arXiv subjects

Kunpeng Zhang

Publications and source records attributed to Kunpeng Zhang.

At least 19 recordsLinked to original sources

arXiv2026

TreeHop: Efficient Embedding-Level Query Rewriter

Retrieval-augmented generation (RAG) systems face significant challenges in multi-hop question answering (MHQA), where complex queries require synthesizing information across multiple document chunks. Existing approaches typically rely on iterative LLM-based query rewriting and routing, resulting in high computational costs due to repeated LLM invocations and multi-stage processes. To address these limitations, we propose TreeHop, an embedding-level framework without the need for LLMs in query refinement. TreeHop dynamically updates query embeddings by fusing semantic information from prior queries and retrieved documents, enabling iterative retrieval through embedding-space operations alone. This method replaces the traditional "Retrieve-Rewrite-Vectorize-Retrieve" cycle with a streamlined "Retrieve-Embed-Retrieve" loop, significantly reducing computational overhead. Moreover, a rule-based stopping criterion is introduced to further prune redundant retrievals, balancing efficiency and recall rate. Experimental results show that TreeHop rivals advanced RAG methods across four open-domain MHQA datasets, achieving comparable performance with only 2.2\%-29.4\% of the parameter size of concurrent solutions and reducing the query latency by 92.8\%-97.8\%. This makes TreeHop a faster and more cost-effective solution for low-resource or latency-sensitive deployment. For reproducibility purposes, codes and data are available here: https://github.com/allen-li1231/TreeHop-RAG

cs.IR↗

arXiv2026

M-RAG: Semantic Key-Value Indexing for Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) turns external documents into evidence for large language models. In practice, this is also a data access problem: a system must decide what to index, what to retrieve, and what evidence to place in the context under a token budget. Most RAG pipelines use text chunks for both lookup and generation. This couples two different objectives. Retrieval benefits from compact and discriminative records, while generation needs contextual and faithful evidence. As a result, small chunks may fragment answer-bearing information, whereas large chunks may introduce noise and waste the context budget. We propose M-RAG, a semantic key-value indexing layer for budget-constrained RAG query processing. M-RAG extracts meta-markers from complete documents, where each record contains a retrieval key, an information value, and provenance pointers. Online retrieval operates over the key field, which can be searched by dense vector retrieval or sparse lexical retrieval; the paired values are returned as generation payloads and assembled under the token budget. Provenance pointers further support coverage validation and position-aware context ordering. This design separates the physical index entry from the evidence payload without changing the underlying retriever or generator. Experiments on LongBench QA subtasks show that M-RAG achieves competitive or better accuracy than representative chunk-based baselines, especially under tight token budgets. Further analyses show high document coverage, stronger robustness under expanding candidate corpora, and lower online retrieval latency. These results suggest that semantic key-value indexing is a practical access method for RAG workloads.

cs.IR↗

arXiv2026

Enhancing Clinical Trial Patient Matching through Knowledge Augmentation and Reasoning with Multi-Agent

Matching patients effectively and efficiently for clinical trials is a significant challenge due to the complexity and variability of patient profiles and trial criteria. This paper introduces \textbf{Multi-Agent for Knowledge Augmentation and Reasoning (MAKAR)}, a novel multi-agent system that enhances patient-trial matching by integrating criterion augmentation with structured reasoning. MAKAR consistently improves performance by an average of 7\% across different datasets. Furthermore, it enables privacy-preserving deployment and maintains competitive performance when using smaller open-source models. Overall, MAKAR can contributes to more transparent, accurate, and privacy-conscious AI-driven patient matching.

cs.MA↗

arXiv2026

Beyond Isolated Investor: Predicting Startup Success via Roleplay-Based Collective Agents

Due to the high value and high failure rates of startups, predicting their success is a critical challenge. Existing approaches typically model startup success from a single decision-maker's perspective, overlooking the collective dynamics that dominate real-world venture capital (VC) decision-making. We propose SimVC-CAS, a collective agent system that simulates VC decisions as a multi-agent interaction process. By designing role-playing agents and a GNN-based supervised interaction module, we reformulate startup financing prediction as a group decision-making task, capturing both enterprise fundamentals and investor network dynamics. Each agent represents an investor with distinct traits and preferences, enabling heterogeneous evaluations and realistic information exchange over a graph-structured co-investment network. Using both proprietary and public VC data with strict anti-leakage controls, we show that SimVC-CAS significantly improves predictive performance, achieving approximately 25% relative improvement in average precision@10, while exhibiting consistency with real investor decisions. The interaction mechanism is particularly effective for network-central startups, confirming the importance of network in VC decision-making. Analysis of agents' reasoning for decision changes further reveals how network environment influence decision quality, demonstrating the system's interpretability. Our approach may generalize to broader group decision-making scenarios.

cs.AI↗

arXiv2026

LLM-Powered Silent Bug Fuzzing in Deep Learning Libraries via Versatile and Controlled Bug Transfer

Deep learning (DL) libraries are widely used in critical applications, where even subtle silent bugs can lead to serious consequences. While existing DL fuzzing techniques have made progress in detecting crashes, they inherently struggle to detect silent bugs due to the lack of effective test programs and corresponding oracles. Building on the observation that historical bug reports contain rich, underutilized information about silent bugs, we leverage large language models (LLMs) to perform versatile yet controlled bug transfer for silent bug fuzzing. Specifically, our approach uses LLMs to extract context-aware bug patterns from historical issues, match semantically related Application Programming Interfaces (APIs) using functionality-based embeddings, and synthesize test cases with customized oracles. This enables proactive detection of silent bugs by transferring high-risk contexts and oracle designs from known buggy APIs to functionally similar target APIs. To ensure the reliability of our context-aware bug transfer, we introduce an LLM-powered self-validation module that systematically evaluates the validity of each transferred bug instance. We implement this methodology in a tool named TransFuzz and evaluate it on three mainstream DL libraries: PyTorch, TensorFlow, and MindSpore. TransFuzz successfully discovers 79 previously unknown bugs (12 confirmed as Common Vulnerabilities and Exposures (CVEs)) in 10 bug types, demonstrating its effectiveness and generalizability in migrating DL library bug discovery capabilities.

cs.SE↗

arXiv2026

Decoding the Hook: A Multimodal LLM Framework for Analyzing the Hooking Period of Video Ads

Video-based ads are a vital medium for brands to engage consumers, with social media platforms leveraging user data to optimize ad delivery and boost engagement. A crucial but under-explored aspect is the 'hooking period', the first three seconds that capture viewer attention and influence engagement metrics. Analyzing this brief window is challenging due to the multimodal nature of video content, which blends visual, auditory, and textual elements. Traditional methods often miss the nuanced interplay of these components, requiring advanced frameworks for thorough evaluation. This study presents a framework using transformer-based multimodal large language models (MLLMs) to analyze the hooking period of video ads. It tests two frame sampling strategies, uniform random sampling and key frame selection, to ensure balanced and representative acoustic feature extraction, capturing the full range of design elements. The hooking video is processed by state-of-the-art MLLMs to generate descriptive analyses of the ad's initial impact, which are distilled into coherent topics using BERTopic for high-level abstraction. The framework also integrates features such as audio attributes and aggregated ad targeting information, enriching the feature set for further analysis. Empirical validation on large-scale real-world data from social media platforms demonstrates the efficacy of our framework, revealing correlations between hooking period features and key performance metrics like conversion per investment. The results highlight the practical applicability and predictive power of the approach, offering valuable insights for optimizing video ad strategies. This study advances video ad analysis by providing a scalable methodology for understanding and enhancing the initial moments of video advertisements.

cs.MM↗

arXiv2026

The Gaining Paths to Investment Success: Information-Driven LLM Graph Reasoning for Venture Capital Prediction

Most venture capital (VC) investments fail, while a few deliver outsized returns. Accurately predicting startup success requires synthesizing complex relational evidence, including company disclosures, investor track records, and investment network structures, through explicit reasoning to form coherent, interpretable investment theses. Traditional machine learning and graph neural networks both lack this reasoning capability. Large language models (LLMs) offer strong reasoning but face a modality mismatch with graphs. Recent graph-LLM methods target in-graph tasks where answers lie within the graph, whereas VC prediction is off-graph: the target exists outside the network. The core challenge is selecting graph paths that maximize predictor performance on an external objective while enabling step-by-step reasoning. We present MIRAGE-VC, a multi-perspective retrieval-augmented generation framework that addresses two obstacles: path explosion (thousands of candidate paths overwhelm LLM context) and heterogeneous evidence fusion (different startups need different analytical emphasis). Our information-gain-driven path retriever iteratively selects high-value neighbors, distilling investment networks into compact chains for explicit reasoning. A multi-agent architecture integrates three evidence streams via a learnable gating mechanism based on company attributes. Under strict anti-leakage controls, MIRAGE-VC achieves +5.0% F1 and +16.6% PrecisionAt5, and sheds light on other off-graph prediction tasks such as recommendation and risk assessment. Code: https://anonymous.4open.science/r/MIRAGE-VC-323F.

cs.AI↗

arXiv2025

Hold Onto That Thought: Assessing KV Cache Compression On Reasoning

Large language models (LLMs) have demonstrated remarkable performance on long-context tasks, but are often bottlenecked by memory constraints. Namely, the KV cache, which is used to significantly speed up attention computations, grows linearly with context length. A suite of compression algorithms has been introduced to alleviate cache growth by evicting unimportant tokens. However, several popular strategies are targeted towards the prefill phase, i.e., processing long prompt context, and their performance is rarely assessed on reasoning tasks requiring long decoding. In particular, short but complex prompts, such as those in benchmarks like GSM8K and MATH500, often benefit from multi-step reasoning and self-reflection, resulting in thinking sequences thousands of tokens long. In this work, we benchmark the performance of several popular compression strategies on long-reasoning tasks. For the non-reasoning Llama-3.1-8B-Instruct, we determine that no singular strategy fits all, and that performance is heavily influenced by dataset type. However, we discover that H2O and our decoding-enabled variant of SnapKV are dominant strategies for reasoning models, indicating the utility of heavy-hitter tracking for reasoning traces. We also find that eviction strategies at low budgets can produce longer reasoning traces, revealing a tradeoff between cache size and inference costs.

cs.CL↗

arXiv2025

PathReasoning: A multimodal reasoning agent for query-based ROI navigation on whole-slide images

Deciphering tumor microenvironment from Whole Slide Images (WSIs) is intriguing as it is key to cancer diagnosis, prognosis and treatment response. While these gigapixel images on one hand offer a comprehensive portrait of cancer, on the other hand, the extremely large size, as much as more than 10 billion pixels, make it challenging and time-consuming to navigate to corresponding regions to support diverse clinical inspection. Inspired by pathologists who conducted navigation on WSIs with a combination of sampling, reasoning and self-reflection, we proposed "PathReasoning", a multi-modal reasoning agent that iteratively navigates across WSIs through multiple rounds of reasoning and refinements. Specifically, starting with randomly sampled candidate regions, PathReasoning reviews current selections with self-reflection, reasoning over the correspondence between visual observations and clinical questions, and concludes by proposing new regions to explore. Across rounds, PathReasoning builds a reasoning chain that gradually directs attention to diagnostically relevant areas. PathReasoning turns each whole slide into a sequence of question-guided views, allowing the model to efficiently find informative ROIs within a fixed number of steps, without the need for dense pixel-level annotations. PathReasoning can substantially outperform strong ROI-selection approaches by 6.7% and 3.1% of AUROC on subtyping and longitudinal analysis tasks. The high-quality ROIs further support accurate report generation on breast cancer, significantly outperforming the standard GPT-4o by 10% in accuracy. PathReasoning prioritizes question-specific regions and constructs interpretable reasoning chains, supporting efficient slide review, consistent diagnostic interpretations, comprehensive reporting, and evidence traceability in digital pathology.

cs.CV↗

arXiv2025

Communication-Efficient Distributed Online Nonconvex Optimization with Time-Varying Constraints

This paper considers distributed online nonconvex optimization with time-varying inequality constraints over a network of agents, where the nonconvex local loss and convex local constraint functions can vary arbitrarily across iterations. For a time-varying directed graph, we propose two distributed bandit online primal--dual algorithm with compressed communication to efficiently utilize communication resources in the one-point and two-point bandit feedback settings, respectively. To measure the performance of the proposed algorithms, we use a network regret metric grounded in the first-order optimality condition associated with the variational inequality. We show that the compressed algorithms establish sublinear network regret and cumulative constraint violation bounds. Moreover, the network cumulative constraint violation bounds are reduced under Slater's condition. Finally, a simulation example is presented to validate the theoretical results.

math.OC↗

arXiv2025

Distributed Constrained Online Nonconvex Optimization with Compressed Communication

This paper considers distributed online nonconvex optimization with time-varying inequality constraints over a network of agents. For a time-varying graph, we propose a distributed online primal-dual algorithm with compressed communication to efficiently utilize communication resources. We show that the proposed algorithm establishes an $\mathcal{O}( {{T^{\max \{ {1 - {θ_1},{θ_1}} \}}}} )$ network regret bound and an $\mathcal{O}( {T^{1 - {θ_1}/2}} )$ network cumulative constraint violation bound, where $T$ is the number of iterations and ${θ_1} \in ( {0,1} )$ is a user-defined trade-off parameter. When Slater's condition holds (i.e, there is a point that strictly satisfies the inequality constraints at all iterations), the network cumulative constraint violation bound is reduced to $\mathcal{O}( {T^{1 - {θ_1}}} )$. These bounds are comparable to the state-of-the-art results established by existing distributed online algorithms with perfect communication for distributed online convex optimization with (time-varying) inequality constraints. Finally, a simulation example is presented to validate the theoretical results.

math.OC↗

arXiv2025

IPGO: Indirect Prompt Gradient Optimization for Parameter-Efficient Prompt-level Fine-Tuning on Text-to-Image Models

Text-to-Image Diffusion models excel at generating images from text prompts but often exhibit suboptimal alignment with content semantics, aesthetics, and human preferences. To address these limitations, this study proposes a novel parameter-efficient framework, Indirect Prompt Gradient Optimization (IPGO), for prompt-level diffusion model fine-tuning. IPGO enhances prompt embeddings by injecting continuously differentiable embeddings at the beginning and end of the prompt embeddings, leveraging low-rank structures with the flexibility and nonlinearity from rotations. This approach enables gradient-based optimization of injected embeddings under range, orthonormality, and conformity constraints, effectively narrowing the search space, promoting a stable solution, and ensuring alignment between the embeddings of the injected embeddings and the original prompt. Its extension IPGO+ adds a parameter-free cross-attention mechanism on the prompt embedding to enforce dependencies between the original prompt and the inserted embeddings. We conduct extensive evaluations through prompt-wise (IPGO) and prompt-batch (IPGO+) training using three reward models of image aesthetics, image-text alignment, and human preferences across three datasets of varying complexity. The results show that IPGO consistently outperforms SOTA benchmarks, including stable diffusion v1.5 with raw prompts, text-embedding-based methods (TextCraftor), training-based methods (DRaFT and DDPO), and training-free methods (DPO-Diffusion, Promptist, and ChatGPT-4o). Specifically, IPGO achieves a win-rate exceeding 99% in prompt-wise learning, and IPGO+ achieves a comparable, but often better performance against current SOTAs (a 75% win rate) in prompt-batch learning. Moreover, we illustrate IPGO's generalizability and its capability to significantly enhance image quality while requiring minimal data and resources.

cs.LG↗

arXiv2025

One-Point Sampling for Distributed Bandit Convex Optimization with Time-Varying Constraints

This paper considers the distributed bandit convex optimization problem with time-varying constraints. In this problem, the global loss function is the average of all the local convex loss functions, which are unknown beforehand. Each agent iteratively makes its own decision subject to time-varying inequality constraints which can be violated but are fulfilled in the long run. For a uniformly jointly strongly connected time-varying directed graph, a distributed bandit online primal-dual projection algorithm with one-point sampling is proposed. We show that sublinear dynamic network regret and network cumulative constraint violation are achieved if the path-length of the benchmark also increases in a sublinear manner. In addition, an $\mathcal{O}({T^{3/4 + g}})$ static network regret bound and an $\mathcal{O}( {{T^{1 - {g}/2}}} )$ network cumulative constraint violation bound are established, where $T$ is the total number of iterations and $g \in ( {0,1/4} )$ is a trade-off parameter. Moreover, a reduced static network regret bound $\mathcal{O}( {T^{2/3 + 4g /3}} )$ is established for strongly convex local loss functions. Finally, a numerical example is presented to validate the theoretical results.

eess.SY↗

arXiv2025

Reduced Network Cumulative Constraint Violation for Distributed Bandit Convex Optimization under Slater Condition

This paper studies the distributed bandit convex optimization problem with time-varying inequality constraints, where the goal is to minimize network regret and cumulative constraint violation. To calculate network cumulative constraint violation, existing distributed bandit online algorithms solving this problem directly use the clipped constraint function to replace its original constraint function. However, the use of the clipping operation renders Slater condition (i.e, there exists a point that strictly satisfies the inequality constraints at all iterations) ineffective to achieve reduced network cumulative constraint violation. To tackle this challenge, we propose a new distributed bandit online primal-dual algorithm. If local loss functions are convex, we show that the proposed algorithm establishes sublinear network regret and cumulative constraint violation bounds. When Slater condition holds, the network cumulative constraint violation bound is reduced. In addition, if local loss functions are strongly convex, for the case where strongly convex parameters are unknown, the network regret bound is reduced. For the case where strongly convex parameters are known, the network regret and cumulative constraint violation bounds are further reduced. To the best of our knowledge, this paper is among the first to establish reduced (network) cumulative constraint violation bounds for (distributed) bandit convex optimization with time-varying constraints under Slater condition. Finally, a numerical example is provided to verify the theoretical results.

eess.SY↗

arXiv2025

Low-Cost and Comprehensive Non-textual Input Fuzzing with LLM-Synthesized Input Generators

Modern software often accepts inputs with highly complex grammars. Recent advances in large language models (LLMs) have shown that they can be used to synthesize high-quality natural language text and code that conforms to the grammar of a given input format. Nevertheless, LLMs are often incapable or too costly to generate non-textual outputs, such as images, videos, and PDF files. This limitation hinders the application of LLMs in grammar-aware fuzzing. We present a novel approach to enabling grammar-aware fuzzing over non-textual inputs. We employ LLMs to synthesize and also mutate input generators, in the form of Python scripts, that generate data conforming to the grammar of a given input format. Then, non-textual data yielded by the input generators are further mutated by traditional fuzzers (AFL++) to explore the software input space effectively. Our approach, namely G2FUZZ, features a hybrid strategy that combines a holistic search driven by LLMs and a local search driven by industrial quality fuzzers. Two key advantages are: (1) LLMs are good at synthesizing and mutating input generators and enabling jumping out of local optima, thus achieving a synergistic effect when combined with mutation-based fuzzers; (2) LLMs are less frequently invoked unless really needed, thus significantly reducing the cost of LLM usage. We have evaluated G2FUZZ on a variety of input formats, including TIFF images, MP4 audios, and PDF files. The results show that G2FUZZ outperforms SOTA tools such as AFL++, Fuzztruction, and FormatFuzzer in terms of code coverage and bug finding across most programs tested on three platforms: UNIFUZZ, FuzzBench, and MAGMA.

cs.SE↗

arXiv2025

Your Fix Is My Exploit: Enabling Comprehensive DL Library API Fuzzing with Large Language Models

Deep learning (DL) libraries, widely used in AI applications, often contain vulnerabilities like buffer overflows and use-after-free errors. Traditional fuzzing struggles with the complexity and API diversity of DL libraries such as TensorFlow and PyTorch, which feature over 1,000 APIs. Testing all these APIs is challenging due to complex inputs and varied usage patterns. While large language models (LLMs) show promise in code understanding and generation, existing LLM-based fuzzers lack deep knowledge of API edge cases and struggle with test input generation. To address this, we propose DFUZZ, an LLM-driven fuzzing approach for DL libraries. DFUZZ leverages two insights: (1) LLMs can reason about error-triggering edge cases from API code and apply this knowledge to untested APIs, and (2) LLMs can accurately synthesize test programs to automate API testing. By providing LLMs with a "white-box view" of APIs, DFUZZ enhances reasoning and generation for comprehensive fuzzing. Experimental results show that DFUZZ outperforms state-of-the-art fuzzers in API coverage for TensorFlow and PyTorch, uncovering 37 bugs, with 8 fixed and 19 under developer investigation.

cs.SE↗

arXiv2024

Enhancing Investment Analysis: Optimizing AI-Agent Collaboration in Financial Research

In recent years, the application of generative artificial intelligence (GenAI) in financial analysis and investment decision-making has gained significant attention. However, most existing approaches rely on single-agent systems, which fail to fully utilize the collaborative potential of multiple AI agents. In this paper, we propose a novel multi-agent collaboration system designed to enhance decision-making in financial investment research. The system incorporates agent groups with both configurable group sizes and collaboration structures to leverage the strengths of each agent group type. By utilizing a sub-optimal combination strategy, the system dynamically adapts to varying market conditions and investment scenarios, optimizing performance across different tasks. We focus on three sub-tasks: fundamentals, market sentiment, and risk analysis, by analyzing the 2023 SEC 10-K forms of 30 companies listed on the Dow Jones Index. Our findings reveal significant performance variations based on the configurations of AI agents for different tasks. The results demonstrate that our multi-agent collaboration system outperforms traditional single-agent models, offering improved accuracy, efficiency, and adaptability in complex financial environments. This study highlights the potential of multi-agent systems in transforming financial analysis and investment decision-making by integrating diverse analytical perspectives.

cs.AI↗

arXiv2024

Sec2Sec Co-attention for Video-Based Apparent Affective Prediction

Video-based apparent affect detection plays a crucial role in video understanding, as it encompasses various elements such as vision, audio, audio-visual interactions, and spatiotemporal information, which are essential for accurate video predictions. However, existing approaches often focus on extracting only a subset of these elements, resulting in the limited predictive capacity of their models. To address this limitation, we propose a novel LSTM-based network augmented with a Transformer co-attention mechanism for predicting apparent affect in videos. We demonstrate that our proposed Sec2Sec Co-attention Transformer surpasses multiple state-of-the-art methods in predicting apparent affect on two widely used datasets: LIRIS-ACCEDE and First Impressions. Notably, our model offers interpretability, allowing us to examine the contributions of different time points to the overall prediction. The implementation is available at: https://github.com/nestor-sun/sec2sec.

cs.MM↗