2026
SpecMind: Enabling Spectrum Intelligence via Multi-Agent Hybrid Retrieval-Augmented Generation
A multi-agent hybrid RAG system that routes spectrum questions to specialised agents over license records, regulations, and proceeding filings, then composes one grounded answer.
SpecMind: Enabling Spectrum Intelligence via Multi-Agent Hybrid Retrieval-Augmented Generation
A multi-agent hybrid RAG system that routes spectrum questions to specialised agents over license records, regulations, and proceeding filings, then composes one grounded answer.
The Past Is Prologue: A Plug-in Controller for Selective Updates in Sequentially Evolving LLM Memory
A plug-in controller that decides whether each candidate memory update is worth deploying, adding 2.7 to 4.6 accuracy points on top of existing memory updaters.
The Past Is Prologue: A Plug-in Controller for Selective Updates in Sequentially Evolving LLM Memory
A plug-in controller that decides whether each candidate memory update is worth deploying, adding 2.7 to 4.6 accuracy points on top of existing memory updaters.
Is One Score Enough? Rethinking the Evaluation of Sequentially Evolving LLM Memory
A diagnostic framework for LLM memory that separates online utility, generalization, backward transfer, and forgetting, showing that headline accuracy often hides substantial forgetting.
Is One Score Enough? Rethinking the Evaluation of Sequentially Evolving LLM Memory
A diagnostic framework for LLM memory that separates online utility, generalization, backward transfer, and forgetting, showing that headline accuracy often hides substantial forgetting.
Training-Free Speedup for Retrieval-Augmented Generation with Staged Parallel Speculation
A training-free RAG framework that overlaps retrieval with generation and picks among parallel candidate chunks by self-consistency, cutting latency by up to 57%.
Training-Free Speedup for Retrieval-Augmented Generation with Staged Parallel Speculation
A training-free RAG framework that overlaps retrieval with generation and picks among parallel candidate chunks by self-consistency, cutting latency by up to 57%.
SceneTrans: A Benchmark for Action-Aware Scene Transformation Captioning
A simulation-built benchmark for action-aware image difference captioning, with object-level spatial grounding and displacement distances that current proprietary VLMs handle poorly.
SceneTrans: A Benchmark for Action-Aware Scene Transformation Captioning
A simulation-built benchmark for action-aware image difference captioning, with object-level spatial grounding and displacement distances that current proprietary VLMs handle poorly.
2023
A hybrid MCDM model with Monte Carlo simulation to improve decision-making stability and reliability
A hybrid MCDM model combining CRITIC, MABAC, and k-means, whose Monte Carlo step resolves disagreeing clustering runs; demonstrated on transport safety across the ASEAN region.
A hybrid MCDM model with Monte Carlo simulation to improve decision-making stability and reliability
A hybrid MCDM model combining CRITIC, MABAC, and k-means, whose Monte Carlo step resolves disagreeing clustering runs; demonstrated on transport safety across the ASEAN region.
Learning Generalized Representations for Open-Set Temporal Action Localization
A one-stage open-set temporal action localization framework trained under sharpness-aware minimization, setting state of the art on THUMOS14 and ActivityNet1.3.
Learning Generalized Representations for Open-Set Temporal Action Localization
A one-stage open-set temporal action localization framework trained under sharpness-aware minimization, setting state of the art on THUMOS14 and ActivityNet1.3.