Skip to content

Commit

Permalink
Merge docs from main
Browse files Browse the repository at this point in the history
  • Loading branch information
sagerpascal committed Sep 29, 2023
1 parent d054e3f commit 3f57910
Show file tree
Hide file tree
Showing 99 changed files with 23 additions and 712 deletions.
Binary file removed docs/_static/results/01_desired_result.png
Binary file not shown.
Binary file removed docs/_static/results/02_test_data.png
Binary file not shown.
Binary file removed docs/_static/results/02_train_data.png
Binary file not shown.
Binary file removed docs/_static/results/03_01.mp4
Binary file not shown.
Binary file removed docs/_static/results/03_02.mp4
Binary file not shown.
Binary file removed docs/_static/results/03_03.mp4
Binary file not shown.
Binary file removed docs/_static/results/03_04.mp4
Binary file not shown.
Binary file removed docs/_static/results/04_01.mp4
Binary file not shown.
Binary file removed docs/_static/results/04_01_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/04_02.mp4
Binary file not shown.
Binary file removed docs/_static/results/04_02_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/04_03.mp4
Binary file not shown.
Binary file removed docs/_static/results/04_03_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/04_04.mp4
Binary file not shown.
Binary file removed docs/_static/results/04_04_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/05_01.mp4
Binary file not shown.
Binary file removed docs/_static/results/05_03.mp4
Binary file not shown.
Binary file removed docs/_static/results/05_04.mp4
Binary file not shown.
Binary file removed docs/_static/results/06_01.mp4
Binary file not shown.
Binary file removed docs/_static/results/06_01_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/06_02_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/06_03.mp4
Binary file not shown.
Binary file removed docs/_static/results/06_03_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/06_04.mp4
Binary file not shown.
Binary file removed docs/_static/results/06_04_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/07_01.mp4
Binary file not shown.
Binary file removed docs/_static/results/07_03.mp4
Binary file not shown.
Binary file removed docs/_static/results/07_04.mp4
Binary file not shown.
Binary file removed docs/_static/results/08_01.mp4
Binary file not shown.
Binary file removed docs/_static/results/08_01_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/08_02.mp4
Binary file not shown.
Binary file removed docs/_static/results/08_02_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/08_03.mp4
Binary file not shown.
Binary file removed docs/_static/results/08_03_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/08_04.mp4
Binary file not shown.
Binary file removed docs/_static/results/08_04_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/09_01.mp4
Binary file not shown.
Binary file removed docs/_static/results/09_03.mp4
Binary file not shown.
Binary file removed docs/_static/results/09_04.mp4
Binary file not shown.
Binary file removed docs/_static/results/10_01.mp4
Binary file not shown.
Binary file removed docs/_static/results/10_01_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/10_02.mp4
Binary file not shown.
Binary file removed docs/_static/results/10_02_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/10_03.mp4
Binary file not shown.
Binary file removed docs/_static/results/10_03_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/10_04.mp4
Binary file not shown.
Binary file removed docs/_static/results/10_04_V2.mp4
Binary file not shown.
Binary file removed docs/_static/results/230601/1_1.png
Binary file not shown.
Binary file removed docs/_static/results/230601/1_2.png
Binary file not shown.
Binary file removed docs/_static/results/230601/1_3.png
Binary file not shown.
Binary file removed docs/_static/results/230601/1_4.png
Binary file not shown.
Binary file removed docs/_static/results/230601/1_5.png
Binary file not shown.
Binary file removed docs/_static/results/230601/1_wta.mp4
Binary file not shown.
Binary file removed docs/_static/results/230601/1_wta.png
Binary file not shown.
Binary file removed docs/_static/results/230601/2_1.jpg
Binary file not shown.
Binary file removed docs/_static/results/230601/2_2.png
Binary file not shown.
Binary file removed docs/_static/results/230601/3_1.jpg
Binary file not shown.
Binary file removed docs/_static/results/230608/1_1.png
Binary file not shown.
Binary file removed docs/_static/results/230608/1_2.png
Binary file not shown.
Binary file removed docs/_static/results/230608/1_3.png
Binary file not shown.
Binary file removed docs/_static/results/230608/2_1.png
Binary file not shown.
Binary file removed docs/_static/results/230608/2_2.png
Binary file not shown.
Binary file removed docs/_static/results/230608/2_3.png
Binary file not shown.
Binary file removed docs/_static/results/230617/mit_L2_unterbrochen.png
Binary file not shown.
Binary file removed docs/_static/results/230617/no_l2.png
Binary file not shown.
Binary file not shown.
Binary file removed docs/_static/results/230617/with_l2.png
Binary file not shown.
Binary file removed docs/_static/results/230620/1_weights_fe.png
Binary file not shown.
Binary file removed docs/_static/results/230620/2_output_fe.png
Binary file not shown.
Binary file removed docs/_static/results/230620/3_lateral_weights.png
Binary file not shown.
Binary file removed docs/_static/results/230620/4_activations.png
Diff not rendered.
Binary file removed docs/_static/results/230623/00.png
Diff not rendered.
Diff not rendered.
Binary file removed docs/_static/results/230623/11.png
Diff not rendered.
Binary file removed docs/_static/results/230623/12.png
Diff not rendered.
Binary file removed docs/_static/results/230623/13.png
Diff not rendered.
Binary file removed docs/_static/results/230623/14.png
Diff not rendered.
Binary file removed docs/_static/results/230623/15.png
Diff not rendered.
Binary file removed docs/_static/results/230623/16.png
Diff not rendered.
Binary file removed docs/_static/results/230623/1_neg_corr.png
Diff not rendered.
Binary file removed docs/_static/results/230623/2_neg_corr_weights.png
Diff not rendered.
Binary file removed docs/_static/results/230623/3.png
Diff not rendered.
Binary file removed docs/_static/results/230623/4.png
Diff not rendered.
Binary file removed docs/_static/results/230623/5.png
Diff not rendered.
Binary file removed docs/_static/results/230623/6.png
Diff not rendered.
Binary file removed docs/_static/results/230623/7.png
Diff not rendered.
Binary file removed docs/_static/results/230623/8.png
Diff not rendered.
Binary file removed docs/_static/results/msc_thesis-v0-9-3.pdf
Binary file not shown.
Binary file not shown.
2 changes: 1 addition & 1 deletion docs/conf.py
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@
author = 'Pascal Sager'

# The full version, including alpha/beta/rc tags
release = '00.00.02'
release = '01.00.00'

# -- General configuration ---------------------------------------------------
# https://www.sphinx-doc.org/en/master/usage/configuration.html#general-configuration
Expand Down
28 changes: 7 additions & 21 deletions docs/index.rst
Original file line number Diff line number Diff line change
Expand Up @@ -13,30 +13,16 @@ Please find the abstract, the full thesis, and videos presenting the results bel
:maxdepth: 1
:caption: Contents

presentation
thesis
results/final_results

---

.. raw:: html

**Download the** :download:`thesis <_static/results/msc_thesis-Pascal_Sager-V1_0.pdf>` **as PDF.**


Abstract
--------
In the past decade, deep learning has established itself as state-of-the-art technology in various automatic image analysis tasks.
Despite impressive results, this technology has several limitations, notably its limited robustness to noise, constrained transformation invariance during object recognition and reliance on a substantial amount of training data.
Conversely, the human brain does not suffer from these limitations due to its non-sequential processing of extracted image features and its ability to perceive visual scenes holistically, i.e. interpret it as more than the sum of its part, as outlined by Gestalt psychology.
This capability stems from the brain's ability to establish internal consistency between each connected cell pair through self-organisation and localised learning, i.e. a consensus is achieved across all features through mutual cell support. This mechanism solves the problem of ``early commitment'' inherent in deep networks as they rely on a global error correction algorithm to establish consistency at a single point between prediction and teaching signal.
This thesis builds upon these insights and proposes a novel image-processing framework inspired by the human brain's functionality.
Accordingly, a significant part of this thesis is devoted to identifying and interpreting neuroscientific findings.
These findings are analysed and translated into a computational framework, thereby linking each model component to the corresponding biological mechanism.
The framework consists of three components: The sensor system *S0*, responsible for extracting low-level features from the images; the feature-building stage *S1*, which uses lateral (intra-layer) connections to form neuron groups, so-called net fragments, fostering mutual support to stabilise known patterns; the prototype stage *S2*, which maps the formed net fragments to object prototypes using projection fibres and provides feedback to *S1*.
The iterative projection process between *S1* and *S2* lasts until consistency is achieved at every point in the network, i.e. until cells and synapses have reached a stable attractor state.
While prior research has demonstrated the efficiency of projection fibres, implementing net fragments still needs to be explored.
Consequently, this thesis analyses the implementation of this component in detail and discusses it by conducting experiments with a simple dataset based on straight lines.
The experimental findings demonstrate that lateral connections trained with Hebbian learning can facilitate cell support effectively.
The network exhibits significant robustness using cell support and can deactivate up to :math:`91.7\%` of unwanted cell activity triggered by noise signals. Furthermore, lateral support can restore discontinuous lines, demonstrating the network's ability to deal with occluded objects. With a range of lateral connections of :math:`11` pixels, interruptions of up to :math:`8` pixels can be reconstructed, and with additional feedback from *S2*, even interruptions of up to :math:`20` pixels can be restored. Improving the proposed framework can potentially reduce several weaknesses of conventional neural networks in the future and is considered a promising alternative research direction.
<embed>
<p><a href="_static/results/msc_thesis-Pascal_Sager-V1_0.pdf" target="_blank">Download</a> the thesis as PDF.</p>
</embed>


5 changes: 5 additions & 0 deletions docs/presentation.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
Presentation
============

.. video:: _static/results/thesis_defence_presentation.mp4
:width: 800
50 changes: 0 additions & 50 deletions docs/results/l2_230617.rst

This file was deleted.

118 changes: 0 additions & 118 deletions docs/results/report_230601.rst

This file was deleted.

75 changes: 0 additions & 75 deletions docs/results/report_230608.rst

This file was deleted.

23 changes: 0 additions & 23 deletions docs/results/report_230620.rst

This file was deleted.

Loading

0 comments on commit 3f57910

Please sign in to comment.