merlihson/scientific-resources - 观星指南 Stargazers.cn

Your one-stop knowledge base for AI/ML research, learning, and reference

Explore Reviews • Learning Materials • Presentations • Quick Start

Highlights

📄 612+

Paper Reviews
_{Deep analysis of cutting-edge
AI/ML research papers}

📚 23

Learning Categories
_{From ML basics to
quantum computing}

🎯 1,496 PDFs

Resources
_{13 GB of books,
cheat sheets & more}

📅 2022-26

Coverage Period
_{Continuously updated
with latest research}

Overview
Quick Start
Paper Reviews
Learning Materials
Presentations
Metadata & Search
Automated Metadata Updates
Daily Review Automation
Collection Statistics
Repository Structure
For Researchers & Students
Contributing
License

Overview

This repository is a curated knowledge base for AI/ML researchers, students, and practitioners. It contains:

612 comprehensive paper reviews covering cutting-edge AI/ML research (2022-2026)
23 learning categories with 1,496 PDFs spanning machine learning, mathematics, algorithms, and more
14 research presentations on deep learning architectures
Searchable metadata indices for quick paper discovery
Python tooling for document processing and automation

Quick Start

Find a Specific Paper

# Search by title in the metadata index
grep -i "transformer" mike-paper-reviews-all/reviews_metadata/all_paper_titles.txt

# Or browse the CSV with paper links
cat mike-paper-reviews-all/reviews_metadata/paper_with_links.csv

Access Reviews

What you want	Where to find it
Individual review files (DOCX)	`mike-paper-reviews-all/split-reviews-docx/Review_001.docx` - `Review_612.docx`
Individual reviews (Markdown)	`mike-paper-reviews-all/split-hebrew-reviews-md/Review_001.md` - `Review_612.md`
All paper titles	`mike-paper-reviews-all/reviews_metadata/all_paper_titles.txt`
Papers with links (100% coverage)	`mike-paper-reviews-all/reviews_metadata/paper_with_links.csv`
Archived PDFs and old formats	`mike-paper-reviews-all/archive/`

Paper Reviews

`mike-paper-reviews-all/`

The core collection containing 612 individual paper reviews in multiple formats.

Type	Count	Description
Individual Reviews	1-208	Deep-dive analyses with paper links
Daily Reviews	209-612	Chronological reviews (May 2024 - Feb 2026)

Formats Available

split-hebrew-reviews-md/ - 612 Hebrew review markdown files (primary format)
split-english-reviews-md/ - 245 English review markdown files
split-reviews-docx/ - 612 DOCX source files (Review_001.docx → Review_612.docx)
reviews_metadata/ - Auto-updated metadata (100% link coverage)
archive/ - Historical PDFs and old batch documents

Research Domains Covered

Domain	Topics
Deep Learning	CNNs, RNNs, Transformers, Novel Architectures
NLP	LLMs, Text Generation, Language Understanding
Computer Vision	Image Recognition, Object Detection, Segmentation
Generative Models	GANs, VAEs, Diffusion Models, Autoregressive
Multimodal	Vision-Language Models, Cross-Modal Learning
Reinforcement Learning	Policy Learning, Decision Making
Optimization	Training Algorithms, Regularization, Efficiency

Learning Materials

`learning-materials/`

22 specialized categories with 1,496 PDFs (13 GB) of educational resources:

Core AI/ML

machine learning/ - 14 subcategories (518 PDFs)
- Broad ML materials (259)
- Deep neural nets (74)
- ML concepts & algorithms (53)
- Reinforcement learning (21)
- NLP (50)
- Conformal prediction, time series, GenAI, and more
ai/ - General AI concepts

Mathematics & CS

math/ - 27 subcategories (511 PDFs)
- Probability & stats, linear algebra, optimization
- Analysis, abstract algebra, physics, calculus
- Combinatorics, topology, game theory, and more
algorithms/ - Data structures & methods
computer science/ - CS fundamentals
signal processing/ - Signal analysis
quantum computing/ - Quantum ML

Programming & Tools

programming/ - 17 subcategories (146 PDFs)
- Python, R & Scala, PyTorch, TensorFlow, scikit-learn
- Software design, Git, NumPy, pandas, and more
python-ml-math/ - Integrated tutorials (47 PDFs)
data engineering/ - Pipelines & processing (50 PDFs)
MLOps/ - ML operations
kubernetes/ - Container orchestration
sql/ - Database queries
excel/ - Spreadsheet techniques
visualisation/ - Data visualization

Applied & Career

data science applications/ - Real-world DS
finance/ - Financial modeling
economics/ - Economic analysis
interview preparation/ - 29 resources
general reading/ - Broad interest
surveys/ - Academic surveys
selected papers/ - Curated readings
repos/ - Book source repos tracking

Presentations

`presentations/`

14 research presentations covering major deep learning topics:

Presentation	Topic	Description
`55-years-500-papers-deep-learning-evolution.pdf`	Deep learning history	A historical retrospective tracing deep learning's evolution through 500+ paper reviews, organized into thematic acts covering GANs, self-supervised learning, multimodal diffusion models, and the search for transformer successors.
`cnn-intro-2020.pdf`	Convolutional Neural Networks	Introductory lecture explaining why fully-connected networks fail for images, and covering core CNN building blocks including convolution operations, pooling layers, and spatial structure exploitation.
`diffusion-models-for-data-generation-intro.pdf`	Diffusion models	Introduction to Denoising Diffusion Probabilistic Models (DDPMs), their mathematical foundations, comparison with GAN/VAE-based approaches, and techniques to overcome their shortcomings.
`generative-adversarial-networks-gan.pdf`	GAN architectures	Covers GAN fundamentals including the generator-discriminator architecture, loss functions, various GAN types, applications in image generation, and known limitations.
`intro-to-transformers-nlp-meetup.pdf`	Transformers for NLP	Introduction to the Transformer architecture starting from RNN limitations, then walking through self-attention, positional embeddings, and the encoder-decoder structure.
`intro-to-rnns-and-transformers.pdf`	RNNs and Transformers	Combined lecture covering Recurrent Neural Networks (basic and advanced architectures, attention layers) and Transformers, showing the progression from sequential RNN processing to the parallel Transformer paradigm.
`llm-fine-tuning-why-and-how.pdf`	LLM fine-tuning	Explains the full LLM training pipeline from pretraining through supervised fine-tuning (SFT) and RLHF, along with alternatives like in-context learning and retrieval-augmented generation (RAG).
`making-data-generation-with-llm-faster.pdf`	Fast LLM inference	Deep dive into accelerating LLM text generation beyond autoregressive decoding, covering speculative decoding, Medusa, Eagle, ring/tree attention for long contexts, and multi-token prediction.
`orchestrating-intelligence-multi-agent-ai-revolution.pdf`	Multi-agent AI	Traces the evolution from standalone LLMs to orchestrated multi-agent systems, covering centralized, decentralized, hierarchical, and hybrid orchestration architectures.
`survey-of-transformers-nlp-meetup.pdf`	Transformer survey	Comprehensive survey of the Transformer architecture's evolution since the 2017 "Attention Is All You Need" paper, examining what truly changed and what continues to be improved.
`survey-of-transformers-short-version.pdf`	Transformer survey (short)	Condensed survey covering Transformer usage across domains, known shortcomings, attention mechanism improvements, positional encoding schemes, and attention-free alternatives.
`teaching-llms-with-rl-beyond-data-mimicry.pdf`	RL for LLMs	Argues that supervised fine-tuning causes LLMs to merely parrot memorized responses, and advocates for reinforcement learning approaches that achieve genuine understanding and adaptability.
`text2any-generative-ddpms-2022-2023.pdf`	Text-to-X generation	Survey of how text-to-image diffusion models (Imagen, DALL-E 2, Stable Diffusion) extend beyond static images to video, editing, personalization, 3D models, and fine-grained controls.
`unsupervised-learning-visual-features-swav.pdf`	Contrastive learning (SwAV)	Presents the SwAV method from Facebook AI Research, which combines contrastive learning with clustering assignments for self-supervised visual representation learning without requiring large numbers of negative pairs.

Metadata & Search

`mike-paper-reviews-all/reviews_metadata/`

Searchable indices for quick paper discovery:

File	Purpose
`all_paper_titles.txt`	546+ indexed paper titles (numbered list)
`clean_titles_for_search.txt`	Sanitized titles for programmatic search
`paper_with_links.csv`	Review number → ArXiv URL mappings
`reviews_1_207_titles.txt`	Index for individual reviews
`reviews_from_208_titles.txt`	Index for daily reviews

Example: Find Papers on Attention Mechanisms

grep -in "attention" mike-paper-reviews-all/reviews_metadata/all_paper_titles.txt
# Output: 42. FlashAttention: Fast and Memory-Efficient Exact Attention
#         156. Attention Is All You Need (Revisited)
#         ...

🤖 Automated Metadata Updates

Git Pre-Commit Hook

This repository includes automated metadata synchronization via a git pre-commit hook. When you commit review files, all metadata indices are automatically updated and included in your commit.

What Gets Auto-Updated

Every time you commit a Review_*.md file, the hook automatically:

✅ Extracts paper title and link from the review
✅ Updates paper_with_links.csv with the new entry
✅ Updates all_paper_titles.txt with numbered title
✅ Updates clean_titles_for_search.txt for search indexing
✅ Updates reviews_from_208_titles.txt (for reviews 208+)
✅ Stages all updated metadata files
✅ Includes them in your commit automatically

How It Works

# 1. Add a new review file
git add mike-paper-reviews-all/split-hebrew-reviews-md/Review_574.md

# 2. Commit (hook runs automatically!)
git commit -m "Add Review_574: Paper Title"

# Output you'll see:
# 📝 Detected review markdown changes, updating metadata...
# Extracting metadata from Hebrew review files...
# Extracted 573 reviews
# ✓ Metadata updated successfully
# ✓ Metadata files staged for commit

# 3. Push to remote
git push

Supported Paper Sources

The hook extracts links from multiple sources:

ArXiv - Primary source for ML/AI papers
Nature - High-impact journal articles
ACM Digital Library - Computer science publications
OpenAI - Direct paper releases and blog posts
Google Research - Research blog publications
OpenReview - Conference submissions
HuggingFace Papers - Community papers
DOI Links - Universal paper identifiers
And more...

Coverage Statistics

Metric	Value
Total Reviews	612
With Paper Links	610 (100% coverage!)
Auto-Extracted	571
Manually Added	1

No manual CSV editing needed! 🎉

🤖 Daily Review Automation

This repository includes automated daily processing of paper reviews. Every day at 5:00 AM, the system automatically checks for new review files and processes them end-to-end.

What It Does

The daily automation:

✅ Scans ~/ReviewsInbox/ for new Review_XXX.docx files
✅ Checks for duplicates - skips reviews already in repo
✅ Copies Hebrew DOCX to split-reviews-docx/
✅ Converts Hebrew DOCX to markdown with title formatting
✅ Converts English DOCX if present
✅ Commits changes with descriptive message
✅ Updates metadata automatically via pre-commit hook
✅ Pushes to GitHub
✅ Logs all actions for monitoring

Setup

The automation is set up via launchd (macOS native scheduler):

# Install daily job (runs at 5:00 AM)
cd .repo-tools/scripts
./schedule_daily_job.sh

Manual Processing

You can also run the processor manually:

# Test without making changes
python3 .repo-tools/scripts/daily_review_processor.py --dry-run

# Process new reviews now
python3 .repo-tools/scripts/daily_review_processor.py

Monitoring

View automation logs:

# View recent activity
tail -f .repo-tools/logs/daily_processor.log

# View errors
cat .repo-tools/logs/daily_processor_error.log

Management

# Check if job is running
launchctl list | grep daily-review

# Run immediately (don't wait for 5 AM)
launchctl start com.user.daily-review-processor

# Uninstall
launchctl unload ~/Library/LaunchAgents/com.user.daily-review-processor.plist

For detailed documentation, see .repo-tools/scripts/README.md

Collection Statistics

Metric	Value
Total Paper Reviews	612
Hebrew Reviews (Markdown)	612 files
English Reviews (Markdown)	245 files
Reviews with Paper Links	610 (100% coverage!)
Daily Reviews	209 (May 2024 - Feb 2026)
Learning Material PDFs	1,496
Learning Categories	23
Presentations	14
Total Repository Size	~17 GB
Coverage Period	2022-2026
Languages	English + Hebrew
License	CC0-1.0 (Public Domain)

Repository Structure

scientific-resources/
├── mike-paper-reviews-all/          # Main review collection
│   ├── split-hebrew-reviews-md/     # 612 Hebrew review markdown files ⭐
│   │   ├── Review_001.md
│   │   ├── Review_002.md
│   │   └── ... → Review_612.md
│   ├── split-english-reviews-md/    # 245 English review markdown files
│   ├── split-reviews-docx/          # 612 DOCX source files
│   │   ├── Review_001.docx
│   │   └── ... → Review_612.docx
│   ├── reviews_metadata/            # Auto-updated metadata (100% coverage) 🤖
│   │   ├── paper_with_links.csv     # 610 reviews with links
│   │   ├── all_paper_titles.txt     # Numbered title list
│   │   ├── clean_titles_for_search.txt
│   │   └── reviews_from_208_titles.txt
│   └── archive/                     # Historical files
│       ├── old-pdf/                 # Old PDF compilations
│       ├── old-docx/                # Old batch DOCX files
│       └── archive-reviews/         # Legacy individual PDFs
│
├── learning-materials/              # Educational resources (13 GB, 1,496 PDFs)
│   ├── machine learning/            # 14 subcategories, 518 PDFs
│   ├── math/                        # 27 subcategories, 511 PDFs
│   ├── programming/                 # 17 subcategories, 146 PDFs
│   ├── python-ml-math/              # Math for ML & DS, 47 PDFs
│   ├── data engineering/            # Pipelines & processing, 50 PDFs
│   ├── algorithms/                  # Data structures & algorithms
│   ├── interview preparation/       # Career resources
│   ├── repos/                       # Book source repos tracking
│   └── ... (22 categories total)
│
├── presentations/                   # Research presentations (32 MB)
│   └── 14 PDF presentations
│
├── .repo-tools/                     # Automation framework
│   └── scripts/update_metadata.py   # Metadata/README/badge updater (via pre-commit hook)
│
├── .git/hooks/pre-commit            # Auto-update git hook 🤖
├── .gitignore                       # Ignore system files
├── METADATA_UPDATE_PROCESS.md       # Automation guide
└── README.md                        # This file

For Researchers & Students

Researchers

Literature Reviews: 612 analyzed papers with critical insights
Trend Analysis: Track AI/ML research evolution (2022-2026)
Methodology Examples: Models for academic analysis
Quick Reference: Searchable metadata for rapid paper discovery

Students

Structured Learning: 22 categories with 1,496 PDFs of educational materials
Academic Writing: Examples of comprehensive paper reviews
Interview Prep: 32+ resources for ML/DS interviews
Practical Skills: Programming, SQL, data engineering tutorials

Practitioners

Implementation Insights: Practical guidance from research
Technology Assessment: Evaluation of emerging techniques
Professional Development: Presentations and tutorials
Quick Lookup: Find relevant papers by topic instantly

Contributing

Adding New Reviews

Thanks to the automated git hook, adding reviews is simple:

Create your review file: mike-paper-reviews-all/split-hebrew-reviews-md/Review_XXX.md
Include the paper link in the review (ArXiv, DOI, or other sources)
Commit the file: git commit -m "Add Review_XXX: Paper Title"
Metadata updates automatically! No manual CSV editing needed.

Guidelines

Follow the Review_NNN.md naming convention
Include paper link (ArXiv, DOI, etc.) in the review text
The git hook will extract title and link automatically
All metadata files are auto-updated on commit

See METADATA_UPDATE_PROCESS.md for detailed documentation on:

Review formatting standards
Manual metadata updates (if needed)
Quality standards
Repository maintenance

License

This work is dedicated to the public domain under the CC0 1.0 Universal license. You are free to use, modify, and distribute this content for any purpose without attribution.

Built with curiosity and dedication to open science