Yang Ruofeng
Explained projects
Diffusion_Code: The One-Line Repo That Exposes the Whole Diffusion Stack
A bare README turns into a map of the hidden machinery behind diffusion models, and why so many simple AI projects are anything but.
DuelingUCB: Learning the Best Choice Without Ever Seeing a Score
A compact Python repo that turns pairwise wins into an Upper Confidence Bound loop, showing why preference feedback can be more useful than numeric reward.
Tencent_test: The Empty Repo That Points to AI-Written Unit Tests
A blank GitHub shell becomes a doorway into TestAgent, a specialized model pipeline that drafts tests and completes missing assertions.
wanshuiyin.github.io: The Smallest Possible Personal Site
A GitHub Pages repo that turns a reserved username into a public homepage with almost no code.
Inside `wanshuiyin/AAAI-2023-The-Learnability-of-Nonlinear-SSL`: A Proof You Can Run for How Nonlinear SSL Learns
A hand-coded NumPy simulator turns an AAAI theory paper into a geometric story about basins of attraction, augmentations, and feature recovery.
AISTATS_VESDE: The Diffusion Repo That Already Knows the Answer
By swapping a learned score network for a closed-form Gaussian mixture, this tiny research codebase turns sampler design into a test of discretization error, correction steps, and variance schedule choice.
OpenReview-Score-Watcher: The Scoreboard Hidden in a Paper Page
A Chrome extension that reads OpenReview review notes, averages the ratings, and keeps the result in a floating window that follows the reader.
Auto-claude-code-research-in-sleep: The End of the Agent Framework: Unpacking ARIS
How a folder of plain Markdown files and adversarial LLM psychology turned Claude Code into an autonomous ML researcher.
Yet to be explained
ARIS-in-AI-Offer
Explain
Bilingual (中文+EN) ML / LLM / diffusion / agent interview cheat sheets for AI 秋招 — generated by ARIS /interview-cheatsheet, rendered by /render-html into single-file HTML, reads anywhere — plus a CV→DBLP-fact-checked academic homepage generator and hand-authored long-form blogs 🌱
HERO-Anti-OverDefense
Explain
HERO = Hashing · Edge cases · Rubrics · Overbuild — the four shapes coding agents over-defend in. A paste-in contract that stops them. Works with Claude Code, Codex, Antigravity, Cursor, Copilot, Windsurf, Gemini CLI.
Anti-Autoresearch
Explain
Don't trust an autoresearch paper at face value. Reviewer-side integrity forensics (self-consistency + fabrication), deterministic verdict. 61 signals: 46 integrity hack-patterns (families A–H, verdict-bearing) + 13 zero-weight AI writing-style impressions (AIS) + 2 advisory. Not an opaque AI-text classifier. The dual of ARIS.
ARIS-Movie-Director
Explain
Agentic, long-horizon visual generation: a fuzzy story → a cross-model-audited image-based movie. Brings ARIS's research-wiki + multi-agent debate to multimodal generation (intelligence lives in the agent; the diffusion model just renders). Image-based today, video next.
ARIS-Anything
Explain
Generalizing ARIS from 科研 to 研究 — same five-step loop (plan · draft · adversarial review · iterate · persist) applied beyond academic research to investment DD, legal, market, self-directed learning, journalism, engineering post-mortems (科研 ⊂ 研究) 🌱
ALIGN-Agentic-Loop-Image-GeneratioN
Explain
Agentic loop image generation without diffusion or autoregressive image models. Coding agents draw in p5.js, using iterative refinement and adversarial visual review for high-fidelity art and method figures. Skills for Claude Code and Codex.