Research and recognition

Latest News

Publications, awards, and milestones from LIKE Lab.
Celebrating the work and achievements of our community.

2026

8 updates

Sep 2026Award
Scholarships
Rang Li receives the PKU Presidential Scholarship
Congratulations to Rang Li on receiving the Peking University Presidential Scholarship for Doctoral Students.
Rang Li’s homepage
Sep 2026Acceptance
NeurIPS1 paper
Claw-Eval accepted at NeurIPS 2026
Claw-Eval joins the Evaluations & Datasets Track, advancing trustworthy evaluation of autonomous agents through task completion, safety, and robustness checks.
View paper
- NeurIPS 2026 · Evaluations & Datasets TrackClaw-Eval: Towards Trustworthy Evaluation of Autonomous AgentsBowen Ye et al.
Source
Jul 2026Conference
ICML2 papers
Two papers at ICML 2026
FIRE develops agents that reason over real-world files through stateful reinforcement learning. Rollout Routing Replay (R3) stabilizes MoE reinforcement learning by aligning inference-time and training-time routing decisions.
View 2 papers
- ICML 2026FIRE: Learning to Navigate and Act on Real-World Files via Stateful Reinforcement LearningJingyuan Ma et al.
- ICML 2026Stabilizing MoE Reinforcement Learning by Aligning Training and Inference RoutersWenhan Ma et al.
Source 1 · Source 2
Jul 2026Publication
ACL5 papers
Five papers at ACL 2026
Two main-conference papers and three Findings papers explore conceptual change, stable MoE reinforcement learning, reasoning safety, token budgeting, and human-centered evaluation.
View 5 papers
- ACL 2026HistLens: Mapping Idea Change across Concepts and CorporaYi Jing et al.
- ACL 2026Towards Stable and Effective Reinforcement Learning for Mixture-of-ExpertsDi Zhang et al.
- Findings of ACL 2026HAUNTATTACK: When Attack Follows Reasoning as a ShadowJingyuan Ma et al.
- Findings of ACL 2026SelfBudgeter: Adaptive Token Allocation for Efficient LLM ReasoningZheng Li et al.
- Findings of ACL 2026SenseJudge: Human-Centric Preference-Driven Judgment FrameworkRui Li et al.
Source 1 · Source 2 · Source 3 · Source 4 · Source 5
Jun 2026Conference
CVPR1 paper
GroundingME at CVPR 2026
GroundingME reveals gaps in multimodal models’ visual grounding through challenging tests of discrimination, spatial relations, limited visibility, and ungroundable queries.
View paper
- CVPR 2026GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional EvaluationRang Li et al.
Source 1 · Source 2 · Source 3
Apr 2026Award
Scholarships
Rang Li receives the Xu Teli Scholarship
Congratulations to Rang Li on receiving the Xu Teli Scholarship from Beijing Institute of Technology, the university’s highest scholarship honor.
Rang Li’s homepage
Mar 2026Publication
AAAI3 papers
Three papers in AAAI 2026
Our work examines unreasonable math problems, selects instruction-tuning data with RICo, and studies how process reward models transfer from mathematical reasoning to code generation.
View 3 papers
- AAAI 2026Large Language Models Struggle with Unreasonability in Math ProblemsJingyuan Ma et al.
- AAAI 2026RICo: Refined In-Context Contribution for Automatic Instruction-Tuning Data SelectionYixin Yang et al.
- AAAI 2026From Mathematical Reasoning to Code: Generalization of Process Reward Models in Test-Time ScalingZhengyu Chen et al.
Source 1 · Source 2 · Source 3
Jan 2026Conference
AAAI Bridge · Oral1 paper
SCoRE selected for an oral at AAAI 2026 Bridge
SCoRE benchmarks long-chain reasoning in commonsense scenarios and is featured as an oral presentation in the Logical and Symbolic Reasoning in Language Models (LMReasoning) Bridge program.
View paper
- AAAI 2026 Bridge · LMReasoning · OralSCoRE: Benchmarking Long-Chain Reasoning in Commonsense ScenariosWeidong Zhan et al.
Source 1 · Source 2

2025

10 updates

Nov 2025Publication
EMNLP2 papers
Two papers at EMNLP 2025
One main-conference paper develops a probabilistic theory of LLM self-correction; one Findings paper evaluates multimodal understanding of implicit narratives in comic strips.
View 2 papers
- EMNLP 2025A Probabilistic Inference Scaling Theory for LLM Self-CorrectionZhe Yang et al.
- Findings of EMNLP 2025Beyond Single Frames: Can LMMs Comprehend Implicit Narratives in Comic Strip?Xiaochen Wang et al.
Source 1 · Source 2
Aug 2025Publication
CCL · Evaluation Report1 paper
SpaCE2025 evaluation report published at CCL 2025
The fifth Spatial Cognition Evaluation brings together five Chinese and English subtasks to assess spatial semantics and reasoning. Its overview report documents the benchmark, participating systems, and evaluation results.
View paper
- CCL 2025 · Evaluation ReportOverview of CCL25-Eval Task 1: The Fifth Spatial Cognition Evaluation (SpaCE2025)Yuhang Qin et al.
Source
Jul 2025Publication
TMLR1 paper
GLAN published in Transactions on Machine Learning Research
Synthetic Data (Almost) from Scratch introduces generalized instruction tuning through a taxonomy of human knowledge and skills, generating diverse instruction data with minimal seed examples.
View paper
- TMLR 2025Synthetic Data (Almost) from Scratch: Generalized Instruction Tuning for Language ModelsHaoran Li et al.
Source
Jul 2025Publication
ACL3 papers
Three papers at ACL 2025
Two main-conference papers investigate self-correction and uncertainty estimation, alongside a Findings paper on persona-based human behavior simulation.
View 3 papers
- ACL 2025Confidence v.s. Critique: A Decomposition of Self-Correction Capability for LLMsZhe Yang et al.
- ACL 2025Towards Harmonized Uncertainty Estimation for Large Language ModelsRui Li et al.
- Findings of ACL 2025How Far are LLMs from Being Our Digital Twins? A Benchmark for Persona-Based Behavior Chain SimulationRui Li et al.
Source 1 · Source 2 · Source 3
Jun 2025Conference
CVPR1 paper
AdaMMS at CVPR 2025
AdaMMS explores merging heterogeneous multimodal large language models with unsupervised optimization of merging coefficients.
View paper
- CVPR 2025AdaMMS: Model Merging for Heterogeneous Multimodal Large Language Models with Unsupervised Coefficient OptimizationYiyang Du et al.
Source 1 · Source 2
Apr 2025Conference
ICLR1 paper
SynPO at ICLR 2025
Self-Boosting Large Language Models with Synthetic Preference Data introduces an iterative approach to improving model alignment through synthetic preferences.
View paper
- ICLR 2025Self-Boosting Large Language Models with Synthetic Preference DataQingxiu Dong et al.
Source 1 · Source 2 · Source 3
Apr 2025Publication
NAACL1 paper
SG-FSM in Findings of NAACL 2025
SG-FSM uses a self-guiding finite-state framework to break down multi-hop questions, correct intermediate errors, and strengthen zero-shot reasoning.
View paper
- Findings of NAACL 2025SG-FSM: A Self-Guiding Zero-Shot Prompting Paradigm for Multi-Hop Question Answering Based on Finite State MachineXiaochen Wang et al.
Source
Apr 2025Publication
AAAI1 paper
Language model activation patterns in AAAI 2025
Exploring Activation Patterns of Parameters in Language Models studies which parameters activate across layers, input domains, and tasks.
View paper
- AAAI 2025Exploring Activation Patterns of Parameters in Language ModelsYudong Wang et al.
Source
Jan 2025Publication
Natural Language Processing1 paper
Dataset evaluation study in Natural Language Processing
Using named entity recognition as a case study, this work examines dataset reliability, difficulty, and validity. It appears in the January 2025 issue, following first online publication in September 2024.
View paper
- Natural Language Processing 31(1), 90–110Statistical dataset evaluation: A case study on named entity recognitionChengwen Wang et al.
Source
Jan 2025Publication
COLING1 paper
Number representations at COLING 2025
Language Models Encode the Value of Numbers Linearly investigates numerical representations in hidden states using linear probes.
View paper
- COLING 2025Language Models Encode the Value of Numbers LinearlyFangwei Zhu et al.
Source

Lab updates · Updated October 21, 2026.
Dates indicate the publication, conference, acceptance, or award month, as labeled.

Explore all publications

Latest News

2026

Rang Li receives the PKU Presidential Scholarship

Claw-Eval accepted at NeurIPS 2026

Two papers at ICML 2026

Five papers at ACL 2026

GroundingME at CVPR 2026

Rang Li receives the Xu Teli Scholarship

Three papers in AAAI 2026

SCoRE selected for an oral at AAAI 2026 Bridge

2025

Two papers at EMNLP 2025

SpaCE2025 evaluation report published at CCL 2025

GLAN published in Transactions on Machine Learning Research

Three papers at ACL 2025

AdaMMS at CVPR 2025

SynPO at ICLR 2025

SG-FSM in Findings of NAACL 2025

Language model activation patterns in AAAI 2025

Dataset evaluation study in Natural Language Processing

Number representations at COLING 2025