What it does

This is Patrick Langechuan Liu’s personal paper-reading repository — a curated, chronologically organized collection of notes on deep-learning and machine-learning research. It spans computer vision, autonomous driving, diffusion models, and robotics, with each entry linking to the original paper and Liu’s own markdown summary. A GitHub Pages site mirrors the content for easier browsing.

The interesting bit

The value isn’t the code (there barely is any — it’s mostly Jupyter notebooks as a format choice). It’s the editorial layer: Liu is Director of AI at Nvidia, and his notes reflect what someone building production autonomous-driving systems actually bothers to read. The “What to read” section even prescribes a first-month curriculum for newcomers, based on what worked for him.

Key highlights

• Heavy focus on autonomous-driving perception: BEV transformers, occupancy networks, 3D lane detection, monocular depth
• Recent expansion into diffusion language models, VLA (vision-language-action) models, and latent reasoning
• Linked blog posts on Medium (“The Thinking Car”) provide longer-form synthesis beyond quick notes
• “Scratchpad” section includes quick reference notes on compute hardware and attention masks — the boring stuff you actually need
• Curated “trustworthy sources” list for when the author runs out of papers to read

Caveats

• No code implementations; this is purely a reading journal with links and commentary
• Update cadence is irregular — some months get 10+ papers, others get zero
• Notes vary in depth; some are detailed, others are just a tagged link

Verdict

Useful if you’re entering computer-vision or autonomous-driving research and want a senior practitioner’s reading list with context. Skip it if you need runnable code or systematic literature reviews; it’s a personal notebook, not a textbook.