arXiv subjects

Ziyu Zhou

Publications and source records attributed to Ziyu Zhou.

At least 37 records · Page 2Linked to original sources

arXiv2026

Lamps: Learning Anatomy from Multiple Perspectives via Self-supervision in Chest Radiographs

Foundation models have been successful in natural language processing and computer vision because they are capable of capturing the underlying structures (foundation) of natural languages. However, in medical imaging, the key foundation lies in human anatomy, as these images directly represent the internal structures of the body, reflecting the consistency, coherence, and hierarchy of human anatomy. Yet, existing self-supervised learning (SSL) methods often overlook these perspectives, limiting their ability to effectively learn anatomical features. To overcome the limitation, we built Lamps (learning anatomy from multiple perspectives via self-supervision) pre-trained on large-scale chest radiographs by harmoniously utilizing the consistency, coherence, and hierarchy of human anatomy as the supervision signal. Extensive experiments across 10 datasets evaluated through fine-tuning and emergent property analysis demonstrate Lamps' superior robustness, transferability, and clinical potential when compared to 10 baseline models. By learning from multiple perspectives, Lamps presents a unique opportunity for foundation models to develop meaningful, robust representations that are aligned with the structure of human anatomy.

cs.CV↗

arXiv2025

RAST: A Retrieval Augmented Spatio-Temporal Framework for Traffic Prediction

Traffic prediction is a cornerstone of modern intelligent transportation systems and a critical task in spatio-temporal forecasting. Although advanced Spatio-temporal Graph Neural Networks (STGNNs) and pre-trained models have achieved significant progress in traffic prediction, two key challenges remain: (i) limited contextual capacity when modeling complex spatio-temporal dependencies, and (ii) low predictability at fine-grained spatio-temporal points due to heterogeneous patterns. Inspired by Retrieval-Augmented Generation (RAG), we propose RAST, a universal framework that integrates retrieval-augmented mechanisms with spatio-temporal modeling to address these challenges. Our framework consists of three key designs: 1) Decoupled Encoder and Query Generator to capture decoupled spatial and temporal features and construct a fusion query via residual fusion; 2) Spatio-temporal Retrieval Store and Retrievers to maintain and retrieve vectorized fine-grained patterns; and 3) Universal Backbone Predictor that flexibly accommodates pre-trained STGNNs or simple MLP predictors. Extensive experiments on six real-world traffic networks, including large-scale datasets, demonstrate that RAST achieves superior performance while maintaining computational efficiency.

cs.LG↗

arXiv2025

Revitalizing Canonical Pre-Alignment for Irregular Multivariate Time Series Forecasting

Irregular multivariate time series (IMTS), characterized by uneven sampling and inter-variate asynchrony, fuel many forecasting applications yet remain challenging to model efficiently. Canonical Pre-Alignment (CPA) has been widely adopted in IMTS modeling by padding zeros at every global timestamp, thereby alleviating inter-variate asynchrony and unifying the series length, but its dense zero-padding inflates the pre-aligned series length, especially when numerous variates are present, causing prohibitive compute overhead. Recent graph-based models with patching strategies sidestep CPA, but their local message passing struggles to capture global inter-variate correlations. Therefore, we posit that CPA should be retained, with the pre-aligned series properly handled by the model, enabling it to outperform state-of-the-art graph-based baselines that sidestep CPA. Technically, we propose KAFNet, a compact architecture grounded in CPA for IMTS forecasting that couples (1) Pre-Convolution module for sequence smoothing and sparsity mitigation, (2) Temporal Kernel Aggregation module for learnable compression and modeling of intra-series irregularity, and (3) Frequency Linear Attention blocks for the low-cost inter-series correlations modeling in the frequency domain. Experiments on multiple IMTS datasets show that KAFNet achieves state-of-the-art forecasting performance, with a 7.2$\times$ parameter reduction and a 8.4$\times$ training-inference acceleration.

cs.LG↗

arXiv2025

Numerical Investigation of Discontinuous Ice Effects on Swept Wings

This study investigates the aerodynamic performance and flow structures of infinite swept wings with artificially simulated discontinuous ice using an enhanced delayed detached-eddy simulation. Comparisons are made among clean, continuous-ice, and discontinuous-ice configurations. Results show that discontinuous ice causes a more severe reduction in lift than continuous ice. While continuous ice forms a large separation bubble that helps maintain lift, discontinuous ice disrupts leading-edge vortex formation through gap jets, resulting in greater lift loss but a smaller drag penalty. Unlike the continuous-ice wing, the discontinuous-ice case does not exhibit a sudden stall-induced lift drop. The flow over the discontinuous-ice wing can be characterized by two canonical patterns: a separating shear layer and Kármán vortex shedding. However, the separating shear layer becomes irregular due to the interference of gap jets. Three characteristic chord-based Strouhal numbers (St)-11.3, 22.6, and 33.9-are identified. The lowest (St=11.3) corresponds to the shedding of vortex pairs; when nondimensionalized by the ice width, it yields St = 0.58, which is higher than that of a canonical cylinder wake. Furthermore, lift and drag fluctuations occur predominantly at St = 22.6, twice the shedding frequency, primarily induced by the gap jets-a phenomenon absent in the continuous-ice case.

physics.flu-dyn↗

arXiv2025

XBench: A Comprehensive Benchmark for Visual-Language Explanations in Chest Radiography

Vision-language models (VLMs) have recently shown remarkable zero-shot performance in medical image understanding, yet their grounding ability, the extent to which textual concepts align with visual evidence, remains underexplored. In the medical domain, however, reliable grounding is essential for interpretability and clinical adoption. In this work, we present the first systematic benchmark for evaluating cross-modal interpretability in chest X-rays across seven CLIP-style VLM variants. We generate visual explanations using cross-attention and similarity-based localization maps, and quantitatively assess their alignment with radiologist-annotated regions across multiple pathologies. Our analysis reveals that: (1) while all VLM variants demonstrate reasonable localization for large and well-defined pathologies, their performance substantially degrades for small or diffuse lesions; (2) models that are pretrained on chest X-ray-specific datasets exhibit improved alignment compared to those trained on general-domain data. (3) The overall recognition ability and grounding ability of the model are strongly correlated. These findings underscore that current VLMs, despite their strong recognition ability, still fall short in clinically reliable grounding, highlighting the need for targeted interpretability benchmarks before deployment in medical practice. XBench code is available at https://github.com/Roypic/Benchmarkingattention

cs.CV↗

arXiv2025

Progressive Scale Convolutional Network for Spatio-Temporal Downscaling of Soil Moisture: A Case Study Over the Tibetan Plateau

Soil moisture (SM) plays a critical role in hydrological and meteorological processes. High-resolution SM can be obtained by combining coarse passive microwave data with fine-scale auxiliary variables. However, the inversion of SM at the temporal scale is hindered by the incompleteness of surface auxiliary factors. To address this issue, first, we introduce validated high temporal resolution ERA5-Land variables into the downscaling process of the low-resolution SMAP SM product. Subsequently, we design a progressive scale convolutional network (PSCNet), at the core of which are two innovative components: a multi-frequency temporal fusion module (MFTF) for capturing temporal dynamics, and a bespoke squeeze-and-excitation (SE) block designed to preserve fine-grained spatial details. Using this approach, we obtained seamless SM products for the Tibetan Plateau (TP) from 2016 to 2018 at 10-km spatial and 3-hour temporal resolution. The experimental results on the TP demonstrated the following: 1) In the satellite product validation, the PSCNet exhibited comparable accuracy and lower error, with a mean R value of 0.881, outperforming other methods. 2) In the in-situ site validation, PSCNet consistently ranked among the top three models for the R metric across all sites, while also showing superior performance in overall error reduction. 3) In the temporal generalization validation, the feasibility of using high-temporal resolution ERA5-Land variables for downscaling was confirmed, as all methods maintained an average relative error within 6\% for the R metric and 2\% for the ubRMSE metric. 4) In the temporal dynamics and visualization validation, PSCNet demonstrated excellent temporal sensitivity and vivid spatial details. Overall, PSCNet provides a promising solution for spatio-temporal downscaling by effectively modeling the intricate spatio-temporal relationships in SM data.

cs.LG↗

arXiv2025

Personalized Federated Dictionary Learning for Modeling Heterogeneity in Multi-site fMRI Data

Data privacy constraints pose significant challenges for large-scale neuroimaging analysis, especially in multi-site functional magnetic resonance imaging (fMRI) studies, where site-specific heterogeneity leads to non-independent and identically distributed (non-IID) data. These factors hinder the development of generalizable models. To address these challenges, we propose Personalized Federated Dictionary Learning (PFedDL), a novel federated learning framework that enables collaborative modeling across sites without sharing raw data. PFedDL performs independent dictionary learning at each site, decomposing each site-specific dictionary into a shared global component and a personalized local component. The global atoms are updated via federated aggregation to promote cross-site consistency, while the local atoms are refined independently to capture site-specific variability, thereby enhancing downstream analysis. Experiments on the ABIDE dataset demonstrate that PFedDL outperforms existing methods in accuracy and robustness across non-IID datasets.

cs.LG↗

arXiv2025

Non-Abelian geometry, topology, and dynamics of a nonreciprocal Su-Schrieffer-Heeger ladder

Non-Hermiticity naturally breaks down the adiabaticity and thus leads to non-Abelian behaviors in multi-band systems. Here, we study how non-Abelian properties emerge in non-Hermitian systems by considering a multi-band non-Hermitian model -- the nonreciprocal Su-Schrieffer-Heeger (SSH) ladder that is formed by coupling two nonreciprocal SSH chains. Under periodic boundary conditions, we analytically obtain the exact phase diagrams of the geometry of band structure classified by its complex value and gap type, and of the non-Abelian topology based on a newly defined gauge-invariant winding number under the chiral symmetry. Under open boundary conditions, we find that the bulk-boundary correspondence survives in the thermodynamic limit but breaks down for finite sizes along with the emergence of critical non-Hermitian skin effects when the inter-leg coupling is weak, where the decaying length $ξ$ of the bulk skin modes varies with the system size $L$, satisfying the scale-free power law $ξ\propto L$. Finally, we demonstrate the non-Abelian dynamics of a Bloch state subject to an external constant force in the pseudo-Hermitian symmetric regime in comparison with the non-Hermitian Wilson lines. Our work may stimulate further interests in nontrivial non-Abelian behaviors in non-Hermitian and open quantum systems.

cond-mat.mes-hall↗

arXiv2025

A numerical investigation of sweep effects on turbulent flow over iced wings

This study employs an improved delayed detached eddy simulation (AMD-IDDES) method to investigate the flow characteristics of an iced swept wing. The AMD-IDDES approach incorporates an anisotropic minimum-dissipation (AMD) model into the traditional IDDES framework to better account for local flow anisotropy and energy dissipation characteristics. The method is validated against the traditional IDDES approach, and the results show that AMD-IDDES more accurately captures the evolution of vortices and separated shear layers. Compared with an iced straight wing, a swept wing exhibits distinct aerodynamic behavior driven mainly by the sweep angle. The sweep angle induces strong spanwise flow, which reshapes the separation region and transforms the flow from two-dimensional to three-dimensional. This spanwise motion significantly alters vortex development and enhances the complexity of the unsteady flow. The shear layer separation, initiated by a Kelvin-Helmholtz instability, dominates the unsteady aerodynamic response. Although wingtip effects remain secondary, their interaction with the leading-edge vortex still contributes to lowering the dominant frequency. Overall, this study highlights that the aerodynamic forces of an iced swept wing are primarily governed by sweep-induced spanwise flow and associated shear layer dynamics, with minimal influence from the root and tip regions.

physics.flu-dyn↗

arXiv2025

On the Interplay of Human-AI Alignment,Fairness, and Performance Trade-offs in Medical Imaging

Deep neural networks excel in medical imaging but remain prone to biases, leading to fairness gaps across demographic groups. We provide the first systematic exploration of Human-AI alignment and fairness in this domain. Our results show that incorporating human insights consistently reduces fairness gaps and enhances out-of-domain generalization, though excessive alignment can introduce performance trade-offs, emphasizing the need for calibrated strategies. These findings highlight Human-AI alignment as a promising approach for developing fair, robust, and generalizable medical AI systems, striking a balance between expert guidance and automated efficiency. Our code is available at https://github.com/Roypic/Aligner.

cs.CV↗

arXiv2025

Evaluating the Effectiveness of Black-Box Prompt Optimization as the Scale of LLMs Continues to Grow

Black-Box prompt optimization methods have emerged as a promising strategy for refining input prompts to better align large language models (LLMs), thereby enhancing their task performance. Although these methods have demonstrated encouraging results, most studies and experiments have primarily focused on smaller-scale models (e.g., 7B, 14B) or earlier versions (e.g., GPT-3.5) of LLMs. As the scale of LLMs continues to increase, such as with DeepSeek V3 (671B), it remains an open question whether these black-box optimization techniques will continue to yield significant performance improvements for models of such scale. In response to this, we select three well-known black-box optimization methods and evaluate them on large-scale LLMs (DeepSeek V3 and Gemini 2.0 Flash) across four NLU and NLG datasets. The results show that these black-box prompt optimization methods offer only limited improvements on these large-scale LLMs. Furthermore, we hypothesize that the scale of the model is the primary factor contributing to the limited benefits observed. To explore this hypothesis, we conducted experiments on LLMs of varying sizes (Qwen 2.5 series, ranging from 7B to 72B) and observed an inverse scaling law, wherein the effectiveness of black-box optimization methods diminished as the model size increased.

cs.CL↗

arXiv2025

Integrated Brain Connectivity Analysis with fMRI, DTI, and sMRI Powered by Interpretable Graph Neural Networks

Multimodal neuroimaging modeling has becomes a widely used approach but confronts considerable challenges due to heterogeneity, which encompasses variability in data types, scales, and formats across modalities. This variability necessitates the deployment of advanced computational methods to integrate and interpret these diverse datasets within a cohesive analytical framework. In our research, we amalgamate functional magnetic resonance imaging, diffusion tensor imaging, and structural MRI into a cohesive framework. This integration capitalizes on the unique strengths of each modality and their inherent interconnections, aiming for a comprehensive understanding of the brain's connectivity and anatomical characteristics. Utilizing the Glasser atlas for parcellation, we integrate imaging derived features from various modalities: functional connectivity from fMRI, structural connectivity from DTI, and anatomical features from sMRI within consistent regions. Our approach incorporates a masking strategy to differentially weight neural connections, thereby facilitating a holistic amalgamation of multimodal imaging data. This technique enhances interpretability at connectivity level, transcending traditional analyses centered on singular regional attributes. The model is applied to the Human Connectome Project's Development study to elucidate the associations between multimodal imaging and cognitive functions throughout youth. The analysis demonstrates improved predictive accuracy and uncovers crucial anatomical features and essential neural connections, deepening our understanding of brain structure and function.

q-bio.NC↗

arXiv2025

Neural Approaches to SAT Solving: Design Choices and Interpretability

In this contribution, we provide a comprehensive evaluation of graph neural networks applied to Boolean satisfiability problems, accompanied by an intuitive explanation of the mechanisms enabling the model to generalize to different instances. We introduce several training improvements, particularly a novel closest assignment supervision method that dynamically adapts to the model's current state, significantly enhancing performance on problems with larger solution spaces. Our experiments demonstrate the suitability of variable-clause graph representations with recurrent neural network updates, which achieve good accuracy on SAT assignment prediction while reducing computational demands. We extend the base graph neural network into a diffusion model that facilitates incremental sampling and can be effectively combined with classical techniques like unit propagation. Through analysis of embedding space patterns and optimization trajectories, we show how these networks implicitly perform a process very similar to continuous relaxations of MaxSAT, offering an interpretable view of their reasoning process. This understanding guides our design choices and explains the ability of recurrent architectures to scale effectively at inference time beyond their training distribution, which we demonstrate with test-time scaling experiments.

cs.LG↗

arXiv2025

WorkTeam: Constructing Workflows from Natural Language with Multi-Agents

Workflows play a crucial role in enhancing enterprise efficiency by orchestrating complex processes with multiple tools or components. However, hand-crafted workflow construction requires expert knowledge, presenting significant technical barriers. Recent advancements in Large Language Models (LLMs) have improved the generation of workflows from natural language instructions (aka NL2Workflow), yet existing single LLM agent-based methods face performance degradation on complex tasks due to the need for specialized knowledge and the strain of task-switching. To tackle these challenges, we propose WorkTeam, a multi-agent NL2Workflow framework comprising a supervisor, orchestrator, and filler agent, each with distinct roles that collaboratively enhance the conversion process. As there are currently no publicly available NL2Workflow benchmarks, we also introduce the HW-NL2Workflow dataset, which includes 3,695 real-world business samples for training and evaluation. Experimental results show that our approach significantly increases the success rate of workflow construction, providing a novel and effective solution for enterprise NL2Workflow services.

cs.CL↗

arXiv2025

SChanger: Change Detection from a Semantic Change and Spatial Consistency Perspective

Change detection is a key task in Earth observation applications. Recently, deep learning methods have demonstrated strong performance and widespread application. However, change detection faces data scarcity due to the labor-intensive process of accurately aligning remote sensing images of the same area, which limits the performance of deep learning algorithms. To address the data scarcity issue, we develop a fine-tuning strategy called the Semantic Change Network (SCN). We initially pre-train the model on single-temporal supervised tasks to acquire prior knowledge of instance feature extraction. The model then employs a shared-weight Siamese architecture and extended Temporal Fusion Module (TFM) to preserve this prior knowledge and is fine-tuned on change detection tasks. The learned semantics for identifying all instances is changed to focus on identifying only the changes. Meanwhile, we observe that the locations of changes between the two images are spatially identical, a concept we refer to as spatial consistency. We introduce this inductive bias through an attention map that is generated by large-kernel convolutions and applied to the features from both time points. This enhances the modeling of multi-scale changes and helps capture underlying relationships in change detection semantics. We develop a binary change detection model utilizing these two strategies. The model is validated against state-of-the-art methods on six datasets, surpassing all benchmark methods and achieving F1 scores of 92.87%, 86.43%, 68.95%, 97.62%, 84.58%, and 93.20% on the LEVIR-CD, LEVIR-CD+, S2Looking, CDD, SYSU-CD, and WHU-CD datasets, respectively.

cs.CV↗

arXiv2025

Gradient Co-occurrence Analysis for Detecting Unsafe Prompts in Large Language Models

Unsafe prompts pose significant safety risks to large language models (LLMs). Existing methods for detecting unsafe prompts rely on data-driven fine-tuning to train guardrail models, necessitating significant data and computational resources. In contrast, recent few-shot gradient-based methods emerge, requiring only few safe and unsafe reference prompts. A gradient-based approach identifies unsafe prompts by analyzing consistent patterns of the gradients of safety-critical parameters in LLMs. Although effective, its restriction to directional similarity (cosine similarity) introduces ``directional bias'', limiting its capability to identify unsafe prompts. To overcome this limitation, we introduce GradCoo, a novel gradient co-occurrence analysis method that expands the scope of safety-critical parameter identification to include unsigned gradient similarity, thereby reducing the impact of ``directional bias'' and enhancing the accuracy of unsafe prompt detection. Comprehensive experiments on the widely-used benchmark datasets ToxicChat and XStest demonstrate that our proposed method can achieve state-of-the-art (SOTA) performance compared to existing methods. Moreover, we confirm the generalizability of GradCoo in detecting unsafe prompts across a range of LLM base models with various sizes and origins.

cs.CL↗

arXiv2025

LF-Steering: Latent Feature Activation Steering for Enhancing Semantic Consistency in Large Language Models

Large Language Models (LLMs) often generate inconsistent responses when prompted with semantically equivalent paraphrased inputs. Recently, activation steering, a technique that modulates LLMs' behaviours by adjusting their latent representations during inference time, has been explored to improve the semantic consistency of LLMs. However, these methods typically operate at the model component level, such as layer hidden states or attention head outputs. They face a challenge due to the ``polysemanticity issue'', where the model components of LLMs typically encode multiple entangled features, making precise steering difficult. To address this challenge, we drill down to feature-level representations and propose LF-Steering, a novel activation steering approach to precisely identify latent feature representations responsible for semantic inconsistency. More specifically, our method maps the hidden states of the relevant transformer layer into a sparsely activated, high-dimensional feature space based on a sparse autoencoder (SAE), ensuring model steering based on decoupled feature representations with minimal interference. Comprehensive experiments on NLU and NLG datasets demonstrate the effectiveness of our method in enhancing semantic consistency, resulting in significant performance gains for various NLU and NLG tasks.

cs.CL↗

arXiv2025

ACE: Anatomically Consistent Embeddings in Composition and Decomposition

Medical images acquired from standardized protocols show consistent macroscopic or microscopic anatomical structures, and these structures consist of composable/decomposable organs and tissues, but existing self-supervised learning (SSL) methods do not appreciate such composable/decomposable structure attributes inherent to medical images. To overcome this limitation, this paper introduces a novel SSL approach called ACE to learn anatomically consistent embedding via composition and decomposition with two key branches: (1) global consistency, capturing discriminative macro-structures via extracting global features; (2) local consistency, learning fine-grained anatomical details from composable/decomposable patch features via corresponding matrix matching. Experimental results across 6 datasets 2 backbones, evaluated in few-shot learning, fine-tuning, and property analysis, show ACE's superior robustness, transferability, and clinical potential. The innovations of our ACE lie in grid-wise image cropping, leveraging the intrinsic properties of compositionality and decompositionality of medical images, bridging the semantic gap from high-level pathologies to low-level tissue anomalies, and providing a new SSL method for medical imaging.

cs.CV↗