arXiv subjects

Jiong Lou

Publications and source records attributed to Jiong Lou.

At least 19 recordsLinked to original sources

arXiv2026

Stop When Memory Suffices: Evidence-Conditioned Progressive Execution for LLM Agents

The continued development of LLMs toward persistent and adaptive intelligence increasingly requires long-term memory mechanisms that preserve and reuse information across interactions. Existing memory systems either compress and structure histories for efficient access or perform deep research over broader trajectories. The former lowers online cost but may omit temporal, causal, or cross-step dependencies, while the latter improves evidence coverage at substantial latency and inference cost. This raises a key question: can a memory system achieve strong answer quality while maintaining low online latency? We introduce Router-Mem, an evidence-conditioned progressive execution framework for long-horizon agent memory. Router-Mem first applies a shared low-cost retrieval prefix to obtain evidence. A lightweight sufficiency router then predicts whether the context supports early termination, which enable a single-token decision at inference time. It is trained with evidence-level supervision and rationale-conditioned representation distillation. When evidence is insufficient, Router-Mem reuses retrieval hits to expand memory blocks and perform deeper analysis and aggregation. Experiments on AMA-Bench and BEAM show that Router-Mem achieves 55.17\% and 38.77\% score while reducing average inference time by 27.3\% and 25.5\% compared with full memory execution.

cs.LG↗

arXiv2026

BRA-Audit: Budgeted Runtime Auditing for LLM Multi-Agent Systems via Cumulative-Exposure Audit-Point Placement

LLM-based multi-agent systems (LLM-MAS) solve complex tasks through specialized collaboration, but inter-agent dependencies can propagate hallucinated or malicious outputs into system-level failures. Auditor agents mitigate these risks, yet existing strategies face an efficiency dilemma: end-only auditing reviews long trajectories and final outputs, potentially weakening audit effectiveness and enlarging rollback scope, while auditing every agent each round improves detection and localization at high token cost. How can guard performance be preserved while minimizing token cost? To address this problem, we propose BRA-Audit, a budget-aware runtime auditing framework that models MAS execution as a dynamic dependency graph and formulates audit scheduling as audit-point placement under a fixed audit-call budget to minimize cumulative unchecked exposure. Its greedy scheduler prioritizes influential and long-unaudited regions, while trusted audit points enable localized recovery. Across structured coordination, complex reasoning, and open-ended tasks, BRA-Audit restores performance close to the clean setting, remains competitive with heavy guard methods and reduces end-to-end token consumption by $17.2\%$--$40.6\%$.

cs.MA↗

arXiv2026

Deep Interaction: An Efficient Human-AI Interaction Method for Large Reasoning Models

The emergence of Chain-of-Thought (CoT) reasoning has significantly enhanced the ability of large language models (LLMs) to tackle complex, multi-step tasks. However, when errors occur, current interaction approaches typically involve re-generating another response that may make mistakes again, or users laboriously flag the faulty step in follow-up turns that may get responses followed by similar errors recurring. To address this issue, we propose an efficient human intervention mechanism for precisely correcting reasoning errors in LLMs, termed Deep Interaction. Our approach enables direct editing of the original response, allowing erroneous parts to be corrected while preserving accurate reasoning steps. We refine the edited CoT into a distilled prompt, which then steers the LLM along the corrected reasoning path. Experimental results show that our method achieves over a 25% improvement in correction success rate and reduces token usage by approximately 40% on STEM tasks reasoning compared to baseline approaches.

cs.AI↗

arXiv2026

E-mem: Multi-agent based Episodic Context Reconstruction for LLM Agent Memory

The evolution of Large Language Model (LLM) agents towards System~2 reasoning, characterized by deliberative, high-precision problem-solving, requires maintaining rigorous logical integrity over extended horizons. However, prevalent memory preprocessing paradigms suffer from destructive de-contextualization. By compressing complex sequential dependencies into pre-defined structures (e.g., embeddings or graphs), these methods sever the contextual integrity essential for deep reasoning. To address this, we propose E-mem, a framework shifting from Memory Preprocessing to Episodic Context Reconstruction. Inspired by biological engrams, E-mem employs a heterogeneous hierarchical architecture where multiple assistant agents maintain uncompressed memory contexts, while a central master agent orchestrates global planning. Unlike passive retrieval, our mechanism empowers assistants to locally reason within activated segments, extracting context-aware evidence before aggregation. Evaluations on the LoCoMo benchmark demonstrate that E-mem achieves over 54\% F1, surpassing the state-of-the-art GAM by 7.75\%, while reducing token cost by over 70\%.

cs.AI↗

arXiv2026

Privacy-Preserving Text Sanitization for Distributed Agents Collaboration via Disentangled Representations

When distributed agents exchange text across organizational boundaries, privacy leakage arises not only from explicit identifiers but also from distributional signatures such as formatting conventions, vocabulary choices, and syntactic patterns. We propose DiSan(Disentangled Sanitization), a privacy-preserving sanitization framework and a built-in component of Intern-Shannon for multi-agent collaboration. DiSan uses a two-stream encoder to factorize text into a source-invariant role subspace that preserves task semantics and a source-identifying style subspace that remains local. Federated proto-type alignment and adversarial regularization enable joint training without centralizing raw text. Experiments show that identifier-level masking is insufficient: masking 19.2% of tokens reduces TF-IDF stylometric attribution by only 18.6%. By contrast, DiSan reduces answer-level PII exposure by 20 times while maintaining 83% answer faithfulness on a distributed multi-agent RAG benchmark, and lowers Enron stylometric attribution by 73.2% under TF-IDF and 70.6% under a neural probe.

cs.CL↗

arXiv2026

HeraSys: Collaborative Serving of Multiple LLM Workflows via Fine-Grained End-to-End Optimization

The proliferation of Large Language Models (LLMs) has shifted serving systems from processing isolated requests to orchestrating high-concurrency, multi-tenant agentic workflows. However, existing solutions typically prioritize intra-workflow optimization, largely neglecting the significant potential for inter-workflow optimization. In this paper, we propose HeraSys, an LLM serving system designed to optimize the end-to-end performance of concurrent workflows. Through fine-grained orchestration, HeraSys eliminates cross-workflow computational redundancy via structural node merging and reuse. Furthermore, HeraSys introduces a load-aware joint scheduling policy that dynamically manages execution order by evaluating both inter- and intra-query priorities. By integrating a resource skewing mechanism with adaptive batching and pipeline decomposition, HeraSys effectively mitigates tail latency while maintaining low average latency, thereby substantially improving system throughput. Extensive experiments demonstrate that HeraSys reduces P99 latency by up to 2.17$\times$ and increases serving throughput by up to 1.85$\times$ under strict latency guarantees.

cs.AI↗

arXiv2026

IntentKV: Cross-Turn Intent-Aware KV Cache Pruning for Agent Inference

Multi-turn LLM agents fan short queries into long trajectories of tool calls, search results, and intermediate reasoning. Both KV memory and KV read bandwidth grow by orders of magnitude across a single trajectory, making the key-value (KV) cache, not parameter compute, the dominant serving bottleneck for long-horizon agents. We introduce IntentKV, learned KV pruning that keeps the base LLM frozen. IntentKV maintains a session-level QueryMemory of cross-turn intent, scores live history tokens with a memory-attention rule, and adds a zero-initialized residual head with cross-attention over current-query K-vectors. To stay composable with prefix caches, eviction is a slot-map redirection: dropped positions route to a sentinel dead slot while surviving K/V rows, RoPE phases, and slot identities stay in place. IntentKV matches the no-pruning full-cache baseline with almost no accuracy drop under tight KV budgets: at an 8k KV budget, mean peak request tokens drop 23.9% on Qwen3-8B and 30.7% on Qwen2.5-14B. On the 100 longest BCP queries that all methods complete on Qwen2.5-14B, IntentKV-8k further cuts worst-case peak request tokens from 92.3k to 20.5k, a 77.8% reduction, and worst-case raw KV reads from 411M to 31M, a 92.6% reduction.

cs.LG↗

arXiv2026

OEP: Poisoning Self-Evolving LLM Agents via Locally Correct but Non-Transferable Experiences

Memory-augmented large language model (LLM) agents use iterative reflection and self-evolution to solve complex tasks, but these mechanisms introduce security risks. Existing agentic memory attacks require privileged access or explicit malicious content, making them detectable by advanced safety filters. This leaves a subtler attack surface underexplored: whether adversaries can induce agent to generate experiences that appear locally correct and semantically plausible yet induce harmful generalization during reflection. We find that reflective agents are vulnerable to such clean experiences, especially when paired with severe but plausible hypothetical consequences. Based on this observation, we introduce Obsessive Experience Poisoning (OEP), a low-privilege black-box attack requiring no direct control over the system prompt or memory database. OEP constructs adversarial clean edge-cases that combine locally correct solutions, non-transferable methods, and severe consequences, biasing reflection toward risk-averse rule formation. During memory consolidation, agents may over-trust self-generated reflections and distill localized experiences into high-priority but over-generalized rules, causing downstream failures. Evaluations across three domains show that OEP achieves ASR above 50\% with GPT-4o agents, and outperforms existing attacks under LLM auditing defense.

cs.CR↗

arXiv2026

ProxyKV: Cross-Model Proxy Pruning for Efficient Long-Context LLM Inference

Efficient long-context inference in Large Language Models (LLMs) is severely constrained by the Key-Value (KV) cache memory wall, yet existing pruning methods force a choice between low-latency heuristics that sacrifice precision and high-precision reconstruction methods that incur prohibitive prefilling overhead. To bridge this scoring-cost--accuracy gap, we propose ProxyKV, a cross-model proxy pruning framework that offloads importance scoring to a lightweight intra-family Small-Model Proxy executed asynchronously to the Large-Model Target. To bridge the architectural gap between heterogeneous models, we design the HybridAxialMapper, which disentangles temporal feature extraction from cross-head alignment, together with a Multi-Granularity Hybrid Loss that shifts the learning objective from rigid regression to relative ranking consistency. Across the Llama-3.1, Qwen-2.5, and Qwen-3 families spanning targets from 7B up to 32B parameters on LongBench, SCBench, and RULER, ProxyKV matches KVZip on aggregate (recovering $\sim$$98.7\%$ of its mean accuracy) while delivering up to a $3.21\times$ prefilling speedup on Llama-3.1-8B (dual-GPU; $\sim$$1.5\times$ shared single-GPU) and sustaining the speedup at contexts up to 170k tokens on Qwen-2.5-7B.

cs.LG↗

arXiv2026

Federated Cross-Modal Retrieval with Missing Modalities via Semantic Routing and Adapter Personalization

Federated cross-modal retrieval faces severe challenges from heterogeneous client data, particularly non-IID semantic distributions and missing modalities. Under such heterogeneity, a single global model is often insufficient to capture both shared cross-modal knowledge and client-specific characteristics. We propose RCSR, a personalization-friendly federated framework that integrates prototype anchoring, retrieval-centric semantic routing, and optional client-specific adapters. Built on a frozen CLIP backbone, RCSR leverages lightweight shared adapters for global knowledge transfer while supporting efficient local personalization. Prototype anchoring helps unimodal clients align with global cross-modal semantics, and a server-side semantic router adaptively assigns aggregation weights based on retrieval consistency to mitigate alignment drift during heterogeneous updates. Extensive experiments on MS-COCO, Flickr30K, and other benchmarks show that RCSR consistently improves global retrieval accuracy and training stability, while further enhancing client-level retrieval performance, especially for clients with incomplete modalities. Code is available at https://github.com/RezinChow/RCSR-Retrieval-Centric-Semantic-Routing.

cs.CV↗

arXiv2026

IEMAS: An Incentive-Efficiency Routing Framework for Open Agentic Web Ecosystems

The transition to open, distributed Multi-Agent Systems (MAS) promises scalable intelligence but introduces a non-trivial tension: maximizing global efficiency requires cooperative, resource-aware scheduling, yet autonomous agents may be self-interested and cannot be managed by a centralized controller. Prior approaches fall short in two key areas: they typically focus on single-query routing, neglecting long-term resource reuse (e.g., KV-caching) and the complexities of system-level many-to-many matching; furthermore, they rely on generic incentive mechanisms that ignore the distinct characteristics of LLM inference. To bridge this gap, we propose IEMAS (Incentive-Efficiency Mechanism for Multi-Agent Systems), a distributed framework that aligns economic incentives with system performance. IEMAS integrates a probabilistic predictive model to estimate Quality of Service (QoS) under uncertainty, which feeds into a VCG-based bipartite matching mechanism. This design guarantees truthful capability reporting and social optimality while explicitly leveraging KV cache-affinity to minimize computational redundancy. We implement IEMAS on top of vLLM and evaluate it via extensive simulations. Results demonstrate that our incentive-efficiency co-design reducing average service cost by 35% and end-to-end latency by up to 2.9 compared to baselines.

cs.NI↗

arXiv2025

AgentTutor: Empowering Personalized Learning with Multi-Turn Interactive Teaching in Intelligent Education Systems

The rapid advancement of large-scale language models (LLMs) has shown their potential to transform intelligent education systems (IESs) through automated teaching and learning support applications. However, current IESs often rely on single-turn static question-answering, which fails to assess learners' cognitive levels, cannot adjust teaching strategies based on real-time feedback, and is limited to providing simple one-off responses. To address these issues, we introduce AgentTutor, a multi-turn interactive intelligent education system to empower personalized learning. It features an LLM-powered generative multi-agent system and a learner-specific personalized learning profile environment that dynamically optimizes and delivers teaching strategies based on learners' learning status, personalized goals, learning preferences, and multimodal study materials. It includes five key modules: curriculum decomposition, learner assessment, dynamic strategy, teaching reflection, and knowledge & experience memory. We conducted extensive experiments on multiple benchmark datasets, AgentTutor significantly enhances learners' performance while demonstrating strong effectiveness in multi-turn interactions and competitiveness in teaching quality among other baselines.

cs.CY↗

arXiv2025

MAS-Shield: A Defense Framework for Secure and Efficient LLM MAS

Large Language Model (LLM)-based Multi-Agent Systems (MAS) are susceptible to linguistic attacks that can trigger cascading failures across the network. Existing defenses face a fundamental dilemma: lightweight single-auditor methods are prone to single points of failure, while robust committee-based approaches incur prohibitive computational costs in multi-turn interactions. To address this challenge, we propose \textbf{MAS-Shield}, a secure and efficient defense framework designed with a coarse-to-fine filtering pipeline. Rather than applying uniform scrutiny, MAS-Shield dynamically allocates defense resources through a three-stage protocol: (1) \textbf{Critical Agent Selection } strategically targets high-influence nodes to narrow the defense surface; (2) \textbf{Light Auditing} employs lightweight sentry models to rapidly filter the majority of benign cases; and (3) \textbf{Global Consensus Auditing} escalates only suspicious or ambiguous signals to a heavyweight committee for definitive arbitration. This hierarchical design effectively optimizes the security-efficiency trade-off. Experiments demonstrate that MAS-Shield achieves a 92.5\% recovery rate against diverse adversarial scenarios and reduces defense latency by over 70\% compared to existing methods.

cs.MA↗

arXiv2025

BAPFL: Exploring Backdoor Attacks Against Prototype-based Federated Learning

Prototype-based federated learning (PFL) has emerged as a promising paradigm to address data heterogeneity problems in federated learning, as it leverages mean feature vectors as prototypes to enhance model generalization. However, its robustness against backdoor attacks remains largely unexplored. In this paper, we identify that PFL is inherently resistant to existing backdoor attacks due to its unique prototype learning mechanism and local data heterogeneity. To further explore the security of PFL, we propose BAPFL, the first backdoor attack method specifically designed for PFL frameworks. BAPFL integrates a prototype poisoning strategy with a trigger optimization mechanism. The prototype poisoning strategy manipulates the trajectories of global prototypes to mislead the prototype training of benign clients, pushing their local prototypes of clean samples away from the prototypes of trigger-embedded samples. Meanwhile, the trigger optimization mechanism learns a unique and stealthy trigger for each potential target label, and guides the prototypes of trigger-embedded samples to align closely with the global prototype of the target label. Experimental results across multiple datasets and PFL variants demonstrate that BAPFL achieves a 35\%-75\% improvement in attack success rate compared to traditional backdoor attacks, while preserving main task accuracy. These results highlight the effectiveness, stealthiness, and adaptability of BAPFL in PFL.

cs.LG↗

arXiv2025

Adaptive AI Agent Placement and Migration in Edge Intelligence Systems

The rise of LLMs such as ChatGPT and Claude fuels the need for AI agents capable of real-time task handling. However, migrating data-intensive, multi-modal edge workloads to cloud data centers, traditionally used for agent deployment, introduces significant latency. Deploying AI agents at the edge improves efficiency and reduces latency. However, edge environments present challenges due to limited and heterogeneous resources. Maintaining QoS for mobile users necessitates agent migration, which is complicated by the complexity of AI agents coordinating LLMs, task planning, memory, and external tools. This paper presents the first systematic deployment and management solution for LLM-based AI agents in dynamic edge environments. We propose a novel adaptive framework for AI agent placement and migration in edge intelligence systems. Our approach models resource constraints and latency/cost, leveraging ant colony algorithms and LLM-based optimization for efficient decision-making. It autonomously places agents to optimize resource utilization and QoS and enables lightweight agent migration by transferring only essential state. Implemented on a distributed system using AgentScope and validated across globally distributed edge servers, our solution significantly reduces deployment latency and migration costs.

cs.AI↗

arXiv2025

LOVO: Efficient Complex Object Query in Large-Scale Video Datasets

The widespread deployment of cameras has led to an exponential increase in video data, creating vast opportunities for applications such as traffic management and crime surveillance. However, querying specific objects from large-scale video datasets presents challenges, including (1) processing massive and continuously growing data volumes, (2) supporting complex query requirements, and (3) ensuring low-latency execution. Existing video analysis methods struggle with either limited adaptability to unseen object classes or suffer from high query latency. In this paper, we present LOVO, a novel system designed to efficiently handle comp$\underline{L}$ex $\underline{O}$bject queries in large-scale $\underline{V}$ide$\underline{O}$ datasets. Agnostic to user queries, LOVO performs one-time feature extraction using pre-trained visual encoders, generating compact visual embeddings for key frames to build an efficient index. These visual embeddings, along with associated bounding boxes, are organized in an inverted multi-index structure within a vector database, which supports queries for any objects. During the query phase, LOVO transforms object queries to query embeddings and conducts fast approximate nearest-neighbor searches on the visual embeddings. Finally, a cross-modal rerank is performed to refine the results by fusing visual features with detailed textual features. Evaluation on real-world video datasets demonstrates that LOVO outperforms existing methods in handling complex queries, with near-optimal query accuracy and up to 85x lower search latency, while significantly reducing index construction costs. This system redefines the state-of-the-art object query approaches in video analysis, setting a new benchmark for complex object queries with a novel, scalable, and efficient approach that excels in dynamic environments.

cs.IR↗

arXiv2025

EAT: QoS-Aware Edge-Collaborative AIGC Task Scheduling via Attention-Guided Diffusion Reinforcement Learning

The growth of Artificial Intelligence (AI) and large language models has enabled the use of Generative AI (GenAI) in cloud data centers for diverse AI-Generated Content (AIGC) tasks. Models like Stable Diffusion introduce unavoidable delays and substantial resource overhead, which are unsuitable for users at the network edge with high QoS demands. Deploying AIGC services on edge servers reduces transmission times but often leads to underutilized resources and fails to optimally balance inference latency and quality. To address these issues, this paper introduces a QoS-aware \underline{E}dge-collaborative \underline{A}IGC \underline{T}ask scheduling (EAT) algorithm. Specifically: 1) We segment AIGC tasks and schedule patches to various edge servers, formulating it as a gang scheduling problem that balances inference latency and quality while considering server heterogeneity, such as differing model distributions and cold start issues. 2) We propose a reinforcement learning-based EAT algorithm that uses an attention layer to extract load and task queue information from edge servers and employs a diffusion-based policy network for scheduling, efficiently enabling model reuse. 3) We develop an AIGC task scheduling system that uses our EAT algorithm to divide tasks and distribute them across multiple edge servers for processing. Experimental results based on our system and large-scale simulations show that our EAT algorithm can reduce inference latency by up to 56\% compared to baselines. We release our open-source code at https://github.com/zzf1955/EAT.

cs.DC↗

arXiv2025

BLOCKS: Blockchain-supported Cross-Silo Knowledge Sharing for Efficient LLM Services

The hallucination problem of Large Language Models (LLMs) has increasingly drawn attention. Augmenting LLMs with external knowledge is a promising solution to address this issue. However, due to privacy and security concerns, a vast amount of downstream task-related knowledge remains dispersed and isolated across various "silos," making it difficult to access. To bridge this knowledge gap, we propose a blockchain-based external knowledge framework that coordinates multiple knowledge silos to provide reliable foundational knowledge for large model retrieval while ensuring data security. Technically, we distill knowledge from local data into prompts and execute transactions and records on the blockchain. Additionally, we introduce a reputation mechanism and cross-validation to ensure knowledge quality and provide incentives for participation. Furthermore, we design a query generation framework that provides a direct API interface for large model retrieval. To evaluate the performance of our proposed framework, we conducted extensive experiments on various knowledge sources. The results demonstrate that the proposed framework achieves efficient LLM service knowledge sharing in blockchain environments.

cs.DC↗