date{ \today } diff --git a/figs/README.md b/figs/README.md new file mode 100644 index 00000000..4a247332 --- /dev/null +++ b/figs/README.md @@ -0,0 +1,159 @@ +Figures (figs) +--- + + + + +# path tree + +``` +~/github/phd-thesis/figs$ tree -d +. +├── appendixA +│   ├── PDF +│   └── vector +├── appendixB +│   ├── PDF +│   └── vector +│   └── sources +├── appendixD +│   ├── PDF +│   └── vector +│   └── v00 +│   ├── ami +│   ├── cao +│   ├── rps +│   ├── rqa +│   ├── rss +│   └── timeseries +├── appendixE +│   ├── PDF +│   └── vector +│   └── v00 +│   ├── rps +│   ├── rss +│   └── timeseries +├── chapter1 +│   ├── PDF +│   └── vector +│   └── sources +├── chapter3 +│   ├── PDF +│   └── vector +├── chapter4 +│   ├── PDF +│   └── vector +│   ├── hii +│   │   ├── blender_model +│   │   └── sources +│   └── hri +├── chapter5 +│   ├── PDF +│   │   └── v00 +│   └── vector +│   ├── v00 +│   │   ├── rps +│   │   ├── rqa-topologies +│   │   ├── rss +│   │   └── timeseries +│   └── v01 +│   ├── ami +│   ├── cao +│   ├── rqa-metrics +│   └── timeseries +├── chapter6 +│   ├── PDF +│   └── vector +│   ├── ami +│   │   ├── v03 +│   │   │   └── sources +│   │   └── v04 +│   ├── cao +│   │   ├── v03 +│   │   │   └── sources +│   │   └── v04 +│   ├── caoami +│   │   └── v04 +│   ├── rps +│   │   ├── v03 +│   │   │   └── sources +│   │   ├── v04 +│   │   └── v05 +│   ├── rqa +│   │   ├── v05 +│   │   └── v06 +│   ├── rqa-topologies +│   │   └── v03 +│   │   ├── p01 +│   │   ├── p02 +│   │   └── p03 +│   └── rss +│   └── v03 +│   └── sources +└── results + ├── rqa + │   ├── hii + │   │   ├── v00 + │   │   │   ├── rp_plots + │   │   │   │   ├── H + │   │   │   │   └── V + │   │   │   ├── rqa_plots + │   │   │   │   ├── H + │   │   │   │   └── V + │   │   │   ├── rqa_topologies + │   │   │   │   ├── p01 + │   │   │   │   │   ├── H + │   │   │   │   │   ├── V + │   │   │   │   │   └── windows + │   │   │   │   ├── p04 + │   │   │   │   │   ├── H + │   │   │   │   │   └── V + │   │   │   │   └── p05 + │   │   │   │   ├── H + │   │   │   │   └── V + │   │   │   └── timeseries_plots + │   │   └── v01 + │   │   └── rqa_plots + │   └── hri + │   ├── v04 + │   │   ├── rp_plots + │   │   ├── rqa_plots + │   │   └── timeseries_plots + │   └── v05 + │   ├── rp_plots + │   │   ├── H + │   │   └── V + │   ├── rqa_plots + │   │   ├── H + │   │   └── V + │   └── timeseries_plots + ├── smoothed-timeseries + │   ├── hii + │   │   └── v00 + │   │   └── hii-sts-svg + │   └── v04 + └── utde + ├── minimum_embedding_parameters + │   ├── ami + │   │   ├── v04 + │   │   └── v05 + │   └── cao + │   ├── v04 + │   └── v05 + ├── minimum_embedding_parameters-hii + │   ├── ami + │   │   ├── v00 + │   │   └── v01 + │   └── cao + │   ├── v00 + │   └── v01 + ├── rss + │   └── v04 + │   ├── w10-dim006tau008 + │   └── w2-dim004tau005 + └── rss-hii + └── v00 + └── w10-dim009tau006 + +145 directories +``` diff --git a/logBOOK.md b/logBOOK.md index 18d93948..a42fc837 100644 --- a/logBOOK.md +++ b/logBOOK.md @@ -9,6 +9,32 @@ # LogBOOK + +* [x] cleaning repository for to commit version v1.0.0 (submission) + + * all comments were deleted for`pre-submission` (Date: Fri Oct 26 19:14:15 2018 +0100) + to the 'submission' version on Mon 29 Oct 00:11:20 GMT 2018 + + +* [ ] learn how to make difference between + + `pre-submission` (Date: Fri Oct 26 19:14:15 2018 +0100) + `submission` (Date: Sun 28 Oct 21:07:37 GMT 2018) + to see the deleted comments that can be either used + as a future work or deleted + + +added: Sun 28 Oct 21:19:22 GMT 2018 + +* [ ] clean README files and add revelent information + from the difference between + `pre-submission` (Date: Fri Oct 26 19:14:15 2018 +0100) + `submission` (Date: Sun 28 Oct 21:07:37 GMT 2018) + + + + + * [x] github commit: thesis draft 1.5 diff --git a/references/README.md b/references/README.md index 34b0b339..09074b97 100644 --- a/references/README.md +++ b/references/README.md @@ -2,6 +2,19 @@ # todo +* [ ] Use scholar R package to understand about the citations of + references in this thesis. + + + +The scholar R package provides functions to extract citation data from Google Scholar. + +* https://cran.r-project.org/web/packages/scholar/index.html +* https://github.com/jkeirstead/scholar + +added: Mon 29 Oct 00:04:07 GMT 2018 + + * [ ] Update references/references.bib with ``` diff --git a/thesis-info.tex b/thesis-info.tex index 2caa3705..58653e18 100644 --- a/thesis-info.tex +++ b/thesis-info.tex @@ -1,53 +1,9 @@ % ************************ Thesis Information & Meta-data ********************** %% The title of the thesis - -% \title{Writing your PhD thesis in \texorpdfstring{\\ \LaTeX2e}{LaTeX2e}} - -% \title{Movement Variability in the context of Human-Humanoid Imitation} - -% \title{Towards the Analysis of Movement Variability in the context of -%Human-Humanoid Imitation} - -% \title{Towards the Analysis of Movement Variability in the context of -% Human-Humanoid Imitation} %August2017 - -% \title{Automatic Analysis of Movement Variability in the Context of -%Human-Humanoid Imitation} %14December2017 - -% \title{Automatic Analysis of Movement Variability} %14December2017 - -%\title{Analysis of Movement Variability Using Nonlinear -%Dynamics Time Series} %24February2017 - -%\title{Analysis of Movement Variability Using Nonlinear Dynamics} -%Wed 20 Jun 16:22:51 BST 2018 - -%\title{Analysis of Movement Variability With Nonlinear Dynamics} -% Thu 21 Jun 09:08:19 BST 2018 - -%\title{Nonlinear Time-series Analysis for Human-robot Movement Variability} -% Thu 21 Jun 10:58:27 BST 2018 -% Title has been inspired from these two book titles: -%* Nonlinear time-series analysis revisited -% E Bradley, H Kantz - arXiv preprint arXiv:1503.07493, 2015 - arxiv.org -%* Nonlinear Analysis for Human Movement Variability Hardcover – 5 Feb 2016 -% Nicholas Stergiou (Editor) - - -%\title{Nonlinear Time-series Analysis for -%Human-Humanoid Movement Variability} -% Mon 25 Jun 09:06:14 BST 2018 - -%\title{Nonlinear Time-series Analysis for -%Movement Variability in Human-humanoid Interaction} -% Mon 3 Sep 13:05:34 BST 2018 - \title{ Nonlinear Analyses to Quantify \\ Movement Variability in \\ Human-humanoid Interaction } -% Wed 19 Sep 13:00:36 BST 2018 - %\texorpdfstring is used for PDF metadata. Usage: @@ -58,9 +14,7 @@ % \subtitle{Using the CUED template} %% The full name of the author -%\author{Miguel Angel Perez Xochicale} -%\author{Miguel P Xochicale} -\author{Miguel Xochicale} %Fri 12 Oct 15:17:39 BST 2018 +\author{Miguel Xochicale} %% Department (eg. Department of Engineering, Maths, Physics) \dept{Department of Electronic, Electrical and Systems Engineering} diff --git a/thesis.tex b/thesis.tex index 945ebabc..f7ac68c7 100644 --- a/thesis.tex +++ b/thesis.tex @@ -146,16 +146,12 @@ \include{chapter1/chapter} %Introduction \include{chapter2/chapter} %Quantifying Movement Variability \include{chapter3/chapter} %Nonlinear Analyses - \include{chapter4/chapter} %Experiments - \include{chapter5/chapter} %Quantifying Human Imitation Activities \include{chapter6/chapter} %Quantifying Human-Humanoid Imitation Activities - \include{chapter7/chapter} %Conclusion and Future Work - %%********************************** Appendices ***************************** \begin{appendices} % Using appendices environment for more functunality \include{appendixA/appendix} % Examples of Uniform TDE diff --git a/xtras/social-nets/README.md b/xtras/social-nets/README.md index 27f26cf1..14b1544b 100644 --- a/xtras/social-nets/README.md +++ b/xtras/social-nets/README.md @@ -2,21 +2,16 @@ Social Networks Messages --- -## Twitter +## Twitter / Facebook / LinkedIn -PhD thesis as submitted! -Many thanks to Chris Baber at @unibirmingham for supervising my scientific endeavours, -and to @Conacyt_MX for funding my curiosity-driven research! -GITHUB: https://github.com/mxochicale/phd-thesis -CODE: https://github.com/mxochicale/phd-thesis-code-data +My PhD thesis as submitted on 26th October 2018 at @unibirmingham! +Enjoy it! +Zenodo: https://doi.org/10.5281/zenodo.1473139 +GitHub: https://github.com/mxochicale/phd-thesis +Code and data: https://github.com/mxochicale/phd-thesis-code-data -## Facebook - - - -## Linkeind diff --git a/xtras/social-nets/main/README.md b/xtras/social-nets/main/README.md index 05790858..1b5ff1e9 100644 --- a/xtras/social-nets/main/README.md +++ b/xtras/social-nets/main/README.md @@ -5,6 +5,7 @@ genesis: Fri 26 Oct 18:38:23 BST 2018 +inkscape --export-png output/t-v01.png t-v01.svg inkscape --export-png output/t-v00.png t-v00.svg diff --git a/xtras/social-nets/main/output/IMG20181029105022.jpg b/xtras/social-nets/main/output/IMG20181029105022.jpg new file mode 100755 index 00000000..17f66d08 Binary files /dev/null and b/xtras/social-nets/main/output/IMG20181029105022.jpg differ diff --git a/xtras/social-nets/main/output/t-v01.png b/xtras/social-nets/main/output/t-v01.png new file mode 100644 index 00000000..9610eb81 Binary files /dev/null and b/xtras/social-nets/main/output/t-v01.png differ diff --git a/xtras/social-nets/main/sources/OAlogo.jpg b/xtras/social-nets/main/sources/OAlogo.jpg new file mode 100644 index 00000000..f775fd7d Binary files /dev/null and b/xtras/social-nets/main/sources/OAlogo.jpg differ diff --git a/xtras/social-nets/main/sources/README.md b/xtras/social-nets/main/sources/README.md index 4ee2a97b..b58769c3 100644 --- a/xtras/social-nets/main/sources/README.md +++ b/xtras/social-nets/main/sources/README.md @@ -24,3 +24,19 @@ convert -density 150 3drqa.pdf -quality 90 3drqa.png +## sources + +https://about.zenodo.org/static/img/logos/zenodo-gradient-round.svg +Mon 29 Oct 01:00:54 GMT 2018 + + +https://about.zenodo.org/static/img/logos/zenodo-gradient-1000.png +Mon 29 Oct 01:03:01 GMT 2018 + + +https://www.drupal.org/project/ghpages +Mon 29 Oct 01:26:18 GMT 2018 + + +https://blogs.napier.ac.uk/open-access/wp-content/uploads/sites/17/2014/11/OAlogo.jpg +Mon 29 Oct 01:31:35 GMT 2018 diff --git a/xtras/social-nets/main/sources/octocat_fluid.png b/xtras/social-nets/main/sources/octocat_fluid.png new file mode 100644 index 00000000..d2e8f551 Binary files /dev/null and b/xtras/social-nets/main/sources/octocat_fluid.png differ diff --git a/xtras/social-nets/main/sources/zenodo-gradient-1000.png b/xtras/social-nets/main/sources/zenodo-gradient-1000.png new file mode 100644 index 00000000..d84adb8e Binary files /dev/null and b/xtras/social-nets/main/sources/zenodo-gradient-1000.png differ diff --git a/xtras/social-nets/main/sources/zenodo-gradient-round.svg b/xtras/social-nets/main/sources/zenodo-gradient-round.svg new file mode 100644 index 00000000..6acd5ef1 --- /dev/null +++ b/xtras/social-nets/main/sources/zenodo-gradient-round.svg @@ -0,0 +1,68 @@ + + + + + + + + + + + diff --git a/xtras/social-nets/main/t-v01.svg b/xtras/social-nets/main/t-v01.svg new file mode 100644 index 00000000..e076cc26 --- /dev/null +++ b/xtras/social-nets/main/t-v01.svg @@ -0,0 +1,485 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + image/svg+xml + + + + + + + + + + + + + + + + + + + + + + + + Nonlinear Analyses to Quantify Movement Variability in Human-Humanoid Interaction + + + + + + + + + + + + + + + + + + Miguel Xochicale + + + + + + PhD thesis as submitted on 26th October 2018 + + + + + + + + https://doi.org/10.5281/zenodo.1473139 + https://github.com/mxochicale/phd-thesis + https://github.com/mxochicale/phd-thesis-code-data + + + +