Institution profile

Intelligent Game and Decision Lab

Research institutionasia · cn
Research library7linked papers
Opportunities0open roles
Selected work

Representative Papers

PreAct-Nav: Agentic Reasoning Before Action for Urban Navigation

Oct 04, 2026

This study addresses the limitation of existing agent navigation systems in translating long-term goals into coherent local decisions, particularly their lack of foresight regarding action consequences and future states. To this end, we propose PreAct-Nav, a framework that equips agents with anticipatory reasoning and error-correction capabilities under a frozen policy. Specifically, the method constructs a predictive world sandbox using an Action-Conditional World Model (AC-WM), leverages Vision-Language Models (VLMs) to anchor mid-range subgoals for reasoning, and introduces a persistent memory module to enable dynamic updates. Experimental results demonstrate that PreAct-Nav significantly improves action selection accuracy in long-distance, multi-turn scenarios and enhances navigation robustness within complex urban environments.

0 citationsRead paper

A Posteriori Error Analysis for Decoupled Neural Approximations of Fully Coupled FBSDEs with Control Mismatch

Jun 28, 2026

This work addresses the control mismatch arising in decoupled neural approximations of fully coupled forward–backward stochastic differential equations (FBSDEs), where inconsistency between the auxiliary control and the backward component compromises solution accuracy, alongside the lack of reliable a posteriori error estimates. For the first time, the control mismatch is explicitly incorporated into the a posteriori error analysis framework for neural approximations. By introducing an auxiliary control process to satisfy the decoupling requirement inherent in deep learning implementations, and leveraging tools from stochastic analysis and numerical FBSDE theory, the authors establish stability estimates in both continuous and discrete time, yielding computable error bounds that depend solely on terminal defects, pathwise residuals, and the mismatch term. Numerical experiments on a linear-quadratic FBSDE with an explicit solution and a high-dimensional Burgers-type FBSDE without a reference solution demonstrate the effectiveness of the proposed error indicators, showing that penalizing control mismatch significantly enhances the consistency and reproducibility of numerical solutions.

0 citationsRead paper

Ternary Gamma Semirings: From Neural Implementation to Categorical Foundations

Mar 15, 2026

Standard neural networks completely fail on compositional generalization tasks (0% accuracy) due to their inability to intrinsically model structured regularities. This work establishes, for the first time, a rigorous correspondence between neural network generalization and ternary Gamma semirings from abstract algebra, leveraging this algebraic framework to impose logical constraints that guide the network to learn representations satisfying symmetry, idempotence, and majority rules. Under identical architectures, these constraints elevate compositional generalization accuracy from 0% to 100%. Furthermore, the learned representations are proven to be isomorphic to the unique Boolean quaternary ternary Gamma semiring (|T| = 4, |Γ| = 1). This study pioneers a novel neuro-symbolic direction—computational Gamma algebra—bridging deep learning with formal algebraic structures.

0 citationsRead paper

FRBNet: Revisiting Low-Light Vision through Frequency-Domain Radial Basis Network

Oct 27, 2025

Low-light imaging severely degrades downstream vision tasks due to significant illumination deficiency. To address this, we propose the Frequency-domain Radial Basis Network (FRBNet), the first framework to model low-light enhancement explicitly in the frequency domain and formulate it as an end-to-end trainable pipeline grounded in an extended Lambertian reflectance model. Our key contributions are: (1) a theoretical frequency-domain channel-ratio principle that characterizes illumination-invariant feature responses; and (2) a learnable, structured frequency-domain filtering module enabling plug-and-play feature enhancement without architectural or loss-function modifications. Extensive experiments on low-light object detection and nighttime semantic segmentation demonstrate consistent improvements—+2.2 mAP and +2.9 mIoU—validating FRBNet’s effectiveness, generalizability, and practical plug-and-play utility across diverse vision tasks.

0 citationsRead paper
Recent publications

Latest Papers

PreAct-Nav: Agentic Reasoning Before Action for Urban Navigation

Oct 04, 2026

This study addresses the limitation of existing agent navigation systems in translating long-term goals into coherent local decisions, particularly their lack of foresight regarding action consequences and future states. To this end, we propose PreAct-Nav, a framework that equips agents with anticipatory reasoning and error-correction capabilities under a frozen policy. Specifically, the method constructs a predictive world sandbox using an Action-Conditional World Model (AC-WM), leverages Vision-Language Models (VLMs) to anchor mid-range subgoals for reasoning, and introduces a persistent memory module to enable dynamic updates. Experimental results demonstrate that PreAct-Nav significantly improves action selection accuracy in long-distance, multi-turn scenarios and enhances navigation robustness within complex urban environments.

0 citationsRead paper

A Posteriori Error Analysis for Decoupled Neural Approximations of Fully Coupled FBSDEs with Control Mismatch

Jun 28, 2026

This work addresses the control mismatch arising in decoupled neural approximations of fully coupled forward–backward stochastic differential equations (FBSDEs), where inconsistency between the auxiliary control and the backward component compromises solution accuracy, alongside the lack of reliable a posteriori error estimates. For the first time, the control mismatch is explicitly incorporated into the a posteriori error analysis framework for neural approximations. By introducing an auxiliary control process to satisfy the decoupling requirement inherent in deep learning implementations, and leveraging tools from stochastic analysis and numerical FBSDE theory, the authors establish stability estimates in both continuous and discrete time, yielding computable error bounds that depend solely on terminal defects, pathwise residuals, and the mismatch term. Numerical experiments on a linear-quadratic FBSDE with an explicit solution and a high-dimensional Burgers-type FBSDE without a reference solution demonstrate the effectiveness of the proposed error indicators, showing that penalizing control mismatch significantly enhances the consistency and reproducibility of numerical solutions.

0 citationsRead paper

Ternary Gamma Semirings: From Neural Implementation to Categorical Foundations

Mar 15, 2026

Standard neural networks completely fail on compositional generalization tasks (0% accuracy) due to their inability to intrinsically model structured regularities. This work establishes, for the first time, a rigorous correspondence between neural network generalization and ternary Gamma semirings from abstract algebra, leveraging this algebraic framework to impose logical constraints that guide the network to learn representations satisfying symmetry, idempotence, and majority rules. Under identical architectures, these constraints elevate compositional generalization accuracy from 0% to 100%. Furthermore, the learned representations are proven to be isomorphic to the unique Boolean quaternary ternary Gamma semiring (|T| = 4, |Γ| = 1). This study pioneers a novel neuro-symbolic direction—computational Gamma algebra—bridging deep learning with formal algebraic structures.

0 citationsRead paper

FRBNet: Revisiting Low-Light Vision through Frequency-Domain Radial Basis Network

Oct 27, 2025

Low-light imaging severely degrades downstream vision tasks due to significant illumination deficiency. To address this, we propose the Frequency-domain Radial Basis Network (FRBNet), the first framework to model low-light enhancement explicitly in the frequency domain and formulate it as an end-to-end trainable pipeline grounded in an extended Lambertian reflectance model. Our key contributions are: (1) a theoretical frequency-domain channel-ratio principle that characterizes illumination-invariant feature responses; and (2) a learnable, structured frequency-domain filtering module enabling plug-and-play feature enhancement without architectural or loss-function modifications. Extensive experiments on low-light object detection and nighttime semantic segmentation demonstrate consistent improvements—+2.2 mAP and +2.9 mIoU—validating FRBNet’s effectiveness, generalizability, and practical plug-and-play utility across diverse vision tasks.

0 citationsRead paper