What it does

This repository is a curated collection of paper summaries in PDF form, covering machine learning research the author has actually sat down and read. Topics span self-supervised and contrastive learning, semi-supervised methods, video understanding, domain adaptation, explainability, and more. Each entry links to the original arXiv or conference paper alongside the author’s own notes.

The interesting bit

The value isn’t the format—it’s the curation. These are handwritten (well, typed) summaries from a single researcher’s reading path, not auto-generated abstracts. That means you get someone’s actual attempt to distill the contribution, not just the abstract restated by a language model.

Key highlights

• ~100+ papers across 10+ subfields, with a heavy tilt toward computer vision
• Self-supervised learning section is particularly deep, running from 2015 (exemplar CNNs, jigsaw puzzles) through 2020 (SimCLR-era contrastive methods)
• Video understanding coverage includes the full transformer wave: ViViT, TimeSformer, Multiscale Vision Transformers
• Each paper gets original PDF notes plus direct link to source
• Semi-supervised and domain adaptation sections include the practical benchmarks (Mean Teacher, MixMatch, UDA) that actually get used

Caveats

• No search, no tagging, no index beyond the README headings—finding a specific paper means scrolling
• PDF notes are opaque blobs in the repo; you can’t preview without downloading
• Coverage is idiosyncratic to the author’s interests; NLP is thinner than the vision sections
• Last major update timing is unclear from the README

Verdict

Worth bookmarking if you’re doing a literature review in self-supervised or video understanding and want a second opinion on what matters. Skip it if you need interactive code, searchable text, or systematic coverage—this is a personal notebook, not a database.