Baidu's UNIQUE unifies retrieval and ranking in one early-fusion model, lifting watch time 0.96% in A/B tests
_reachsumit · x · 2026-09-22
Baidu published UNIQUE, a unified retrieval and ranking system for large-scale feed recommendation.
- Problem: conventional retrieval-ranking pipelines suffer from hierarchical quantization instability and information loss between stages, hurting long-tail and cold-start recommendation.
- Approach: UNIQUE fuses generative code-based candidate retrieval with target-aware ranking in a single early-fusion architecture trained end-to-end, plus a balanced quantization mechanism to mitigate codebook imbalance and improve long-tail representation.
- Deployment: live in Mobile Baidu's homepage feed, discovery page, and short-video recommendation; online A/B tests show a 0.96% gain in total watch duration and 1.08% in total consumption.
More from Research
- SVEET: streaming video editing with a diffusion model hits 15 FPS on a single H100 — SJTU · 2026-09-22
- Semantic cache feedback loop quietly breaks its own guaranteed 2% error rate, worst case 3x worse — Reasonable_Royal_621 · 2026-09-22
- Stanford turns papers into AI agents; two unrelated studies surface unreported ADHD genetic link — VraserX · 2026-09-22
- Xiaomi MiMo 2.6 Pro formalizes Li–Yorke chaos theorem in 6,000+ lines of verified Lean — bookwormengr · 2026-09-22
- Apollo, the first advanced LLM for Ancient Greek, aims to restore tattered papyri — nordicinst · 2026-09-22
- Two Minute Papers: Jev claims 200x AI speedup via System One models — but there's a catch — Two Minute Papers · 2026-09-22