Institution profile

Chengdu University of Information Technology

Academic institutionasia · cn
Official website
Research library29linked papers
Opportunities0open roles
Selected work

Representative Papers

Actively Obtaining Environmental Feedback for Autonomous Action Evaluation Without Predefined Measurements

Jan 04, 2026arXiv.org

In open and dynamic environments, agents often struggle to evaluate their actions due to the absence of predefined feedback. This work proposes an active feedback acquisition model that autonomously discovers, filters, and validates effective feedback signals by analyzing the environmental changes induced by its actions, without relying on external rewards or pre-specified metrics. The approach incorporates an intrinsic-goal-driven self-triggering mechanism—guided by objectives such as accuracy and efficiency—to enable autonomous action planning. Experimental results demonstrate that the model substantially enhances the efficiency and robustness of feedback identification, allowing agents to rapidly focus on and acquire high-quality feedback without external supervision.

3 citations1 influentialRead paper

Human Simulation Computation: A Human-Inspired Framework for Adaptive AI Systems

Jan 20, 2026

This work addresses the limitations of large language models, which, trained exclusively on static text, struggle to verify reasoning or adapt in open, dynamic environments. To overcome this, the paper proposes a human cognition-inspired closed-loop intelligence framework that unifies thinking, acting, learning, reflection, and task scheduling into a cohesive internal reasoning process. The framework enables an action-driven self-optimization mechanism through prototype-guided reasoning, action-mediated expansion of perceptual boundaries, and immediate learning from environmental feedback. Theoretical analysis demonstrates that this approach effectively compensates for the inherent deficiencies of pure language models in reasoning validation and environmental adaptation, substantially enhancing the robustness and interactive efficiency of AI systems in real-world scenarios.

2 citationsRead paper

Learning Situation-Conditioned Thinking Policies for Long-Term LLM Agents

Oct 07, 2026

This study addresses the challenges of unbounded historical memory growth and the absence of context-aware thought activation mechanisms in intelligent agents. To this end, we propose a Context-Conditioned Thought Memory framework that transcends conventional retrieval-compression paradigms. Specifically, it employs a lightweight thought policy network to internalize cross-cycle reasoning experiences into predictive strategies, thereby enabling spatiotemporal evolution-based long-range pattern discovery and dynamic knowledge integration. Experimental results demonstrate that the proposed framework achieves perfect scores of 1.0 in both temporal rule generalization F1 and relation discovery accuracy. Furthermore, it yields substantial improvements in overall reasoning performance while reducing online processing latency by approximately 90%.

0 citationsRead paper

Autoresearch in Mixed-Integer Linear and Nonlinear Programming

Sep 30, 2026

This study addresses the challenges of managing competing ideas and controlling long-cycle experimental trajectories in automated research for mixed-integer programming (MIP) by proposing the AutoMIP framework. This method is the first to integrate a persistent idea pool with an algorithmic tree search mechanism, leveraging an agent skill framework to enable the collaborative management of diverse research concepts and structured experimental exploration while continuously refining research directions and preserving valid hypotheses. Evaluated on the MIPLib and MINLPLib benchmarks, AutoMIP discovers numerous new optimal solutions and achieves success rates that significantly surpass those of existing automated research baselines.

0 citationsRead paper

VLM Fine-Tuning for End-to-End Combinatorial Optimization

Sep 29, 2026

This study addresses the limitations of text-serialized representations in capturing spatial and relational structures within combinatorial optimization problems by proposing a general-purpose vision-language solver. The proposed method integrates textual and visual inputs, incorporating gold-standard-free visual representations, and employs a vision-language model (VLM) jointly optimized through supervised fine-tuning and verifier-guided reinforcement learning. Experimental results demonstrate that on complex tasks such as the Capacitated Vehicle Routing Problem (CVRP), visual information significantly enhances solution quality, with this advantage becoming increasingly pronounced as problem scale grows. These findings reveal the core value of visual representations in large-scale combinatorial optimization.

0 citationsRead paper
Recent publications

Latest Papers

Learning Situation-Conditioned Thinking Policies for Long-Term LLM Agents

Oct 07, 2026

This study addresses the challenges of unbounded historical memory growth and the absence of context-aware thought activation mechanisms in intelligent agents. To this end, we propose a Context-Conditioned Thought Memory framework that transcends conventional retrieval-compression paradigms. Specifically, it employs a lightweight thought policy network to internalize cross-cycle reasoning experiences into predictive strategies, thereby enabling spatiotemporal evolution-based long-range pattern discovery and dynamic knowledge integration. Experimental results demonstrate that the proposed framework achieves perfect scores of 1.0 in both temporal rule generalization F1 and relation discovery accuracy. Furthermore, it yields substantial improvements in overall reasoning performance while reducing online processing latency by approximately 90%.

0 citationsRead paper

Autoresearch in Mixed-Integer Linear and Nonlinear Programming

Sep 30, 2026

This study addresses the challenges of managing competing ideas and controlling long-cycle experimental trajectories in automated research for mixed-integer programming (MIP) by proposing the AutoMIP framework. This method is the first to integrate a persistent idea pool with an algorithmic tree search mechanism, leveraging an agent skill framework to enable the collaborative management of diverse research concepts and structured experimental exploration while continuously refining research directions and preserving valid hypotheses. Evaluated on the MIPLib and MINLPLib benchmarks, AutoMIP discovers numerous new optimal solutions and achieves success rates that significantly surpass those of existing automated research baselines.

0 citationsRead paper

VLM Fine-Tuning for End-to-End Combinatorial Optimization

Sep 29, 2026

This study addresses the limitations of text-serialized representations in capturing spatial and relational structures within combinatorial optimization problems by proposing a general-purpose vision-language solver. The proposed method integrates textual and visual inputs, incorporating gold-standard-free visual representations, and employs a vision-language model (VLM) jointly optimized through supervised fine-tuning and verifier-guided reinforcement learning. Experimental results demonstrate that on complex tasks such as the Capacitated Vehicle Routing Problem (CVRP), visual information significantly enhances solution quality, with this advantage becoming increasingly pronounced as problem scale grows. These findings reveal the core value of visual representations in large-scale combinatorial optimization.

0 citationsRead paper