Selected work

Representative Papers

SYNAPSE: Empowering LLM Agents with Episodic-Semantic Memory via Spreading Activation

Jan 06, 2026arXiv.org

Existing retrieval-augmented approaches struggle to handle the fragmented nature of long-term agent memory and inadequately support complex temporal and multi-hop reasoning tasks. This work proposes a dynamic schematic memory architecture that emulates the spreading activation mechanism from cognitive science, dynamically selecting relevant subgraphs through lateral inhibition and temporal decay to enable synergistic retrieval of semantic and episodic memory. By moving beyond static vector similarity, the model mitigates the “context tunneling” problem and integrates geometric embedding with activation-driven graph traversal to form a tripartite hybrid retrieval strategy. Evaluated on the LoCoMo benchmark, the proposed method significantly outperforms current state-of-the-art approaches, demonstrating superior performance in complex reasoning scenarios.

1 citationsRead paper

Who is Responsible? The Data, Models, Users or Regulations? Responsible Generative AI for a Sustainable Future

Jan 15, 2025

This paper addresses the implementation gap in ethical governance of generative AI (Gen AI) in the post-ChatGPT era. Methodologically, it introduces the first end-to-end Responsible Gen AI (RAI) practice framework—spanning governance, technology, evaluation, and deployment—integrating philosophical responsibility theory, eXplainable AI (XAI), benchmark alignment, cross-sector application modeling, and KPI-based quantitative assessment. It pioneers an AI-readiness-oriented testbed evaluation methodology and establishes a comprehensive RAI Key Performance Indicator (KPI) system. Contributions include: (1) systematically bridging the chasm between normative ethical principles and engineering practice while redefining accountability structures; and (2) releasing an open-source resource repository—including standards, tools, and benchmark datasets—to provide researchers, policymakers, and industry practitioners with scalable, reusable, and trustworthy implementation guidance.

1 citationsRead paper

MLLM-Routed Heterogeneous Ensembles for Robust Cross-Dataset Image Classification

Aug 13, 2026

This work addresses the limited generalization of existing image classification models in cross-dataset and cross-domain scenarios. The authors propose a heterogeneous vision model ensemble system that leverages a multimodal large language model (MLLM) as a dynamic router to adaptively select the optimal backbone network—such as ResNet, self-supervised models, or vision-language models—at the sample level based on the input image. This approach enables seamless integration of new knowledge and unification of multi-source label spaces without requiring retraining. Evaluated across multiple heterogeneous datasets, the method achieves performance comparable to routers trained specifically for each task, while significantly enhancing cross-domain adaptability, interpretability, and scalability.

0 citationsRead paper

Hybrid-Policy Self-Editing for Composable Unstructured Knowledge Editing

Aug 12, 2026

This work addresses the challenge that large language models struggle with atomic fact-based question answering and multi-hop reasoning after injecting unstructured knowledge, primarily due to poor compositional generalization of newly acquired facts. To overcome this limitation, the paper proposes a Hybrid Policy Self-Editing (HPSE) approach that leverages a combination of on-policy and off-policy rollouts to enable unsupervised, active self-distillation for knowledge editing. HPSE precisely inserts missing factual knowledge in regions where the student model’s trajectory coverage is insufficient. Experimental results demonstrate that HPSE significantly improves post-editing performance across four prominent large language models and two distinct editors, enhancing both factual accuracy in question answering and multi-hop reasoning capabilities while effectively promoting the compositional use of newly integrated knowledge.

0 citationsRead paper

Multitask Scanning Probe Microscopy

Aug 10, 2026

This work addresses the low efficiency and high risk of probe and sample damage in large-scale, multimodal scanning probe microscopy by proposing a closed-loop autonomous workflow. The approach leverages multitask Gaussian process modeling to capture both spatial and cross-modal correlations, extending active learning from spatial sampling to dynamic selection of measurement modalities for non-colocated multimodal characterization. By integrating minimally invasive imaging with contact-based measurements within an automated large-sample atomic force microscopy platform, the method enables efficient, simultaneous acquisition of tapping-mode topography and DART (Dual AC Resonance Tracking) response maps across an AlScN composition-gradient wafer. This significantly enhances the throughput and intelligence of large-area multimodal nanoscale characterization.

0 citationsRead paper

Recent publications

Latest Papers

MLLM-Routed Heterogeneous Ensembles for Robust Cross-Dataset Image Classification

Aug 13, 2026

This work addresses the limited generalization of existing image classification models in cross-dataset and cross-domain scenarios. The authors propose a heterogeneous vision model ensemble system that leverages a multimodal large language model (MLLM) as a dynamic router to adaptively select the optimal backbone network—such as ResNet, self-supervised models, or vision-language models—at the sample level based on the input image. This approach enables seamless integration of new knowledge and unification of multi-source label spaces without requiring retraining. Evaluated across multiple heterogeneous datasets, the method achieves performance comparable to routers trained specifically for each task, while significantly enhancing cross-domain adaptability, interpretability, and scalability.

0 citationsRead paper

Hybrid-Policy Self-Editing for Composable Unstructured Knowledge Editing

Aug 12, 2026

This work addresses the challenge that large language models struggle with atomic fact-based question answering and multi-hop reasoning after injecting unstructured knowledge, primarily due to poor compositional generalization of newly acquired facts. To overcome this limitation, the paper proposes a Hybrid Policy Self-Editing (HPSE) approach that leverages a combination of on-policy and off-policy rollouts to enable unsupervised, active self-distillation for knowledge editing. HPSE precisely inserts missing factual knowledge in regions where the student model’s trajectory coverage is insufficient. Experimental results demonstrate that HPSE significantly improves post-editing performance across four prominent large language models and two distinct editors, enhancing both factual accuracy in question answering and multi-hop reasoning capabilities while effectively promoting the compositional use of newly integrated knowledge.

0 citationsRead paper

Multitask Scanning Probe Microscopy

Aug 10, 2026

This work addresses the low efficiency and high risk of probe and sample damage in large-scale, multimodal scanning probe microscopy by proposing a closed-loop autonomous workflow. The approach leverages multitask Gaussian process modeling to capture both spatial and cross-modal correlations, extending active learning from spatial sampling to dynamic selection of measurement modalities for non-colocated multimodal characterization. By integrating minimally invasive imaging with contact-based measurements within an automated large-sample atomic force microscopy platform, the method enables efficient, simultaneous acquisition of tapping-mode topography and DART (Dual AC Resonance Tracking) response maps across an AlScN composition-gradient wafer. This significantly enhances the throughput and intelligence of large-area multimodal nanoscale characterization.

0 citationsRead paper

Algorithmic Asymmetry in Zero-Sum Games: Unilateral Recovery of Fast Convergence Against a Slow Opponent

Aug 10, 2026

In zero-sum games, achieving fast convergence of the overall learning dynamics poses a significant challenge when one player employs standard gradient descent (GD) with slow convergence. This work proposes a modified Alternating Optimistic Gradient Descent (AOGD) method that enables the other player to unilaterally compensate for the slower learner without requiring coordinated algorithm selection. For the first time, it is shown that under this asymmetric setting, the time-averaged strategies still converge to the Nash equilibrium at an $O(1/T)$ rate, substantially improving upon the $O(1/\sqrt{T})$ rate typical of symmetric algorithms. This result breaks the prevailing theoretical reliance on mutual coordination between players’ learning algorithms.

0 citationsRead paper

SuperNeuroMAT: An Efficient Matrix-based Simulator for Spiking Neural Networks

Aug 09, 2026

This work addresses the limitations of existing spiking neural network (SNN) simulation frameworks—namely, their inadequate speed, usability, and generality—which hinder broader adoption in energy-efficient AI and brain-inspired computing. To overcome these challenges, we introduce SuperNeuroMAT, an open-source, Python-based SNN simulator that pioneers a matrix-based leaky integrate-and-fire (LIF) neuron model and a hybrid dense-sparse execution architecture. This design enables unified support for both high-performance simulation and general neuromorphic algorithms, such as shortest-path computation and arithmetic operations. Distributed via PyPI, SuperNeuroMAT efficiently simulates networks ranging from tens of thousands to over one hundred thousand neurons on standard hardware. Benchmark evaluations demonstrate its superior runtime speed and memory efficiency across varying network scales and connection densities compared to NEST, Brian2, BindsNET, and snnTorch, with successful applications in image classification, graph neural networks, and event-camera data processing.

0 citationsRead paper