Kuaishou's DIRECTOR Framework Uses Optimal Transport to Cut RecSys Compute by 66.7%
_reachsumit · x · 2026-07-30
Kuaishou's team published DIRECTOR (Dynamic Index-based Recommendation with Transport-Optimized Retrieval) to address efficiency and coordination issues in the reranking phase of recommendation systems.
- Background: Traditional Autoregressive (AR) models generate recommendation lists sequentially, incurring high latency and premature pruning. Non-Autoregressive (NAR) methods are faster but often lead to duplicate or conflicting item selections.
- Core Method: DIRECTOR maps candidate items into a continuous latent space and generates dynamic retrieval indices in parallel. It uses entropy-regularized Optimal Transport during training for conflict-aware supervision and performs global hard matching during inference.
- Performance: Compared to traditional AR search, the framework produces duplicate-free results and reduces serving compute by 66.7%.
More from Research
- Study: Agentic Harnesses Significantly Improve LLM Compositional Generalization — lateinteraction · 2026-07-30
- Stanford's Free CS336 Course: Building LLMs from Scratch to Become an AI Architect — techNmak · 2026-07-30
- Handroid: A Reconfigurable Robot Shifting Between Humanoid and Dexterous Hand — carlosdponx · 2026-07-30
- RATs: Multi-Agent Robot Lifelong Skill Learning Without Gradients or RL — rsasaki0109 · 2026-07-30
- Cisco's Open-Source Security Model Antares-3B Nears GPT-5.5 at Fraction of Cost — joshua_saxe · 2026-07-30
- DeepMind's Andrew Trask: LLMs Are Like Libraries Without an Index — iamtrask · 2026-07-30