arXiv subjects

Zhimin Lin

Publications and source records attributed to Zhimin Lin.

6 recordsLinked to original sources

arXiv2026

Coarse Indexing, Fine Evidence: Decoupling Temporal Granularity in Long-Video RAG

Graph-based retrieval-augmented generation (RAG) provides a scalable paradigm for long-video understanding, but existing systems typically inherit a fixed temporal granularity from video segmentation when constructing their retrieval index. We argue that this design unnecessarily couples indexing granularity with evidence granularity: coarse representations can often suffice for locating relevant temporal regions, while fine-grained evidence remains important for downstream reasoning. We propose \textbf{Density-Aware Graph Construction (DAGC)}, a training-free approach that decouples a query-independent coarse retrieval index from the original fine-grained evidence space. DAGC constructs a compact, density-adaptive graph index by merging visually redundant neighboring chunks, while preserving mappings to the original temporal units. Retrieved coarse regions are subsequently expanded back to the original chunk granularity for fine-grained evidence refinement and answer generation. Experiments on MLVU, VideoMME, and LongVideoBench show that DAGC retains only about 40--50\% of the original graph nodes and achieves $1.3$--$1.7\times$ end-to-end wall-clock acceleration while preserving approximately 99\% of the original QA performance. The gains transfer across different LVLM backbones and video RAG pipelines, suggesting that long-video RAG need not maintain the same temporal granularity for indexing and evidence reasoning.

cs.CV↗

arXiv2026

Agent-as-Peer-Debriefer: A Multi-Agent Framework with Perspective-Based Refinement for Qualitative Analysis

Large language models (LLMs) are increasingly used for qualitative data analysis (QDA), yet their outputs often miss the depth and nuance of human analysis. We argue this gap reflects a missing credibility practice from human QDA: peer debriefing, in which an analyst seeks feedback from a disinterested peer and uses it to refine their coding. To bring this practice into LLM-assisted QDA, we propose Agent-as-Peer-Debriefer, a multi-agent QDA framework that builds peer debriefing into key coding steps. In our framework, a Hierarchical Coding Agent follows the standard QDA process to generate codes, sub-themes, and themes, along with self-explanations and reflection memos. It then shares these outputs with three Peer-Debriefing Agents, each applying a distinct analytical perspective (Theory-Driven, Data-Driven, or Applied) and refining the codes by keeping, renaming, reassigning, merging, or splitting them. These perspectives are drawn from established human QDA practices that generalize across domains and datasets. To evaluate the framework, we test it on three datasets across two domains with three LLMs, measuring semantic similarity to human-annotated codes. Across all settings, perspective-based, peer-debriefing refinement aligns more closely with human codes than a single-LLM baseline, and an ablation further shows the gain is not merely from additional refinement. The three perspectives also produce distinct trade-offs, showing that the choice of perspective is a meaningful and controllable design decision. More broadly, these findings suggest that simulating peer debriefing with explicit perspectives is a promising route to more credible LLM-assisted QDA.

cs.AI↗

arXiv2026

When to Vote, When to Rewrite: Disagreement-Guided Strategy Routing for Test-Time Scaling

Large Reasoning Models (LRMs) achieve strong performance on mathematical reasoning tasks but remain unreliable on challenging instances. Existing test-time scaling methods, such as repeated sampling, self-correction, and tree search, improve performance at the cost of increased computation, yet often exhibit diminishing returns on hard problems. We observe that output disagreement is strongly correlated with instance difficulty and prediction correctness, providing a useful signal for guiding instance-level strategy selection at test time. Based on this insight, we propose a training-free framework that formulates test-time scaling as an instance-level routing problem, rather than allocating more computation within a single strategy, dynamically selecting among different scaling strategies based on output disagreement. The framework applies lightweight resolution for consistent cases, majority voting for moderate disagreement, and rewriting-based reformulation for highly ambiguous instances. Experiments on seven mathematical benchmarks and three models show that our method improves accuracy by 3% - 7% while reducing sampling cost compared to existing approaches.

cs.AI↗

arXiv2026

LongCat-Flash-Prover: Advancing Native Formal Reasoning via Agentic Tool-Integrated Reinforcement Learning

We introduce LongCat-Flash-Prover, a flagship 560-billion-parameter open-source Mixture-of- Experts (MoE) model that advances Native Formal Reasoning in Lean4 through agentic tool-integrated reasoning (TIR). We decompose the native formal reasoning task into three independent formal capabilities, i.e., auto-formalization, sketching, and proving. To facilitate these capabilities, we propose a Hybrid-Experts Iteration Framework to expand high-quality task trajectories, including generating a formal statement based on a given informal problem, producing a whole-proof directly from the statement, or a lemma-style sketch. During agentic RL, we present a Hierarchical Importance Sampling Policy Optimization (HisPO) algorithm, which aims to stabilize the MoE model training on such long-horizon tasks. It employs a gradient masking strategy that accounts for the policy staleness and the inherent train-inference engine discrepancies at both sequence and token levels. Additionally, we also incorporate theorem consistency and legality detection mechanisms to eliminate reward hacking issues. Extensive evaluations show that our LongCat-Flash-Prover sets a new state-of-the-art for open-weights models in both auto-formalization and theorem proving. Demonstrating remarkable sample efficiency, it achieves a 97.1% pass rate on MiniF2F-Test using only 72 inference budget per problem. On more challenging benchmarks, it solves 70.8% of ProverBench and 41.5% of PutnamBench with no more than 220 attempts per problem, significantly outperforming existing open-weights baselines.

cs.AI↗

arXiv2025

Solution for Meta KDD Cup'25: A Comprehensive Three-Step Framework for Vision Question Answering

Vision Large Language Models (VLLMs) have improved multi-modal understanding and visual question answering (VQA), but still suffer from hallucinated answers. Multi-modal Retrieval-Augmented Generation (RAG) helps address these issues by incorporating external information, yet challenges remain in visual context comprehension, multi-source retrieval, and multi-turn interactions. To address these challenges, Meta constructed the CRAG-MM benchmark and launched the CRAG-MM Challenge at KDD Cup 2025, which consists of three tasks. This paper describes the solutions of all tasks in Meta KDD Cup'25 from BlackPearl team. We use a single model for each task, with key methods including data augmentation, RAG, reranking, and multi-task fine-tuning. Our solution achieve automatic evaluation rankings of 3rd, 3rd, and 1st on the three tasks, and win second place in Task3 after human evaluation.

cs.IR↗

arXiv2021

Multimodal Feature Fusion for Video Advertisements Tagging Via Stacking Ensemble

Automated tagging of video advertisements has been a critical yet challenging problem, and it has drawn increasing interests in last years as its applications seem to be evident in many fields. Despite sustainable efforts have been made, the tagging task is still suffered from several challenges, such as, efficiently feature fusion approach is desirable, but under-explored in previous studies. In this paper, we present our approach for Multimodal Video Ads Tagging in the 2021 Tencent Advertising Algorithm Competition. Specifically, we propose a novel multi-modal feature fusion framework, with the goal to combine complementary information from multiple modalities. This framework introduces stacking-based ensembling approach to reduce the influence of varying levels of noise and conflicts between different modalities. Thus, our framework can boost the performance of the tagging task, compared to previous methods. To empirically investigate the effectiveness and robustness of the proposed framework, we conduct extensive experiments on the challenge datasets. The obtained results suggest that our framework can significantly outperform related approaches and our method ranks as the 1st place on the final leaderboard, with a Global Average Precision (GAP) of 82.63%. To better promote the research in this field, we will release our code in the final version.

cs.CV↗