Institution profile

East China Normal University

Academic institutionasia · cn
Official website
Research library1,283linked papers
Opportunities0open roles
Selected work

Representative Papers

Towards Understanding Retrieval Accuracy and Prompt Quality in RAG Systems

Nov 29, 2024arXiv.org

The impact of key design decisions—RAG activation, retrieval granularity, and knowledge integration strategy—on RAG system performance remains poorly understood. Method: We conduct systematic ablation studies across three code/qa benchmarks and two state-of-the-art LLMs, quantitatively evaluating how document type, recall rate, document selection strategy, and prompt engineering jointly affect answer correctness and confidence via multi-dimensional analysis, cross-model/dataset comparison, and joint prompt-retrieval analysis. Contribution/Results: We identify precise interaction patterns and operational boundaries among these factors and propose nine actionable, empirically grounded guidelines for diagnosing and optimizing RAG failures. Our findings significantly improve RAG system stability, debuggability, and reliability, offering rigorous empirical evidence and a principled methodology to support the engineering deployment of LLM-augmented systems.

7 citationsRead paper

Explore with Long-term Memory: A Benchmark and Multimodal LLM-based Reinforcement Learning Framework for Embodied Exploration

Jan 11, 2026arXiv.org

This work addresses the common oversight in existing embodied intelligence approaches—the lack of synergistic integration between exploration processes and long-term memory—which hinders lifelong learning and complex, long-horizon decision-making. To bridge this gap, the authors propose the LMEE framework, introducing LMEE-Bench, the first evaluation benchmark that jointly incorporates multi-object navigation and memory-based question answering, emphasizing both procedural and outcome-oriented assessment. They further present MemoryExplorer, a method that unifies active exploration and memory retrieval by leveraging a multimodal large language model within a reinforcement learning paradigm, guided by a multi-task reward mechanism encompassing action prediction, frontier selection, and question answering. Experiments demonstrate that MemoryExplorer significantly outperforms current methods on long-horizon embodied tasks, markedly improving both exploration efficiency and memory utilization.

3 citationsRead paper

UniCorn: Towards Self-Improving Unified Multimodal Models through Self-Generated Supervision

Jan 06, 2026arXiv.org

Although unified multimodal models excel at cross-modal understanding, they struggle to translate this capability into high-quality, controllable generation—a limitation akin to “conductive aphasia.” To address this, this work proposes UniCorn, a framework that, for the first time, enhances generative capacity within a unified model through fully self-supervised learning. UniCorn partitions the model into three roles—proposer, solver, and critic—to establish a self-play mechanism that generates high-quality interactive data. It further introduces cognitive mode reconstruction to explicitly convert implicit understanding into actionable generative signals. The proposed method achieves state-of-the-art performance across six image generation benchmarks, setting new records on TIIF, DPG, CompBench, and UniCycle, while improving scores by 5.0 and 6.5 points on WISE and OneIG, respectively.

3 citationsRead paper

CATCH: Channel-Aware multivariate Time Series Anomaly Detection via Frequency Patching

Oct 16, 2024arXiv.org

Existing reconstruction-based methods for heterogeneous subsequence anomaly detection in multivariate time series struggle to simultaneously achieve fine-grained frequency modeling and dynamic inter-channel dependency learning. Method: We propose a frequency-domain patching reconstruction framework. It introduces *frequency patching*—a novel mechanism that partitions the frequency spectrum into patches for band-level fine-grained feature representation—and a *Channel Fusion Module (CFM)* that employs patch-wise masking and masked attention to adaptively learn and cluster inter-channel dependencies. The framework integrates frequency-domain transformation, patch-based encoding, masked attention, and a two-level multi-objective optimization. Results: Evaluated on 10 real-world and 12 synthetic datasets, our method achieves state-of-the-art performance, significantly improving detection rates for subtle, localized, and cross-channel anomalies.

3 citationsRead paper

E2LLM: Encoder Elongated Large Language Models for Long-Context Understanding and Reasoning

Sep 10, 2024arXiv.org

Addressing the “impossible triangle” in long-context modeling for large language models—where high performance, computational efficiency, and compatibility with pretrained models are mutually exclusive—this paper proposes an encoder-extension architecture. It freezes a pretrained text encoder (e.g., BERT or CLIP) to compress long inputs into soft prompts, then introduces a learnable adapter coupled with dual-objective adaptation training: reconstruction loss and long-context instruction fine-tuning. This enables decoder-only LLMs to model extended contexts efficiently without modifying the backbone architecture. The method is fully plug-and-play compatible with existing decoder-only models. Evaluated on multi-turn dialogue and long-document summarization, it outperforms state-of-the-art long-context approaches, achieving a 2.3× inference speedup and 37% reduction in GPU memory consumption—demonstrating strong performance, high efficiency, and seamless integration.

3 citationsRead paper
Recent publications

Latest Papers