Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Yu–Gang JiangSuggested from co-authorship
- Suggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works91 from public data
- The rise and potential of large language model based agents: a survey906
A comprehensive survey on LLM-based agents is conducted, covering their construction frameworks, application scenarios, and the exploration of societies built upon LLM-based agents.
- Secrets of RLHF in Large Language Models Part I: PPO290
The PPO-max, an advanced version of PPO algorithm, is explored, to efficiently improve the training stability of the policy model, and a comprehensive analysis of RLHF abilities compared with SFT models and ChatGPT is performed.
- Memory in the Age of AI Agents288
This work aims to provide an up-to-date landscape of current agent memory research, clearly delineating the scope of agent memory and distinguishing it from related concepts such as LLM memory, retrieval augmented generation (RAG), and context engineering, and articulate a forward-looking perspective on emerging research frontiers.
- LoRAMoE: Alleviating World Knowledge Forgetting in Large Language Models via MoE-Style Plugin186
Experimental results show that, as the instruction data increases, Lo-RAMoE can significantly improve the ability to process downstream tasks, while maintaining the world knowledge stored in the LLM.
- Secrets of RLHF in Large Language Models Part II: Reward Modeling179
This report introduces contrastive learning to enhance the ability of reward models to distinguish between chosen and rejected responses, thereby improving model generalization and employs meta-learning to enable the reward model to maintain the ability to differentiate subtle differences in out-of-distribution samples.
- AgentGym: Evolving Large Language Model-based Agents across Diverse Environments113
This paper proposes AgentGym, a new framework featuring a variety of environments and tasks for broad, real-time, uni-format, and concurrent agent exploration, and proposes a novel method, AgentEvol, to investigate the potential of agent self-evolution beyond previously seen data across tasks and environments.
- StepCoder: Improve Code Generation with Reinforcement Learning from Compiler Feedback98
StepCoder is introduced, a novel RL framework for code generation, consisting of two main components: CCCS addresses the exploration challenge by breaking the long sequences code generation task into a Curriculum of Code Completion Subtasks, while FGO only optimizes the model by masking the unexecuted code segments to provide Fine-Grained Optimization.
- 90
- AgentPRM: Process Reward Models for LLM Agents via Step-Wise Promise and Progress81
This work explores constructing process reward models (PRMs) to evaluate each decision and guide the agent's decision-making process, and proposes a re-defined PRM for agent tasks, named AgentPRM, to capture both the interdependence between sequential decisions and their contribution to the final goal.
- StepCoder: Improving Code Generation with Reinforcement Learning from Compiler Feedback76
StepCoder is introduced, a novel RL framework for code generation, consisting of two main components: CCCS addresses the exploration challenge by breaking the long sequences code generation task into a C urriculum of C ode C ompletion S ubtasks, while FGO only optimizes the model by masking the unexecuted code segments to provide F ine-G rained O ptimization.
- Below Chance: Code and Derived Data for an Audit of the TRACE Continual-Learning Benchmark and a Census of Published Sequential Baselines70
TRACE is introduced, a novel benchmark designed to evaluate continual learning in LLMs and the Reasoning-augmented Continual Learning (RCL) approach, which integrates task-specific cues with meta-rationales, effectively reducing catastrophic forgetting in LLM while expediting convergence on novel tasks.
- AgentGym: Evaluating and Training Large Language Model-based Agents across Diverse Environments57
A GENT G YM is proposed, a framework featuring 7 real-world scenarios, 14 environments, and 89 tasks for unified, real-time, and concurrent agent interaction that takes the initial step to develop LLM-based agents that can handle diverse tasks via methods like self-improvement or reinforcement learning.
- BAPO: Stabilizing Off-Policy Reinforcement Learning for LLMs via Balanced Policy Optimization with Adaptive Clipping56
This work proposes BAlanced Policy Optimization with Adaptive Clipping with Adaptive Clipping (BAPO), a simple yet effective method that dynamically adjusts clipping bounds to adaptively re-balance positive and negative contributions, preserve entropy, and stabilize RL optimization.
- Enhancing LLM Reasoning via Critique Models with Test-Time and Training-Time Supervision54
This paper delves into a two-player paradigm that separates the roles of reasoning and critique models, where the critique model provides step-level feedback to supervise the reasoning (actor) model during both test-time and train-time, and proposes a critique-in-the-loop self-improvement method.
- MouSi: Poly-Visual-Expert Vision-Language Models28
This paper proposes the use of ensemble experts technique to synergizes the capabilities of individual visual encoders, including those skilled in image-text matching, OCR, image segmentation, etc, while bridging the gap between image encoders and pre-trained LLMs.
- Improving Generalization of Alignment with Human Preferences through Group Invariant Learning28
This work proposes a novel approach that can learn a consistent policy via RL across various data groups or domains, and significantly enhances training stability and model generalization.
- Have the VLMs Lost Confidence? A Study of Sycophancy in VLMs18
To mitigate sycophancy, this work proposes a synthetic dataset for training and employ methods based on prompts, supervised fine-tuning, and DPO that effectively alleviate sycophancy in VLMs, and demonstrates that enhancing image attention at high layers proves beneficial in mitigating this issue.
- Nex-N1: Agentic Models Trained via a Unified Ecosystem for Large-Scale Environment Construction16
This work introduces a comprehensive method designed to systematically scale the diversity and complexity of interactive environments and trains Nex-N1 upon the diverse and complex interactive environments established by the infrastructure.
- Reward Modeling Requires Automatic Adjustment Based on Data Quality16
A method is introduced that automatically adjusts reward mod-019 eling based on data quality, reducing the impact of noise and making full use of dataset, and significantly enhances the alignment performance of RLHF.
- 16
- 9
- 7
- AI Deception: Risks, Dynamics, and Controls6
This project provides a comprehensive and up-to-date overview of the AI deception field, covering its core concepts, methodologies, genesis, and potential mitigations, and organizes the landscape of AI deception research as a deception cycle, consisting of two key components: deception emergence and deception treatment.
- 6
- 5
- CritiQ: Mining Data Quality Criteria from Human Preferences5
This work introduces CritiQ, a novel data selection method that automatically mines criteria from human preferences for data quality with only ~30 human-annotated pairs and performs efficient data selection, and analyzes how criteria evolve and the effectiveness of majority voting.
- 5
- Improving Discriminative Capability of Reward Models in RLHF Using Contrastive Learning5
Experimental results demonstrate that the proposed contrastive learning-based reward modeling method effectively enhances the generalization of the reward model, stabilizes the reinforcement learning training process, and improves the final alignment with human preferences.
- Characterizing the Impacts of Instances on Robustness5
It is shown that robust and non-robust instances in the training dataset, though are both important for test performance, have contrary impacts on robustness, which makes it possible to build a highly robust model by leveraging the training dataset in a more effective way.
- 4
- 3
- 2
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Parrot: A Training Pipeline Enhances Both Program CoT and Natural Language CoT for Reasoning–
Parrot is proposed, a novel training pipeline for mathematical problems that significantly enhances both the performance of N-CoT and P-CoT, especially on N-CoT.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Connectivity Patterns are Task Embeddings–
This work draws inspiration from the operating mechanism of deep neural networks (DNNs) and biological brains, where neuronal activations are sparse and task-specific, and uses the connectivity patterns of neurons as a unique identifier associated with the task.
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-13. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.