Kuaishou (China)

5 related articles

Subscribe to updates

arXiv Nov 5, 2026
Search pipelines can run clean and still miss the point: WISE traces the bad page back to ranking and recall
Industrial search pipelines can return results on time and still show pages that fail the user's intent, while existing monitoring watches latency, availability and errors. WISE splits the diagnosis in two: it first judges whether a page has a quality problem using only user-facing multimodal evidence, then pulls internal pipeline evidence to attribute confirmed problems to supply, recall, relevance scoring, ranking or reranking. On WISE-Bench, 12,000 real search cases, Qwen3.5-35B-A3B rises from 35.25% to 75.20% end-to-end exact match after trajectory-supervised fine-tuning and AHPO.
Industrial search pipelines can return results on time and still show pages that fail the user's intent, while existing monitoring watches latency, availability and errors. WISE splits the diagnosis in two: it first judges whether a page has a quality problem using only user-facing multimodal evidence, then pulls internal pipeline evidence to attribute confirmed problems to supply, recall, relevance scoring, ranking or reranking. On WISE-Bench, 12,000 real search cases, Qwen3.5-35B-A3B rises from 35.25% to 75.20% end-to-end exact match after trajectory-supervised fine-tuning and AHPO.
Kuaishou (China)
Wenyi Xu（First author）
WISE System Quality Diagnosis WISE-Bench ReTrace Adaptive Harness–Policy Optimization
arXiv Oct 6, 2026
Kuaishou and Nanjing University propose BPS: adding reverse preference pairs for boundary failures so the same response is rejected where wrong and chosen where right
Kuaishou Technology and Nanjing University introduce Bidirectional Preference Synthesis (BPS): for each validated boundary failure, BPS keeps the conventional forward pair under the original prompt and adds a reverse pair under a synthesized achieved prompt, so the same response is rejected where it is wrong and chosen where it is right; on Qwen3-4B-Instruct-2507, BPS preserves original-side pairwise ranking while raising achieved-side ranking accuracy on held-out crossed anchors from 6.8% to 62.3%, without changing the DPO objective, training a reward model, or requiring online sampling.
Kuaishou Technology and Nanjing University introduce Bidirectional Preference Synthesis (BPS): for each validated boundary failure, BPS keeps the conventional forward pair under the original prompt and adds a reverse pair under a synthesized achieved prompt, so the same response is rejected where it is wrong and chosen where it is right; on Qwen3-4B-Instruct-2507, BPS preserves original-side pairwise ranking while raising achieved-side ranking accuracy on held-out crossed anchors from 6.8% to 62.3%, without changing the DPO objective, training a reward model, or requiring online sampling.
Nanjing University Kuaishou (China)
Junbo Wang（First author）
Preference Optimization Instruction Following Bidirectional Preference Synthesis
arXiv Oct 2, 2026
Kuaishou team's DARA weights rewards by active-group density, cutting tool-calling format steps by up to 26% and math length-compliance steps by up to 65%
The work characterizes each reward's contribution to policy updates in multi-reward RL through "advantage energy," shows that under idealized GDPO normalization a reward's advantage energy is proportional to its active-group density (the fraction of rollout groups in which it yields nonzero relative advantages), and proposes DARA, an inverse-square-root density weighting that reaches target behaviors faster than GDPO: up to 26% fewer training steps for format compliance in tool calling and up to 65% fewer steps for near-saturated length compliance in mathematical reasoning, while remaining competitive in final performance.
The work characterizes each reward's contribution to policy updates in multi-reward RL through "advantage energy," shows that under idealized GDPO normalization a reward's advantage energy is proportional to its active-group density (the fraction of rollout groups in which it yields nonzero relative advantages), and proposes DARA, an inverse-square-root density weighting that reaches target behaviors faster than GDPO: up to 26% fewer training steps for format compliance in tool calling and up to 65% fewer steps for near-saturated length compliance in mathematical reasoning, while remaining competitive in final performance.
Beijing Foreign Studies University University of Minnesota University of Wisconsin–Madison Dalian University of Technology Kuaishou (China)Southeast University Stony Brook University
Tong Zheng（First author）
Density-Aware Reward Aggregation Advantage energy Multi-reward reinforcement learning tool calling Mathematical Reasoning
arXiv Oct 1, 2026
ZJU team runs 25 same-family on-policy distillation pairs on Qwen2.5 from 0.5B to 14B, finding peak capability is predictable from student scale and teacher score, and that weaker teachers can teach stronger students
The study systematically characterizes the scaling properties of same-family on-policy distillation (OPD) on Qwen2.5 (0.5B–14B) math reasoning, finding a regular early useful-transfer regime in which held-out accuracy rises approximately linearly with the square root of token-level reverse KL from the student initialization, and fitting power laws in student scale, teacher scale, and teacher gold score that predict peak accuracy and transfer rate, with every weak-to-strong student peaking above its own teacher.
The study systematically characterizes the scaling properties of same-family on-policy distillation (OPD) on Qwen2.5 (0.5B–14B) math reasoning, finding a regular early useful-transfer regime in which held-out accuracy rises approximately linearly with the square root of token-level reverse KL from the student initialization, and fitting power laws in student scale, teacher scale, and teacher gold score that predict peak accuracy and transfer rate, with every weak-to-strong student peaking above its own teacher.
Zhejiang University Kuaishou (China)
Yuntai Bao（First author）
On-Policy Distillation Scaling Law Weak-to-Strong Transfer Capability Transfer Qwen2.5
arXiv Sep 27, 2026
SEEK externalizes search evaluation criteria into a routable skill bank, improving listwise quality evaluation and attribution diagnosis in Kuaishou short-video search
The work proposes SEEK (Skill-routed Evaluation with Evolvable Knowledge), which externalizes specific search evaluation criteria into a skill bank, dynamically routes relevant skills for each query-result list pair, and employs a task-adapted listwise evaluator to produce page-level judgments and failure mode attribution; a two-stage training pipeline teaches the evaluator to align evaluation criteria with human preferences, while a replay-gated skill bank allows recurring evaluation knowledge gaps to be incorporated without model retraining; experiments on industrial short-video search show that SEEK improves listwise quality evaluation accuracy and achieves significant progress in attribution diagnosis, and SEEK has been deployed at Kuaishou, a platform with over 400 million daily activ
The work proposes SEEK (Skill-routed Evaluation with Evolvable Knowledge), which externalizes specific search evaluation criteria into a skill bank, dynamically routes relevant skills for each query-result list pair, and employs a task-adapted listwise evaluator to produce page-level judgments and failure mode attribution; a two-stage training pipeline teaches the evaluator to align evaluation criteria with human preferences, while a replay-gated skill bank allows recurring evaluation knowledge gaps to be incorporated without model retraining; experiments on industrial short-video search show that SEEK improves listwise quality evaluation accuracy and achieves significant progress in attribution diagnosis, and SEEK has been deployed at Kuaishou, a platform with over 400 million daily activ
Kuaishou (China)Peking University
Zhongxin Huang（First author）
SEEK Search quality evaluation Listwise evaluator

You’re all caught up

Search pipelines can run clean and still miss the point: WISE traces the bad page back to ranking and recall

Kuaishou and Nanjing University propose BPS: adding reverse preference pairs for boundary failures so the same response is rejected where wrong and chosen where right

Kuaishou team's DARA weights rewards by active-group density, cutting tool-calling format steps by up to 26% and math length-compliance steps by up to 65%

ZJU team runs 25 same-family on-policy distillation pairs on Qwen2.5 from 0.5B to 14B, finding peak capability is predictable from student scale and teacher score, and that weaker teachers can teach stronger students

SEEK externalizes search evaluation criteria into a routable skill bank, improving listwise quality evaluation and attribution diagnosis in Kuaishou short-video search