arXiv subjects

Chao Zhang

Publications and source records attributed to Chao Zhang.

At least 199 records · Page 11Linked to original sources

arXiv2026

Hierarchical Dual-Strategy Unlearning for Biomedical and Healthcare Intelligence Using Imperfect and Privacy-Sensitive Medical Data

Large language models (LLMs) exhibit exceptional performance but pose substantial privacy risks due to training data memorization, particularly within healthcare contexts involving imperfect or privacy-sensitive patient information. We present a hierarchical dual-strategy framework for selective knowledge unlearning that precisely removes specialized knowledge while preserving fundamental medical competencies. Our approach synergistically integrates geometric-constrained gradient updates to selectively modulate target parameters with concept-aware token-level interventions that distinguish between preservation-critical and unlearning-targeted tokens via a unified four-level medical concept hierarchy. Comprehensive evaluations on the MedMCQA (surgical) and MHQA (anxiety, depression, trauma) datasets demonstrate superior performance, achieving an 82.7% forgetting rate and 88.5% knowledge preservation. Notably, our framework maintains robust privacy guarantees while requiring modification of only 0.1% of parameters, addressing critical needs for regulatory compliance, auditability, and ethical standards in clinical research.

cs.LG↗

arXiv2026

Gradient estimates for nonlinear elliptic equations with Orlicz growth and measure data

We establish gradient estimates of solutions to a class of nonlinear elliptic equations with measure data under Orlicz-type growth conditions. The growth is governed by the structural condition \[ 0<i_a\le t g'(t)/g(t)\le s_a<1. \] We obtain two types of regularity results: pointwise Wolff potential estimates for the gradient of solutions in the singular regime $i_a \in \big(\frac{n-1}{2n-1},1\big)$, and Lipschitz regularity of the solutions in the regime $i_a \in (0,1)$. In the power-type case $g(t)=t^{p-1}$, our results recover the known gradient estimates for the singular $p$-Laplace equation.

math.AP↗

arXiv2026

Superfluid transition of bond bipolarons with long-range Coulomb repulsion in two dimensions

Using numerically exact diagrammatic Monte Carlo simulations in the two-electron (single-bipolaron) sector, we explore the impact of long-range Coulomb repulsion on the dilute-limit Berezinskii--Kosterlitz--Thouless (BKT) transition temperature $T_c$ of bipolarons on a two-dimensional square lattice. We study the bond Su--Schrieffer--Heeger model, in which bond phonons modulate the electron hopping. In the absence of long-range repulsion, this model was shown to support small, light bipolarons with a comparatively high transition temperature \cite{PhysRevX.13.011010}. Here we find that long-range Coulomb repulsion suppresses the optimal $T_c$ but leaves it appreciable over a broad parameter window, including the adiabatic regime $ω/t=0.5$ at a representative Coulomb strength $V=U/10$ (with $U$ the on-site repulsion). Our results provide controlled single-bipolaron inputs for dilute-limit $T_c$ estimates in the presence of long-range repulsion.

cond-mat.supr-con↗

arXiv2026

Adaptation of Agentic AI: A Survey of Post-Training, Memory, and Skills

Large language model (LLM) agents are moving beyond prompting alone. ChatGPT marked the rise of general-purpose LLM assistants, DeepSeek showed that on-policy reinforcement learning with verifiable rewards can improve reasoning and tool use, and OpenClaw highlights a newer direction in which agents accumulate persistent memory and reusable skills. Yet the research landscape remains fragmented across post-training, retrieval, memory, and skill systems. This survey studies these developments under a single notion of \emph{adaptation}: improving an agent, its tools, or their interaction after pretraining. We organize the field with a four-paradigm framework spanning agent adaptation and tool adaptation. On the agent side, A1 (tool-execution-signaled) and A2 (agent-output-signaled) improve the agent itself through supervised fine-tuning, preference optimization, and reinforcement learning with verifiable rewards. On the tool side, T1 (agent-agnostic) provides reusable pre-trained modules any agent can call, while T2 (agent-supervised) uses the agent's outputs to train memory systems, skill libraries, or lightweight subagents. Using this framework, we review post-training methods, adaptive memory architectures, and agent skills; compare their trade-offs in cost, flexibility, and generalization; and summarize evaluation practices across deep research, software development, computer use, and drug discovery. We conclude by outlining open problems in agent-tool co-adaptation, continual learning, safety, and efficient deployment.

cs.AI↗

arXiv2026

EVLF: Early Vision-Language Fusion for Generative Dataset Distillation

Dataset distillation (DD) aims to synthesize compact training sets that enable models to achieve high accuracy with significantly fewer samples. Recent diffusion-based DD methods commonly introduce semantic guidance through late-stage cross-attention, where textual prompts tend to dominate the generative process. Although this strategy enforces label relevance, it diminishes the contribution of visual latents, resulting in over-corrected samples that mirror prompt patterns rather than reflecting intrinsic visual features. To solve this problem, we introduce an Early Vision-Language Fusion (EVLF) method that aligns textual and visual embeddings at the transition between the encoder and the generative backbone. By incorporating a lightweight cross-attention module at this transition, the early representations simultaneously encode local textures and global semantic directions across the denoising process. Importantly, EVLF is plug-and-play and can be easily integrated into any diffusion-based dataset distillation pipeline with an encoder. It works across different denoiser architectures and sampling schedules without any task-specific modifications. Extensive experiments demonstrate that EVLF generates semantically faithful and visually coherent synthetic data, yielding consistent improvements in downstream classification accuracy across varied settings. Source code is available at https://github.com/wenqi-cai297/earlyfusion-for-dd/.

cs.CV↗

arXiv2026

PerfGuard: A Performance-Aware Agent for Visual Content Generation

The advancement of Large Language Model (LLM)-powered agents has enabled automated task processing through reasoning and tool invocation capabilities. However, existing frameworks often operate under the idealized assumption that tool executions are invariably successful, relying solely on textual descriptions that fail to distinguish precise performance boundaries and cannot adapt to iterative tool updates. This gap introduces uncertainty in planning and execution, particularly in domains like visual content generation (AIGC), where nuanced tool performance significantly impacts outcomes. To address this, we propose PerfGuard, a performance-aware agent framework for visual content generation that systematically models tool performance boundaries and integrates them into task planning and scheduling. Our framework introduces three core mechanisms: (1) Performance-Aware Selection Modeling (PASM), which replaces generic tool descriptions with a multi-dimensional scoring system based on fine-grained performance evaluations; (2) Adaptive Preference Update (APU), which dynamically optimizes tool selection by comparing theoretical rankings with actual execution rankings; and (3) Capability-Aligned Planning Optimization (CAPO), which guides the planner to generate subtasks aligned with performance-aware strategies. Experimental comparisons against state-of-the-art methods demonstrate PerfGuard's advantages in tool selection accuracy, execution reliability, and alignment with user intent, validating its robustness and practical utility for complex AIGC tasks. The project code is available at https://github.com/FelixChan9527/PerfGuard.

cs.AI↗

arXiv2026

Impurity Self-Trapping in Lattice Bose systems

We map out the global phase diagram of a single mobile impurity in the two-dimensional Bose-Hubbard model, spanning the bath evolution from a compressible superfluid (SF) to an incompressible Mott insulator (MI) and the full range of impurity-bath coupling. Using sign-problem-free worm-algorithm quantum Monte Carlo, we identify two distinct self-trapping mechanisms that organize the entire diagram. In the compressible SF, increasing impurity-bath coupling $|U_{\mathrm{ib}}|$ drives an interaction-driven self-trapping crossover signaled by a collapse of the \emph{impurity} winding number: a light, extended polaron evolves continuously into a heavy polaron and ultimately into a self-trapped state -- a repulsive \emph{saturated bubble} or an attractive \emph{bound cluster} -- even while the bath remains globally superfluid, demonstrating self-trapping without any bath phase transition. By contrast, when the bath is tuned across the SF-MI transition at fixed $U_{\mathrm{ib}}$, localization is compressibility controlled. The vanishing bath compressibility quenches long-wavelength density redistribution and suppresses polaronic dressing, converting the SF polaron into a weakly dressed, nearly free defect upon entering the MI when $|U_{\mathrm{ib}}| \le 8.0$. Then increasing $|U_{\mathrm{ib}}|$ triggers a distinct Mott-specific route: the impurity binds a quantized vacancy or particle excitation, manifested by discrete changes $ΔN_b=\pm1$ in the total bath occupation. Together, our results provide a unified microscopic picture of impurity self-trapping in correlated lattice bosons, governed by winding collapse in the SF and by compressibility loss and defect quantization across the SF-MI boundary.

cond-mat.quant-gas↗

arXiv2026

A successive difference-of-convex method for a class of two-stage nonconvex nonsmooth stochastic conic program via SVI

We consider a class of two-stage nonconvex nonsmooth stochastic conic program, where the objective functions in both stages can contain nonsmooth terms that are functions with easily computed proximal mappings, further composed with affine mappings. This kind of problem is capable of modeling various applications. Solving these problems, however, can be challenging due to the two-stage structure with possibly large number of scenarios, the nonconvex, nonsmooth and even non-Lipschitz discontinuous terms, as well as the conic constraints. In this paper, we define a KKT point of the problem, show that it is a necessary optimality condition under mild conditions, and transform it to an equivalent nonmonotone nonsmooth two-stage stochastic variational inequality (SVI). We then propose a successive difference-of-convex (SDC) method by making use of Moreau envelope to solve it, the subproblems of which are approximately solved by the progressive hedging method for solving maximal monotone two-stage SVI. We show the rigorous convergence of our method under suitable assumptions. An extension of Markowitz's mean-variance model is provided as an application and numerical results on it demonstrate the effectiveness of the model and the SDC method.

math.OC↗

arXiv2026

iScript: A Domain-Adapted Large Language Model and Benchmark for Physical Design Tcl Script Generation

Modern EDA flows rely heavily on Tcl scripting, yet general LLMs perform poorly in this domain due to extreme data scarcity, domain-specific semantics, and the high reliability required in physical design. We present iScript, a domain-adapted Qwen3-8B model for Innovus Tcl script generation, and iScript-Bench, a comprehensive benchmark covering five task categories and three difficulty levels. To overcome the lack of training data, we introduce a multi-stage data synthesis pipeline that integrates command extraction, static linting, requirement back-inference, and Chain-of-Thought generation, producing a 10K-tuple (requirement, CoT, script) dataset. iScript is trained through a two-stage strategy combining domain-adaptive pretraining and supervised fine-tuning. To evaluate script correctness efficiently, we further propose a two-step verification framework consisting of static syntax verification and LLM-based functional evaluation. On our benchmark, iScript shows higher pass@k scores than currently state-of-the-art LLMs on average. These results demonstrate the effectiveness of domain adaptation and data synthesis for EDA scripting tasks.

cs.SE↗

arXiv2026

Pricing for Information Revelation in Demand Response: A Strategic Communication Approach

Many smart grid frameworks, such as demand response programs, require accurate information about consumers' parameters (e.g., flexibility) at the aggregator side to optimize grid operations. Existing works typically rely on perfect information assumptions or complex incentive-compatible mechanisms; however, in voluntary settings, and in the presence of strategic consumers, possibly implemented by automated intelligent agents, private parameters may be misreported due to strategic incentives. We analyze this communication setting using cheap-talk game theory, delivering four key insights. First, the nontrivial scenario of multiple strategic transmitters (consumers) turns out to be tractable for the case study of interest: we prove that complex strategic interactions among multiple consumers decouple into independent subgames. Second, we demonstrate that a pre-announced retail price can be exploited as a design lever to control the information revealed by the consumers and therefore the overall system efficiency. Third, we derive a closed-form expression for the optimal uniform price that maximizes information revelation. Finally, we characterize the equilibrium structure to identify when communication is informative. Simulations show that a properly designed price for the communication scheme can recover up to 95% of the ideal system utility (i.e., under perfect information reporting), whereas a price-unaware choice leads to significant losses in social welfare.

cs.GT↗

arXiv2026

DAM-VLA: A Dynamic Action Model-Based Vision-Language-Action Framework for Robot Manipulation

In dynamic environments such as warehouses, hospitals, and homes, robots must seamlessly transition between gross motion and precise manipulations to complete complex tasks. However, current Vision-Language-Action (VLA) frameworks, largely adapted from pre-trained Vision-Language Models (VLMs), often struggle to reconcile general task adaptability with the specialized precision required for intricate manipulation. To address this challenge, we propose DAM-VLA, a dynamic action model-based VLA framework. DAM-VLA integrates VLM reasoning with diffusion-based action models specialized for arm and gripper control. Specifically, it introduces (i) an action routing mechanism, using task-specific visual and linguistic cues to select appropriate action models (e.g., arm movement or gripper manipulation), (ii) a dynamic action model that fuses high-level VLM cognition with low-level visual features to predict actions, and (iii) a dual-scale action weighting mechanism that enables dynamic coordination between the arm-movement and gripper-manipulation models. Across extensive evaluations, DAM-VLA achieves superior success rates compared to state-of-the-art VLA methods in simulated (SIMPLER, FurnitureBench) and real-world settings, showing robust generalization from standard pick-and-place to demanding long-horizon and contact-rich tasks.

cs.RO↗

arXiv2026

Polarization-aware Reconfigurable Antenna Aided Wireless Communications

Reconfigurable antennas (RAs) have emerged as a promising technology for future wireless networks, offering additional flexibility for wireless communications. Among existing designs, rotatable antennas are particularly effective in improving directional gain via boresight alignment only. However, conventional rotatable RAs often overlook a critical physical coupling: the mechanical rotation inevitably alters the radiated polarization orientation, potentially leading to polarization mismatch. To address this challenge, we investigate a novel RA architecture that simultaneously supports 3D rotation and polarization state reconfiguration, ensuring alignment in both spatial and polarization domains. To quantify the performance gains, we analyze a simplified single-user LoS scenario to compare the optimized rotatable design against a fixed scheme. This analysis attributes the performance improvement to three aspects: directional and projection gain arising from boresight steering, polarization direction alignment gain enabled by roll adjustment, and polarization state matching gain provided by polarization reconfiguration. Furthermore, for general multipath multi-user systems, we formulate a joint power minimization problem by optimizing digital beamforming alongside rotation and polarization designs, subject to rate and hardware constraints. To solve the resulting non-convex problem efficiently, we develop an alternating optimization framework, where the digital beamforming is solved via semidefinite relaxation and difference-of-convex techniques, while the rotation and polarization designs are updated using Riemannian conjugate gradient on their respective manifolds. Simulation results demonstrate that the proposed RA outperforms both rotation-only and boresight-only benchmarks, achieving lower transmit power under the same rate constraints by joint spatial-polarization design.

eess.SP↗

arXiv2026

Audio-Conditioned Diffusion LLMs for ASR and Deliberation Processing

Diffusion-based large language models (DLLMs) have recently attracted growing interest as an alternative to autoregressive decoders. In this work, we present an empirical study on using the diffusion-based large language model LLaDA for automatic speech recognition (ASR). We first investigate its use as an external deliberation-based processing module for Whisper-LLaMA transcripts. By leveraging the bidirectional attention and denoising capabilities of LLaDA, we explore random masking, low-confidence masking, and semi-autoregressive strategies, showing that Whisper-LLaDA substantially reduces WER compared with the baseline. On LibriSpeech, the best cascade system achieves 2.25%/4.94% WER on test-clean/test-other, representing a 12.3% relative improvement over the Whisper-LLaMA baseline on the test-other split. In contrast, a plain-text LLaDA without acoustic features fails to improve accuracy, highlighting the importance of audio-conditioned embeddings. We further evaluate Whisper-LLaDA as a standalone decoder for ASR with diffusion-based and semi-autoregressive decoding. Most experimental configurations achieve faster inference than the Whisper-LLaMA baseline, although recognition accuracy is slightly lower. These findings offer an empirical view of diffusion-based LLMs for ASR and point to promising directions for improvements. Code and model are open-sourced at https://github.com/liuzhan22/Diffusion-ASR.

eess.AS↗

arXiv2026

SciTS: Scientific Time Series Understanding and Generation with LLMs

The scientific reasoning ability of large language models (LLMs) has recently attracted significant attention. Time series, as a fundamental modality in scientific data, presents unique challenges that are often overlooked in current multimodal LLMs, which either encode numerical sequences as text or convert them into images. Such approaches may be insufficient for comprehensive scientific time series understanding and generation. Existing unified time series models typically specialise in either forecasting or analysis, and their effectiveness on non-periodic, heterogeneous scientific signals remains unclear. To address these gaps, we introduce SciTS, a benchmark spanning 12 scientific domains and 43 tasks, with over 50k+ instances, both univariate and multivariate signals ranging from $10^0$ to $10^7$ in length and up to 10~MHz in frequency. We benchmark 17 models, including text-only LLMs, multimodal LLMs, and unified time series models, and find that general-purpose LLMs exhibit stronger generalisability than specialised time series models, while representing time series as text or images limits their performance due to excessively long sequences and loss of numerical precision, respectively. We then introduce TimeOmni, a framework that equips LLMs with the ability to understand and generate time series while remaining compatible with general-purpose LLM training. This work fills a gap in both dedicated benchmarks and modelling frameworks for scientific time series, paving the way for LLMs to understand and generate complex temporal scientific data.

cs.LG↗

arXiv2026

One Brain, Omni Modalities: Towards Unified Non-Invasive Brain Decoding with Large Language Models

Deciphering brain function through non-invasive recordings requires synthesizing complementary high-frequency electromagnetic (EEG/MEG) and low-frequency metabolic (fMRI) signals. However, despite their shared neural origins, extreme discrepancies have traditionally confined these modalities to isolated analysis pipelines, hindering a holistic interpretation of brain activity. To bridge this fragmentation, we introduce \textbf{NOBEL}, a \textbf{n}euro-\textbf{o}mni-modal \textbf{b}rain-\textbf{e}ncoding \textbf{l}arge language model (LLM) that unifies these heterogeneous signals within the LLM's semantic embedding space. Our architecture integrates a unified encoder for EEG and MEG with a novel dual-path strategy for fMRI, aligning non-invasive brain signals and external sensory stimuli into a shared token space, then leverages an LLM as a universal backbone. Extensive evaluations demonstrate that NOBEL serves as a robust generalist across standard single-modal tasks. We also show that the synergistic fusion of electromagnetic and metabolic signals yields higher decoding accuracy than unimodal baselines, validating the complementary nature of multiple neural modalities. Furthermore, NOBEL exhibits strong capabilities in stimulus-aware decoding, effectively interpreting visual semantics from multi-subject fMRI data on the NSD and HAD datasets while uniquely leveraging direct stimulus inputs to verify causal links between sensory signals and neural responses. NOBEL thus takes a step towards unifying non-invasive brain decoding, demonstrating the promising potential of omni-modal brain understanding.

q-bio.NC↗

arXiv2026

WAVE: Learning Unified & Versatile Audio-Visual Embeddings with Multimodal LLM

While embeddings from multimodal large language models (LLMs) excel as general-purpose representations, their application to dynamic modalities like audio and video remains underexplored. We introduce WAVE (\textbf{u}nified \& \textbf{v}ersatile \textbf{a}udio-\textbf{v}isual \textbf{e}mbeddings), the first LLM-based embedding that creates a unified representation space for text, audio, and video modalities. WAVE employs a novel hierarchical feature fusion strategy and a joint multi-modal, multi-task training approach to enable two key capabilities: any-to-any cross-modal retrieval and the generation of prompt-aware embeddings tailored to user instructions. Experimentally, WAVE sets a new state-of-the-art on the MMEB-v2 video benchmark and achieves superior results in audio and video-to-audio retrieval. Its prompt-aware nature also yields remarkable performance in multimodal question answering, significantly outperforming existing embedding models. Ablation studies validate our joint training strategy, demonstrating improved performance across all modalities. With a newly introduced benchmark for versatile audio-visual learning, WAVE opens up broad possibilities for cross-modal, any-to-any applications. Our code and checkpoints are released at \href{https://github.com/TCL606/WAVE}{https://github.com/TCL606/WAVE}.

cs.CV↗

arXiv2026

Nuclear cluster structure effect in $^{16}$O+$^{16}$O collisions at the top RHIC energy

Using the improved string-melting version of a Multi-Phase Transport model, we investigated the impact of nuclear geometry of $^{16}$O on anisotropic flows in O+O collisions at $\sqrt{s_{\rm NN}} = 200$ GeV. To evaluate the influence of nuclear structure and potential alpha clustering, we implemented four candidate configurations: Woods-Saxon, tetrahedron, square, and Nuclear Lattice Effective Field Theory. Initial-state geometry is quantified via the eccentricity cumulant ratio $\varepsilon_{2}\{4\}/\varepsilon_{2}\{2\}$, which provides a robust and evolution-independent measure sensitive to configuration differences. The model reproduces $v_{2}(p_{\rm T})$ at low $p_{\rm T}$ and $v_{3}(p_{\rm T})$ across the full $p_{\rm T}$ range, with integrated $v_{2}\{2\}$ and $v_{3}\{2\}$ matching the STAR data, demonstrating that transport dynamics captures the essential collectivity in this intermediate-size system. These findings establish a baseline for extending nuclear-structure studies in O+O collisions to other energies and differential observables within a unified transport model framework.

nucl-th↗

arXiv2026

Label-Consistent Dataset Distillation with Detector-Guided Refinement

Dataset distillation (DD) aims to generate a compact yet informative dataset that achieves performance comparable to the original dataset, thereby reducing demands on storage and computational resources. Although diffusion models have made significant progress in dataset distillation, the generated surrogate datasets often contain samples with label inconsistencies or insufficient structural detail, leading to suboptimal downstream performance. To address these issues, we propose a detector-guided dataset distillation framework that explicitly leverages a pre-trained detector to identify and refine anomalous synthetic samples, thereby ensuring label consistency and improving image quality. Specifically, a detector model trained on the original dataset is employed to identify anomalous images exhibiting label mismatches or low classification confidence. For each defective image, multiple candidates are generated using a pre-trained diffusion model conditioned on the corresponding image prototype and label. The optimal candidate is then selected by jointly considering the detector's confidence score and dissimilarity to existing qualified synthetic samples, thereby ensuring both label accuracy and intra-class diversity. Experimental results demonstrate that our method can synthesize high-quality representative images with richer details, achieving state-of-the-art performance on the validation set.

cs.CV↗