Hao Zhang

The Parameterized Diffusion Policy (PDP) framework transforms diffusion models from general stochastic generators into precise, steerable tools for learning and adapting complex robotic behaviors by embedding them on a semantically structured latent manifold.

The paper introduces BioConCal, a supervised scoring mechanism that evaluates biomedical NER candidates surfaced by multiple LLMs, significantly improving the quality of the candidate pool for human curators.

GaMi is a multimodal material identification system that uses mmWave and acoustic sensing with a cross-modal subtractive disentanglement framework to achieve high accuracy (95.2%) for material identification regardless of geometric variations.

The paper introduces Adaptive Context Management (AdaCoM), an external context manager that uses reinforcement learning to improve the performance of frozen LLM agents on long-horizon tasks by intelligently managing and pruning accumulated context.

The paper proposes using an auxiliary reconstruction task, specifically one that captures intra-state feature dependencies, to improve the quality of state representations learned by the encoder in neural algorithmic reasoning.

SkillRevise is an execution-grounded framework that iteratively refines initial, imperfect LLM agent skills by diagnosing defects from execution evidence and applying empirically validated edits, significantly boosting agent performance.

The Implicit Drifting Policy (IDP) is a novel one-step action generation framework that implicitly enforces trajectory correction constraints by analyzing local expert action geometry, overcoming the difficulties of explicitly estimating a training-time drifting field.

The paper introduces SkillHarm, a comprehensive benchmark and automated framework for evaluating skill-based attacks across the entire agent skill-use lifecycle, demonstrating that current agents remain highly vulnerable to both fixed-payload and self-mutating poisoning attacks.

QUBRIC introduces a co-design framework that simultaneously optimizes queries and rubrics, overcoming the bottleneck of vague rubrics derived from open-ended questions, leading to significant gains in RL performance.

The paper proposes OneReason, a framework that enhances the reasoning capability of generative recommendation models by focusing on improving item perception and structuring user behavior into coherent latent interests.

This paper introduces CORE-Bench, a comprehensive benchmark for code retrieval in agentic coding.

This paper presents video-SALMONN-R$^3$, an end-to-end video-LLM that enables re-watch through reinforcement learning, improving question answering performance with a two-stage paradigm.

This paper introduces DyadEE, a dataset for emotional entrainment detection in conversational interactions, and TRACE, a window-level framework for modeling dyadic interaction using emotion fine-tuned Whisper representations. TRACE achieves the highest accuracy of 97.01% on DyadEE.

The paper introduces PlanRAG, a framework for Retrieval-Augmented Generation (RAG) that models exploratory reasoning problems as logical query trees, addressing representation and optimization gaps between structured SQL and unstructured natural language.

The paper introduces SPEARBench, a benchmark for evaluating naturalness in speech-to-speech language models using a multidimensional protocol.

The paper presents GIFT, a method for reducing communication volume in large language model pretraining by transforming gradients into a near-isotropic space before quantization.

The paper proposes SALMONN-2, an ALLM built on a unified SSL encoder, and presents a multi-layer feature fusion adapter to better exploit hierarchical SSL encoder representations. It also explores multimodal in-context learning in ALLMs and shows that a general-purpose SSL encoder achieves comparable performance to specialized audio encoders.

The paper introduces ReferTrack, a method for embodied visual tracking using a single forward-facing camera, achieving state-of-the-art performance on EVT-Bench.

The paper introduces an agentic soundscape construction framework for controllable compositional audio generation, which makes explicit the scene planning, source selection, temporal layout, and rendering steps.

This paper presents SpecBox, a runtime system for LLM agents that uses speculative sandbox preallocation to improve resource utilization and reduce interactive tail latency.