paper.dinq.meAI 与前沿科学顶会 · 学术新星雷达
登录 DINQ
返回新星榜
TC

Tat-Seng Chua

ICLR 2026×15CVPR 2026×5KDD 2026×3AAAI 2026×2SIGIR 2026×2COLM 2026多模态推理对齐与偏好检索与知识评测与基准
OROpenReview
新星分资深
48
一作贡献0
核心/参与56
方向广度19
合作网络12
0 一作 · 28 篇 · ICLR 2026 / CVPR 2026 / KDD 2026 / AAAI 2026 / SIGIR 2026 / COLM 2026

获取完整研究档案

首次获取该候选人的完整档案及 StartLux 评估共 20 积分；每位用户单独解锁，之后查看免费。主动刷新 20 积分。

联系方式

实时查找电子邮箱，费用不超过20积分；未找到邮箱不扣费。

2026 顶会论文

28
作者位次 #9/9COLM 2026Poster· Poster Session 3· Grand Ballroom· #129

Unifying Group-Relative and Self-Distillation Policy Optimization via Sample Routing

Gengsheng Li · Tianyu Yang · Junfeng Fang · Mingyang Song · Mao Zheng · Haiyun Guo · Dan Zhang · Jinqiao Wang · Tat-Seng Chua

We diagnose two causes of self-distillation instability in LLM post-training and propose SRPO, routing correct samples to GRPO and incorrect ones to entropy-weighted SDPO, achieving better efficiency, stability, and peak performance than either alone

摘要

Reinforcement learning with verifiable rewards (RLVR) has become a standard paradigm for post-training large language models. While Group Relative Policy Optimization (GRPO) is widely adopted, its coarse credit assignment uniformly penalizes failed rollouts, lacking the token-level focus needed to efficiently address specific deviations. Self-Distillation Policy Optimization (SDPO) addresses this by providing denser, more targeted logit-level supervision that facilitates rapid early improvement, yet it frequently collapses during prolonged training. We trace this late-stage instability to two intrinsic flaws: self-distillation on already-correct samples introduces optimization ambiguity, and the self-teacher's signal reliability progressively degrades. To resolve these issues, we propose Sample-Routed Policy Optimization (SRPO), a unified on-policy framework that routes correct samples to GRPO's reward-aligned reinforcement and failed samples to SDPO's targeted logit-level correction. SRPO further incorporates an entropy-aware dynamic weighting mechanism to suppress unreliable, high-entropy distillation targets. Evaluated across five benchmarks and two model scales, SRPO achieves both the rapid early improvement of SDPO and the long-horizon stability of GRPO. It consistently surpasses the peak performance of both baselines, raising the five-benchmark average on Qwen3-8B by 3.4\% over GRPO and 6.3\% over SDPO, while simultaneously yielding moderate response lengths and lowering per-step compute costs by up to 17.2\%.

强化学习与后训练评测与基准对齐与偏好Self-DistillationOn Policy DistillationReinforcement LearningPost-TrainingOpenReview
作者位次 #10/11SIGIR 2026Oral· Information Retrieval (Full)

Verifiable Reasoning for LLM-based Generative Recommendation

Xinyu Lin · Hanqing Zeng · Hanchao Yu · Yinglong Xia · Jiang Zhang · Aashu Singh · Fei Liu · Wenjie Wang · Fuli Feng · Tat-Seng Chua · Qifan Wang

摘要

Reasoning in Large Language Models (LLMs) has recently shown strong potential in enhancing generative recommendation through deep understanding of complex user preference. Existing approaches follow a reason-then-recommend paradigm, where LLMs perform step-by-step reasoning before item generation...

推理对齐与偏好效率与系统OpenReview
作者位次 #15/16SIGIR 2026Oral· Information Retrieval (Full)

Towards Knowledgeable Deep Research: Framework and Benchmark

Wenxuan Liu · Zixuan Li · Long Bai · Chunmao Zhang · Fenghui Zhang · Zhuo Chen · Wei Li · Yuxin Zuo · Fei Wang · Bingbing Xu · Xuhui Jiang · Jin Zhang · Xiaolong Jin · Jiafeng Guo · Tat-Seng Chua · Xueqi Cheng

摘要

Deep Research (DR) requires LLM agents to autonomously perform multi-step information seeking, processing, and reasoning to generate comprehensive reports. In contrast to existing studies that mainly focus on unstructured web content, a more challenging DR task should additionally utilize structu...

检索与知识智能体与工具多模态OpenReview

常见合作者

WWWenjie Wang×7FFFuli Feng×5SWShengqiong Wu×4XXXiaobo Xia×4LQLeigang Qu×4XZXiaoyan Zhao×3YZYang Zhang×3KLKai Liu×3HFHao (Scofield) Fei×3TYTianyu Yang×2JFJunfeng Fang×2HGHaiyun Guo×2