Repolode
投稿する
ATOM00blue

ATOM00blue/machine-learning-library

A curated, Markdown-based ML knowledge base of 923 papers, lectures and articles, plus a small NumPy-only custom classifier and regressor called APRL.

GitHub で見る
スター
255
フォーク
34
未解決の Issue
0
最終コミット
2 か月前

これは何？

The repository combines two layers. The first is APRL (Adaptive Prototype Residual Learning), an experimental classifier and regressor for numeric tabular data. It pairs a global ridge model with relevance-weighted prototypes and local polynomial experts that correct residual errors, and it offers distance-aware uncertainty, missing-value handling and incremental batch updates on a NumPy-only runtime. The project presents it as an experimental composition rather than a state-of-the-art method.

The second layer is a knowledge base of about 11 million tokens. It contains 391 arXiv papers (78 full text, 313 abstract plus metadata), 474 lecture transcripts from courses such as MIT 6.S191, Stanford CS224n and CS231n, and 58 web explainers. Every file is Markdown with YAML frontmatter that records source, URL, authors and controlled topic tags. The corpus is organized into a 17-topic map and can be opened as an Obsidian vault.

Both layers are meant to be used together or separately: you can train and test APRL, read or grep the corpus, or build a retrieval system over it that cites specific papers and lectures.

主な機能

  • 01Custom APRL classifier and regressor for numeric tabular data, implemented with NumPy only.
  • 02Distance-aware uncertainty estimates and missing-value handling in the APRL model.
  • 03Corpus of 923 Markdown documents spanning papers, lecture transcripts and articles.
  • 04Every file carries YAML frontmatter with source, URL, authors and controlled topic tags.
  • 05Organized into a 17-topic map with hub pages and reading paths, usable as an Obsidian vault.

ユースケース

  • Learning ML fundamentals and frontier research through lectures and papers with consistent metadata.
  • Building a retrieval-augmented Q&A assistant that cites specific papers or lectures.
  • Experimenting with a small, inspectable tabular classifier or regressor with uncertainty estimates.
  • Using the clean corpus as a text dataset for embedding benchmarks or model fine-tuning.

トピック

#machine-learning#study-resources#dataset#nlp#llm#education#arxiv#deep-learning#corpus#transformers

よくある質問

machine-learning-library とは？
A curated, Markdown-based ML knowledge base of 923 papers, lectures and articles, plus a small NumPy-only custom classifier and regressor called APRL. The repository combines two layers. The first is APRL (Adaptive Prototype Residual Learning), an experimental classifier and regressor for numeric tabular data. It pairs a global ridge model with relevance-weighted prototypes and local polynomial experts that correct residual errors, and it offers distance-aware uncertainty, missing-value handling and incremental batch updates on a NumPy-only runtime. The project presents it as an experimental composition rather than a state-of-the-art method.
machine-learning-library は無料で使えますか？ライセンスは？
machine-learning-library は Other ライセンスのオープンソースです。商用製品で使う前にライセンス条件を確認してください。
machine-learning-library は今もメンテナンスされていますか？
半年以内に更新あり。machine-learning-library の最終コミットは 2026年8月10日 です。
machine-learning-library は何に使えますか？
Learning ML fundamentals and frontier research through lectures and papers with consistent metadata. Building a retrieval-augmented Q&A assistant that cites specific papers or lectures. Experimenting with a small, inspectable tabular classifier or regressor with uncertainty estimates. Using the clean corpus as a text dataset for embedding benchmarks or model fine-tuning.
machine-learning-library はどこで入手できますか？
ソースコードとインストール手順は GitHub にあります：https://github.com/ATOM00blue/machine-learning-library

概要はプロジェクトの README をもとに AI が生成しています。GitHub の数値は毎日更新されます。

関連プロジェクト