arXiv subjects

Jiaqi Shao

Publications and source records attributed to Jiaqi Shao.

14 recordsLinked to original sources

arXiv2026

Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI

Agent benchmarks increasingly evaluate repository editing, web research, terminal use, and long-horizon interaction. Their scores support capability claims only when the evaluation protocol keeps the intended capability necessary for success. Recent reward-hacking benchmarks and system reports show that agents can instead recover public solutions, read evaluation artifacts, infer generator structure, manipulate feedback, or benefit from invalid scoring paths; existing responses do not provide a common procedure for attributing these shortcuts and quantifying their effect across benchmarks. We formulate protocol validity and introduce HackDetect, a post-hoc audit that identifies an exposure, determines how the agent used it, and assesses whether the resulting score is misleading. We quantify score inflation with the Mislead gap, defined as the exploit score minus the intended score. We audit 2,385 traces across 15 agent benchmarks and find evidence of exposures and reward hacking in 67.0% of Frontier Science traces and 66.7% of AutoLab tasks. Across paired comparisons, we measure score inflation of 0.45-1.00, showing that benchmark reports should provide evidence that scores reflect the intended capability.

cs.AI↗

arXiv2026

Do More Agents Help? Controlled and Protocol-Aligned Evaluation of LLM Agent Workflows

Does adding more agents help an LLM workflow once compared systems share the same benchmark loader, tool access, answer contract, usage accounting, and trajectory logging? We introduce BenchAgent, an evaluation framework that places single-agent, fixed multi-agent (MAS), and evolving MAS workflows under one normalized execution and logging protocol. BenchAgent evaluates these substrate-internal workflows across ten reasoning, coding, and tool-use benchmarks with GPT-4.1, and separately reports a Protocol-Aligned External (PAE) GAIA study of a runtime-generated workflow. Under SI conditions, at most one of six tested MAS exceeds the matched single-agent anchor on benchmark-balanced average accuracy: EvoAgent lies within the Wilson one-run guidance, while the remaining five trail by 2.56-11.29 points and occupy more expensive accuracy-cost trade-offs. On the PAE GAIA snapshot, a Claude-Code-style runtime workflow reaches 66.72% overall and 69.23% on Level 3, more than 20 points above the strongest non-Claude baseline, Jarvis, a fixed MAS.

cs.AI↗

arXiv2026

When Stored Evidence Stops Being Usable: Scale-Conditioned Evaluation of Agent Memory

Memory-agent evaluations report fixed-snapshot accuracy or retrieval quality, but these scores do not show whether evidence remains usable as irrelevant sessions (sessions not annotated as task-relevant evidence for the query) accumulate. We present a scale-conditioned evaluation protocol for agent memory under evidence-preserving growth: for each query, task evidence is held fixed while irrelevant sessions are added. The protocol logs agent--memory trajectories and reports four diagnostics: budget-compliant reliability, tail memory-call burden, failure-regime decomposition, and the usable-scale boundary where reliability falls below the target. Applied to LongMemEval and LoCoMo across flat, planar, and hierarchical memory interfaces, the protocol shows reliability loss is not a single phenomenon. On LongMemEval, HippoRAG stays within the two-call budget but loses 16--20 percentage points in budget-compliant reliability as irrelevant sessions are added; LiCoMemory's observed failures depend strongly on the agent, with Qwen3-8B exceeding the budget while Qwen3-32B and Qwen3-235B remain reliable in the tested range. The result supports a framework for making scalable-memory claims conditional on agent, interface, scale range, and interaction budget.

cs.AI↗

arXiv2026

Breaking Lock-In: Preserving Steerability under Low-Data VLA Post-Training

Have you ever post-trained a generalist vision-language-action (VLA) policy on a small demonstration dataset, only to find that it stops responding to new instructions and is limited to behaviors observed during post-training? We identify this phenomenon as lock-in: after low-data, supervised fine-tuning (SFT), the policy becomes overly specialized to the post-training data and fails to generalize to novel instructions, manifesting as concept lock-in (fixation on training objects/attributes) and spatial lock-in (fixation on training spatial targets). Many existing remedies introduce additional supervision signals, such as those derived from foundation models or auxiliary objectives, or rely on augmented datasets to recover generalization. In this paper, we show that the policy's internal pre-trained knowledge is sufficient: DeLock mitigates lock-in by preserving visual grounding during post-training and applying test-time contrastive prompt guidance to steer the policy's denoising dynamics according to novel instructions. Across eight simulation and real-world evaluations, DeLock consistently outperforms strong baselines and matches or exceeds the performance of a state-of-the-art generalist policy post-trained with substantially more curated demonstrations.

cs.RO↗

arXiv2026

From $O(mn)$ to $O(r^2)$: Two-Sided Low-Rank Communication for Adam in Distributed Training with Memory Efficiency

As foundation models continue to scale, pretraining increasingly relies on data-parallel distributed optimization, making bandwidth-limited gradient synchronization a key bottleneck. Orthogonally, projection-based low-rank optimizers were mainly designed for memory efficiency, but remain suboptimal for communication-limited training: one-sided synchronization still transmits an $O(rn)$ object for an $m\times n$ matrix gradient and refresh steps can dominate peak communicated bytes. We propose TSR, which brings two-sided low-rank communication to Adam-family updates (TSR-Adam) by synchronizing a compact core $U^\top G V\in\mathbb{R}^{r\times r}$, reducing the dominant per-step payload from $O(mn)$ to $O(r^2)$ while keeping moment states in low-dimensional cores. To further reduce the peak communication from subspace refresh, TSR-Adam adopts a randomized SVD-based refresh that avoids full-gradient synchronization. We additionally extend low-rank communication to embedding gradients with embedding-specific ranks and refresh schedules, yielding additional communication and memory savings over keeping embeddings dense. Across pretraining from 60M to 1B model scales, TSR-Adam reduces average communicated bytes per step by $13\times$, and on GLUE fine-tuning it reduces communication by $25\times$, while achieving comparable performance; we further provide a theoretical stationarity analysis for the proposed update. Code is available at https://github.com/DKmiyan/TSR-Adam.

cs.LG↗

arXiv2025

FoldAct: Efficient and Stable Context Folding for Long-Horizon Search Agents

Long-horizon reinforcement learning (RL) for large language models faces critical scalability challenges from unbounded context growth, leading to context folding methods that compress interaction history during task execution. However, existing approaches treat summary actions as standard actions, overlooking that summaries fundamentally modify the agent's future observation space, creating a policy-dependent, non-stationary observation distribution that violates core RL assumptions. This introduces three fundamental challenges: (1) gradient dilution where summary tokens receive insufficient training signal, (2) self-conditioning where policy updates change summary distributions, creating a vicious cycle of training collapse, and (3) computational cost from processing unique contexts at each turn. We introduce \textbf{FoldAct}\footnote{https://github.com/SHAO-Jiaqi757/FoldAct}, a framework that explicitly addresses these challenges through three key innovations: separated loss computation for independent gradient signals on summary and action tokens, full context consistency loss to reduce distribution shift, and selective segment training to reduce computational cost. Our method enables stable training of long-horizon search agents with context folding, addressing the non-stationary observation problem while improving training efficiency with 5.19$\times$ speedup.

cs.LG↗

arXiv2025

Do LLM Agents Know How to Ground, Recover, and Assess? A Benchmark for Epistemic Competence in Information-Seeking Agents

Recent work has explored training Large Language Model (LLM) search agents with reinforcement learning (RL) for open-domain question answering (QA). However, most evaluations focus solely on final answer accuracy, overlooking how these agents reason with and act on external evidence. We introduce SeekBench, the first benchmark for evaluating the \textit{epistemic competence} of LLM search agents through step-level analysis of their response traces. SeekBench comprises 190 expert-annotated traces with over 1,800 response steps generated by LLM search agents, each enriched with evidence annotations for granular analysis of whether agents (1) generate reasoning steps grounded in observed evidence, (2) adaptively reformulate searches to recover from low-quality results, and (3) have proper calibration to correctly assess whether the current evidence is sufficient for providing an answer.

cs.AI↗

arXiv2025

MorphAgent: Empowering Agents through Self-Evolving Profiles and Decentralized Collaboration

Large Language Model (LLM) based multi-agent systems (MAS) have shown promise in tackling complex tasks, but often rely on predefined roles and centralized coordination, limiting their adaptability to evolving challenges. This paper introduces MorphAgent, a novel Autonomous, Self-Organizing, and Self-Adaptive Multi-Agent System for decentralized agent collaboration that enables agents to dynamically evolve their roles and capabilities. Our approach employs self-evolving agent profiles, optimized through three key metrics, guiding agents in refining their individual expertise while maintaining complementary team dynamics. MorphAgent implements a two-phase process: a Profile Update phase for profile optimization, followed by a Task Execution phase where agents continuously adapt their roles based on task feedback. Our experimental results show that MorphAgent outperforms existing frameworks in terms of task performance and adaptability to changing requirements, paving the way for more robust and versatile multi-agent collaborative systems.

cs.AI↗

arXiv2025

Beyond Right to be Forgotten: Managing Heterogeneity Side Effects Through Strategic Incentives

Federated Unlearning (FU) enables the removal of specific clients' data influence from trained models. However, in non-IID settings, removing clients creates critical side effects: remaining clients with similar data distributions suffer disproportionate performance degradation, while the global model's stability deteriorates. These vulnerable clients then have reduced incentives to stay in the federation, potentially triggering a cascade of withdrawals that further destabilize the system. To address this challenge, we develop a theoretical framework that quantifies how data heterogeneity impacts unlearning outcomes. Based on these insights, we model FU as a Stackelberg game where the server strategically offers payments to retain crucial clients based on their contribution to both unlearning effectiveness and system stability. Our rigorous equilibrium analysis reveals how data heterogeneity fundamentally shapes the trade-offs between system-wide objectives and client interests. Our approach improves global stability by up to 6.23\%, reduces worst-case client degradation by 10.05\%, and achieves up to 38.6\% runtime efficiency over complete retraining.

cs.GT↗

arXiv2025

Cognitive Insights and Stable Coalition Matching for Fostering Multi-Agent Cooperation

Cognitive abilities, such as Theory of Mind (ToM), play a vital role in facilitating cooperation in human social interactions. However, our study reveals that agents with higher ToM abilities may not necessarily exhibit better cooperative behavior compared to those with lower ToM abilities. To address this challenge, we propose a novel matching coalition mechanism that leverages the strengths of agents with different ToM levels by explicitly considering belief alignment and specialized abilities when forming coalitions. Our proposed matching algorithm seeks to find stable coalitions that maximize the potential for cooperative behavior and ensure long-term viability. By incorporating cognitive insights into the design of multi-agent systems, our work demonstrates the potential of leveraging ToM to create more sophisticated and human-like coordination strategies that foster cooperation and improve overall system performance.

cs.MA↗

arXiv2025

Ten Challenging Problems in Federated Foundation Models

Federated Foundation Models (FedFMs) represent a distributed learning paradigm that fuses general competences of foundation models as well as privacy-preserving capabilities of federated learning. This combination allows the large foundation models and the small local domain models at the remote clients to learn from each other in a teacher-student learning setting. This paper provides a comprehensive summary of the ten challenging problems inherent in FedFMs, encompassing foundational theory, utilization of private data, continual learning, unlearning, Non-IID and graph data, bidirectional knowledge transfer, incentive mechanism design, game mechanism design, model watermarking, and efficiency. The ten challenging problems manifest in five pivotal aspects: ``Foundational Theory," which aims to establish a coherent and unifying theoretical framework for FedFMs. ``Data," addressing the difficulties in leveraging domain-specific knowledge from private data while maintaining privacy; ``Heterogeneity," examining variations in data, model, and computational resources across clients; ``Security and Privacy," focusing on defenses against malicious attacks and model theft; and ``Efficiency," highlighting the need for improvements in training, communication, and parameter efficiency. For each problem, we offer a clear mathematical definition on the objective function, analyze existing methods, and discuss the key challenges and potential solutions. This in-depth exploration aims to advance the theoretical foundations of FedFMs, guide practical implementations, and inspire future research to overcome these obstacles, thereby enabling the robust, efficient, and privacy-preserving FedFMs in various real-world applications.

cs.LG↗

arXiv2024

Demo: FedCampus: A Real-world Privacy-preserving Mobile Application for Smart Campus via Federated Learning & Analytics

In this demo, we introduce FedCampus, a privacy-preserving mobile application for smart \underline{campus} with \underline{fed}erated learning (FL) and federated analytics (FA). FedCampus enables cross-platform on-device FL/FA for both iOS and Android, supporting continuously models and algorithms deployment (MLOps). Our app integrates privacy-preserving processed data via differential privacy (DP) from smartwatches, where the processed parameters are used for FL/FA through the FedCampus backend platform. We distributed 100 smartwatches to volunteers at Duke Kunshan University and have successfully completed a series of smart campus tasks featuring capabilities such as sleep tracking, physical activity monitoring, personalized recommendations, and heavy hitters. Our project is opensourced at https://github.com/FedCampus/FedCampus_Flutter. See the FedCampus video at https://youtu.be/k5iu46IjA38.

cs.CR↗

arXiv2024

Federated Unlearning: a Perspective of Stability and Fairness

This paper explores the multifaceted consequences of federated unlearning (FU) with data heterogeneity. We introduce key metrics for FU assessment, concentrating on verification, global stability, and local fairness, and investigate the inherent trade-offs. Furthermore, we formulate the unlearning process with data heterogeneity through an optimization framework. Our key contribution lies in a comprehensive theoretical analysis of the trade-offs in FU and provides insights into data heterogeneity's impacts on FU. Leveraging these insights, we propose FU mechanisms to manage the trade-offs, guiding further development for FU mechanisms. We empirically validate that our FU mechanisms effectively balance trade-offs, confirming insights derived from our theoretical analysis.

cs.AI↗

arXiv2024

Privacy-Preserving Federated Heavy Hitter Analytics for Non-IID Data

Federated heavy-hitter analytics involves the identification of the most frequent items within distributed data. Existing methods for this task often encounter challenges such as compromising privacy or sacrificing utility. To address these issues, we introduce a novel privacy-preserving algorithm that exploits the hierarchical structure to discover local and global heavy hitters in non-IID data by utilizing perturbation and similarity techniques. We conduct extensive evaluations on both synthetic and real datasets to validate the effectiveness of our approach. We also present FedCampus, a demonstration application to showcase the capabilities of our algorithm in analyzing population statistics.

cs.DC↗