arXiv subjects

Zhencan Peng

Publications and source records attributed to Zhencan Peng.

4 recordsLinked to original sources

arXiv2026

Conformal Path Reasoning: Trustworthy Knowledge Graph Question Answering via Path-Level Calibration

Knowledge Graph Question Answering (KGQA) offers grounded, interpretable reasoning, but existing methods often fail to provide reliable coverage guarantees over retrieved answers. While Conformal Prediction (CP) offers a principled framework for producing prediction sets with statistical guarantees, prior conformal KGQA methods suffer from two critical pitfalls: violated coverage guarantees due to invalid calibration, and weak score discriminability that yields excessively large prediction sets. We propose Conformal Path Reasoning (CPR), a novel trustworthy KGQA framework built on two key innovations. First, query-level conformal calibration over path-level scores preserves exchangeability to ensure valid coverage guarantees. Second, we introduce the Residual Conformal Value Network (RCVNet), a lightweight module trained via PUCT-guided exploration to learn discriminative path-level nonconformity scores. Extensive experiments show that CPR significantly improves the Empirical Coverage Rate by 45% while reducing prediction set size by 52% on average over conformal baselines across benchmark datasets, highlighting its effectiveness for reliable conformal reasoning over knowledge graphs.

cs.CL↗

arXiv2026

CSAttention: Centroid-Scoring Attention for Accelerating LLM Inference

Long-context LLMs increasingly rely on extended, reusable prefill prompts for agents and domain Q&A, pushing attention and KV-cache to become the dominant decode-time bottlenecks. While sparse attention reduces computation and transfer costs, it often struggles to maintain accuracy at high sparsity levels due to the inherent distribution shift between Queries and Keys. We propose Centroid-Scoring Attention (CSAttention), a training-free sparse attention method optimized for high-throughput serving of reusable contexts. CSAttention adopts a storage-for-computation strategy tailored to the offline-prefill/online-decode setting: it front-loads computation into a one-time offline prefill phase that can be amortized across multiple queries, while aggressively optimizing per-step decoding latency. Specifically, CSAttention constructs query-centric lookup tables during offline prefill, whose size remains fixed during decoding, and enables online decoding to replace full-context scans with efficient table lookups and GPU-friendly score accumulation. Extensive experiments demonstrate that CSAttention achieves near-identical accuracy to full attention. Under high sparsity (95%) and long-context settings (32K-128K), CSAttention consistently outperforms state-of-the-art sparse attention methods in both model accuracy and inference speed, achieving up to 4.6x inference speedup over the most accurate baseline at a context length of 128K.

cs.LG↗

arXiv2025

Cache Mechanism for Agent RAG Systems

Recent advances in Large Language Model (LLM)-based agents have been propelled by Retrieval-Augmented Generation (RAG), which grants the models access to vast external knowledge bases. Despite RAG's success in improving agent performance, agent-level cache management, particularly constructing, maintaining, and updating a compact, relevant corpus dynamically tailored to each agent's need, remains underexplored. Therefore, we introduce ARC (Agent RAG Cache Mechanism), a novel, annotation-free caching framework that dynamically manages small, high-value corpora for each agent. By synthesizing historical query distribution patterns with the intrinsic geometry of cached items in the embedding space, ARC automatically maintains a high-relevance cache. With comprehensive experiments on three retrieval datasets, our experimental results demonstrate that ARC reduces storage requirements to 0.015% of the original corpus while offering up to 79.8% has-answer rate and reducing average retrieval latency by 80%. Our results demonstrate that ARC can drastically enhance efficiency and effectiveness in RAG-powered LLM agents.

cs.CL↗

arXiv2025

Near-Duplicate Text Alignment under Weighted Jaccard Similarity

Near-duplicate text alignment is the task of identifying, among the texts in a corpus, all the subsequences (substrings) that are similar to a given query. Traditional approaches rely on seeding-extension-filtering heuristics, which lack accuracy guarantees and require many hard-to-tune parameters. Recent methods leverage min-hash techniques under a hash-based framework: group subsequences by their min-hash, and for any query, find all sketches similar to the query's sketch. These methods guarantee to report all subsequences whose estimated unweighted Jaccard similarity with the query exceeds a user-provided threshold and are efficient. However, they fail to account for token importance or frequency, which limits their use in real scenarios where tokens carry weights, such as TF-IDF. To address this, we propose MONO, an approach that supports weighted Jaccard similarity using consistent weighted sampling. MONO achieves optimality within the hash-based framework. For example, when token weights are proportional to frequencies, MONO generates O(n + n log f) groups in expectation for a text of length n, where f is the maximum token frequency. Each group takes O(1) space and represents a few subsequences sharing the same sampling. We prove this bound is tight: any algorithm must produce Omega(n + n log f) groups in expectation in the worst case. Experiments show that MONO outperforms the state of the art by up to 26x in index construction time, reduces index size by up to 30 percent, and improves query latency by up to 3x, while scaling well.

cs.DB↗