msc placeholder: not afraid of assembly

[synctext]

brainstorm placeholder

Embedded systems master. Start Q2 or perhaps (possible) Literature Survey earlier.
On progress for Q1 finishing all course work+survey. Then Q2 start thesis full-time, ~nov.

Likes Rust, C coding. Completed machine learning, evolutionary algorithms (ongoing) {not deep learning}. Interested in decentralised machine learning. See our Delft/EPFL work: MoDeST: Bridging the Gap between Federated and Decentralized Learning with Decentralized Sampling. master thesis on arXiv: G-Rank: Unsupervised Continuous Learn-to-Rank for Edge Devices in a P2P Network. CODE: https://github.com/awrgold/G-Rank (Jupyter Notebook file)

Scholar Google search for anything related to "LLM Android"
DroidBot-GPT: GPT-powered UI Automation for Android. Clusion after 2 minutes of effort: nobody is combining LLM for Android yet. Homework for reader: "nanoGPT Android".

We evaluate DroidBot-GPT with a self-created dataset that contains 33 tasks collected
from 17 Android applications spanning 10 categories. It can successfully complete 39.39% of the
tasks, and the average partial completion progress is about 66.76%.

EDIT: another potential direction to explore, hardware-accelerated machine learning on Android https://towardsdatascience.com/gpu-accelerated-machine-learning-in-your-mobile-applications-using-the-android-ndk-vulkan-kompute-1e9da37b7617

EDIT2: lots of activity on federated learning and swarm learning. Enough for a "literature survey". Have a broader scope than thesis, gives a broad understanding, before doing a deep dive for thesis.

• https://deepai.org/publication/peer-to-peer-federated-learning-on-graphs
• https://deepai.org/publication/federated-learning-at-the-network-edge-when-not-all-nodes-are-created-equal
• https://deepai.org/publication/decentralized-federated-learning-a-segmented-gossip-approach
• https://deepai.org/publication/gfl-a-decentralized-federated-learning-framework-based-on-blockchain
• https://deepai.org/publication/a-blockchain-based-decentralized-federated-learning-framework-with-committee-consensus
• Sybil and security issues
  • "Towards Sybil Resilience in Decentralized Learning"
  • https://deepai.org/publication/collaborative-domain-blocking-using-federated-nlp-to-detect-malicious-domains
• Models
  • https://www.tensorflow.org/lite/models/modify/model_maker/text_searcher
  • https://github.com/google-research/google-research/tree/master/scann
• on-device machine learning
  • Edge-Adaptable Serverless Acceleration for Machine Learning IoT Applications
  • Advantages and limitations of on-device ML
• Mobile Wifi and 5G limitations for on-device learning
• literature survey examples from prior students
  • Survey on social reputation mechanisms: Someone told me I can trust you
  • Survey of stable coins {52 citations}
  • The Universal Trust Machine: A survey on the Web3 path towards enabling long term digital cooperation through decentralised trust
• Literature survey has always 1 essential taxonomy table. Taxonomy in a single glance. Systematic overview of past years of innovation. Key milestones identified. Scientific grounding, 1 or more scientific article per entry/line/milestone. Table with overview and literature, see brilliant With Honours example:

Very concrete possible thesis direction. Starting point: take the proof-of-principle implementation of a Web3 recommender with full decentralisation and monte-carlo basic machine learning. Replace the recommender with TFLite model, add hardware-acceleration, tune the memory usage toward huge trust graphs, tune algorithm to giant content databases, and make it "industrial-proof". Actual deployment to Google Play store and watch the crash reports. Starting point: "Web3Recommend: Decentralised recommendations with trust and relevance" by Delft student Rohan. His repo: https://github.com/rmadhwal/trustchain-superapp/tree/TrustedRecommendations

update: https://www.economist.com/leaders/2023/05/11/what-does-a-leaked-google-memo-reveal-about-the-future-of-ai
update2: BeyondFederated by master student Quinten:
update3: First month of thesis time is wisely spend actively {e.g. coding&reading} exporing 2 diffirent thesis direction and may end up inspired for your 3rd direction. {Decentralised federated: hardware-acceleration, tune stochastic gradient decent for infinite scalability, focus specifically on decentralised content recommendation, or learning rate and parameter tuning magic}
Update4: 2011 earliest work. Gossip Learning with Linear Models on Fully Distributed Data

[AkiSchmatzler]

book on advances of federated learning

Update: current idea is to do a cross platform federated learning tool (For Arduino, Raspberry Pi, Phones)

[synctext]

Slow exploration of thesis, booked with 3 courses. Ambition level is more phd thesis instead of master thesis 🎲
Quick search on existing code:

• Decide on the balance you want between science and engineering. New distributed machine learning methodology or full-time cross-platform engineering?
• as starting point: get TensorFlow going on your Raspberry Pi:
  • https://github.com/tertiarycourses/RaspberryPITensorflow
  • https://github.com/smirko-dev/machine-learning-rpi
  • https://github.com/Qengineering/RPi-image
    • Vision is challenging, thus only realistic thing is small incremental improvement on some OpenCV code
    • Facebook infinite scroll AI or addiction ML from Tiktok has tons of related work and libraries, you can start from cratch using federated learning {But prefer latest science: self-supervised self-organising learning}