Xin Zhang

ClawGuard is a novel runtime security framework that deterministically enforces user-confirmed rules at tool-call boundaries to protect LLM agents from indirect prompt injection.

ZK-Value introduces a practical, scalable zero-knowledge system for calculating data valuations (Shapley values) in data marketplaces, significantly reducing proving time while maintaining high accuracy.

SafeHarbor is a novel, hierarchical memory-augmented framework that establishes context-aware decision boundaries for LLM agents, achieving state-of-the-art safety while minimizing over-refusal.

This paper proposes four guidelines and two novel data ordering methods (STR and SAW) to systematically optimize data organization, significantly enhancing the stability and performance of LLM training.

The paper introduces RoboWits, a new bi-manual robotic benchmark designed to test a robot's cognitive reasoning and adaptability to unexpected challenges, revealing that current Vision-Language-Action (VLA) models are brittle when faced with mutated or constrained tasks.

The paper proposes a unified framework that decouples long-video reasoning into semantic and visual evidence, significantly improving performance on the HD-EPIC VQA Challenge.

This paper introduces Electric Flow Sampling (elfs) as a zero-error quantum walk primitive and uses it to derive improved quantum algorithms for various graph problems, including semi-supervised learning.

The paper introduces ERGeoBench, a comprehensive diagnostic benchmark designed to evaluate the fine-grained capabilities of multimodal large language models (MLLMs) for embodied geo-localization across various viewing conditions.

The paper audits six LLMs across four languages, finding that their gender stereotyping is significantly wider than human baselines and that cross-lingual translation fundamentally alters the nature of the bias.

The paper proposes FedMChain, a novel federated learning framework that structures multimodal training into sequential phases to mitigate modality competition and improve model performance while reducing communication overhead.

The paper introduces RUBAS, a rubric-based reinforcement learning framework that improves agent safety by providing fine-grained, multi-dimensional rewards for complex tool-use scenarios.

The paper proposes a framework to harvest unused computation resources on AI chips for general-purpose tasks using neural architecture search and approximation techniques.

This paper proposes TRIAGE, a role-typed credit assignment framework for agentic reinforcement learning to address the structural incompleteness of standard GRPO.

The paper introduces MADB, a large-scale dataset and benchmark for music aesthetic assessment with 9,999 tracks annotated by 30 trained annotators across 10 perceptual dimensions.

This paper proves that every simple planar graph with no copy of C8 has at most 69/25(n-2) edges, improving the previous bound.

This paper proposes PeakFlow, a framework for dynamic affective trajectory prediction in EEG data using a peak-guided coarse-to-refined approach.

This paper proposes CoHarden, a co-generation framework for automated program repair that uses a lax signal as an in-loop convergence criterion to prevent lax regressions.

This paper introduces GS-Agent, an end-to-end multi-agent framework that generates realistic, dynamic, and controllable 4D physical worlds from natural language descriptions by emulating human creation process using physics engines.