Bilingual (EN/ZH) ML learning resource — paper deep-dives, job hunt tools, curated resources, and an AI timeline.
Live site: PaperTrace on GitHub Pages
Each paper gets a full breakdown with formula walkthroughs, interactive widgets, and connections to related work.
| Paper | Key Idea |
|---|---|
| Attention Is All You Need | The Transformer architecture |
| BERT | Masked LM + bidirectional pretraining |
| CLIP | Contrastive vision-language pretraining |
| LoRA | Low-rank adapter fine-tuning |
| FlashAttention | IO-aware exact attention |
| LLaVA | Visual instruction tuning |
| GPT-2 | Autoregressive language modeling at scale |
| PPO | Proximal policy optimization for RLHF |
| DPO | Direct preference optimization (no RL) |
| GRPO | Group relative policy optimization (DeepSeek-R1) |
| MDLM | Masked diffusion language models |
| LLaDA | Masked diffusion for LLMs |
| Block Diffusion | AR between blocks + diffusion within blocks |
| Fast DLLM | Adaptive denoising — 3-10x fewer steps |
/— Landing page with paper grid, category filter/daily— Paper feed + News & Events (行业动态)/resources— Curated learning resources: YouTube, Bilibili, newsletters, blogs, X/Twitter, tools/interview— Job hunt hub: LeetCode resources, company job openings, ML interview Q&A (20 questions, 5 sections)/interview/notes— "Notes on the Job Hunt / 写在找工路上"
/guide— Study guide/timeline— AI/ML timeline
- Next.js 14 — App Router, static export for GitHub Pages
- React — Interactive widgets (custom, no heavy chart libs)
- KaTeX — Math rendering
- Tailwind CSS — Styling with dark mode
- Custom i18n —
t(en, zh)hook, no external library
git clone https://github.com/yayajjiang/PaperTrace.git
cd PaperTrace
npm install
npm run devOpen http://localhost:3000.
Pushes to main automatically deploy to GitHub Pages via GitHub Actions.
Settings → Pages → Source → GitHub Actions.
- Add metadata to
src/lib/papers.ts - Create
src/app/papers/<slug>/page.tsx - Key components:
<Math tex="..." />/<Math display tex="..." />— KaTeX math<Widget title="...">— interactive visualizations<Collapsible title="...">— expandable sections<KeyInsight>— highlighted callout boxes<SectionHeader>— colored section headers (used in interview page)
Scan the QR code to join our WeChat community for ML learners and job seekers.
扫码加入微信交流群，一起学习 ML、交流找工经验。
MIT