arXiv subjects

Shiyuan He

Publications and source records attributed to Shiyuan He.

At least 19 recordsLinked to original sources

arXiv2026

MASRA: MLLM-Assisted Semantic-Relational Consistent Alignment for Video Temporal Grounding

Video Temporal Grounding (VTG) faces a cross-modal semantic gap that often leads to background features being incorrectly aligned with the query, while directly matching the query to moments results in insufficient discriminability and consistency of temporal semantics. To address this issue, we propose MLLM-Assisted Semantic-Relational Consistent Alignment (MASRA), a training-time MLLM-based optimization framework for VTG. MASRA leverages an MLLM during training to produce two forms of textual priors, namely event-level descriptions with temporal spans and clip-level captions, and instantiates two MLLM-assisted alignments. Event Semantic Temporal Alignment (ESTA) aligns temporal context with event semantics to explicitly strengthen the correspondence between semantics and temporal events and improve span-level separability. Local Relational Consistency Alignment (LRCA) constructs a textual relation matrix derived from clip-level captions and aligns it with the temporal feature similarity matrix in the model, enhancing temporal consistency while capturing local structural information. MASRA includes two simple supporting modules, semantic-guided enhancement and second-order relational attention, to better utilize the learned semantic context and relational structure. Moreover, we introduce Decoupled Alignment Interaction (DAI) with a context-aware codebook to adaptively absorb query-irrelevant semantics and alleviate the cross-modal gap. The MLLM is only invoked during training and is not used at inference. Extensive experiments show that MASRA outperforms existing methods, and ablation studies validate its effectiveness.

cs.CV↗

arXiv2025

Revisiting Madigan and Mosurski: Collapsibility via Minimal Separators

Collapsibility provides a principled approach for dimension reduction in contingency tables and graphical models. Madigan and Mosurski (1990) pioneered the study of minimal collapsible sets in decomposable models, but existing algorithms for general graphs remain computationally demanding. We show that a model is collapsible onto a target set precisely when that set contains at least one minimal separator between its non-adjacent vertices. This insight motivates the Close Minimal Separator Absorption (CMSA) algorithm, which constructs minimal collapsible sets using only local separator searches at very low costs. Simulations confirm substantial efficiency gains, making collapsibility analysis practical in high-dimensional settings.

stat.ME↗

arXiv2025

MM-R1: Unleashing the Power of Unified Multimodal Large Language Models for Personalized Image Generation

Multimodal Large Language Models (MLLMs) with unified architectures excel across a wide range of vision-language tasks, yet aligning them with personalized image generation remains a significant challenge. Existing methods for MLLMs are frequently subject-specific, demanding a data-intensive fine-tuning process for every new subject, which limits their scalability. In this paper, we introduce MM-R1, a framework that integrates a cross-modal Chain-of-Thought (X-CoT) reasoning strategy to unlock the inherent potential of unified MLLMs for personalized image generation. Specifically, we structure personalization as an integrated visual reasoning and generation process: (1) grounding subject concepts by interpreting and understanding user-provided images and contextual cues, and (2) generating personalized images conditioned on both the extracted subject representations and user prompts. To further enhance the reasoning capability, we adopt Grouped Reward Proximal Policy Optimization (GRPO) to explicitly align the generation. Experiments demonstrate that MM-R1 unleashes the personalization capability of unified MLLMs to generate images with high subject fidelity and strong text alignment in a zero-shot manner.

cs.CV↗

arXiv2025

Pushing the Limits of Safety: A Technical Report on the ATLAS Challenge 2025

Multimodal Large Language Models (MLLMs) have enabled transformative advancements across diverse applications but remain susceptible to safety threats, especially jailbreak attacks that induce harmful outputs. To systematically evaluate and improve their safety, we organized the Adversarial Testing & Large-model Alignment Safety Grand Challenge (ATLAS) 2025}. This technical report presents findings from the competition, which involved 86 teams testing MLLM vulnerabilities via adversarial image-text attacks in two phases: white-box and black-box evaluations. The competition results highlight ongoing challenges in securing MLLMs and provide valuable guidance for developing stronger defense mechanisms. The challenge establishes new benchmarks for MLLM safety evaluation and lays groundwork for advancing safer multimodal AI systems. The code and data for this challenge are openly available at https://github.com/NY1024/ATLAS_Challenge_2025.

cs.CR↗

arXiv2025

InPK: Infusing Prior Knowledge into Prompt for Vision-Language Models

Prompt tuning has become a popular strategy for adapting Vision-Language Models (VLMs) to zero/few-shot visual recognition tasks. Some prompting techniques introduce prior knowledge due to its richness, but when learnable tokens are randomly initialized and disconnected from prior knowledge, they tend to overfit on seen classes and struggle with domain shifts for unseen ones. To address this issue, we propose the InPK model, which infuses class-specific prior knowledge into the learnable tokens during initialization, thus enabling the model to explicitly focus on class-relevant information. Furthermore, to mitigate the weakening of class information by multi-layer encoders, we continuously reinforce the interaction between learnable tokens and prior knowledge across multiple feature levels. This progressive interaction allows the learnable tokens to better capture the fine-grained differences and universal visual concepts within prior knowledge, enabling the model to extract more discriminative and generalized text features. Even for unseen classes, the learned interaction allows the model to capture their common representations and infer their appropriate positions within the existing semantic structure. Moreover, we introduce a learnable text-to-vision projection layer to accommodate the text adjustments, ensuring better alignment of visual-text semantics. Extensive experiments on 11 recognition datasets show that InPK significantly outperforms state-of-the-art methods in multiple zero/few-shot image classification tasks.

cs.CV↗

arXiv2025

Functional BART with Shape Priors: A Bayesian Tree Approach to Constrained Functional Regression

Motivated by the remarkable success of Bayesian additive regression trees (BART) in regression modelling, we propose a novel nonparametric Bayesian method, termed Functional BART (FBART), tailored specifically for function-on-scalar regression. FBART leverages spline-based representations for functional responses coupled with a flexible tree-based partitioning structure, effectively capturing complex and heterogeneous relationships between response curves and scalar predictors. To facilitate efficient posterior inference, we develop a customized Bayesian backfitting algorithm. Additionally, we extend FBART by introducing shape constraints (e.g., monotonicity or convexity) on the response curves, enabling enhanced estimation and prediction when prior shape information is available. The use of shape priors ensures that posterior samples respect the specified functional constraints. Under mild regularity conditions, we establish posterior convergence rates for both FBART and its shape-constrained variant, demonstrating rate adaptivity to unknown smoothness. Extensive simulation studies and analyses of two real datasets illustrate the superior estimation accuracy and predictive performance of our proposed methods compared to existing state-of-the-art alternatives.

stat.ME↗

arXiv2024

LoLDU: Low-Rank Adaptation via Lower-Diag-Upper Decomposition for Parameter-Efficient Fine-Tuning

The rapid growth of model scale has necessitated substantial computational resources for fine-tuning. Existing approach such as Low-Rank Adaptation (LoRA) has sought to address the problem of handling the large updated parameters in full fine-tuning. However, LoRA utilize random initialization and optimization of low-rank matrices to approximate updated weights, which can result in suboptimal convergence and an accuracy gap compared to full fine-tuning. To address these issues, we propose LoLDU, a Parameter-Efficient Fine-Tuning (PEFT) approach that significantly reduces trainable parameters by 2600 times compared to regular PEFT methods while maintaining comparable performance. LoLDU leverages Lower-Diag-Upper Decomposition (LDU) to initialize low-rank matrices for faster convergence and orthogonality. We focus on optimizing the diagonal matrix for scaling transformations. To the best of our knowledge, LoLDU has the fewest parameters among all PEFT approaches. We conducted extensive experiments across 4 instruction-following datasets, 6 natural language understanding (NLU) datasets, 8 image classification datasets, and image generation datasets with multiple model types (LLaMA2, RoBERTa, ViT, and Stable Diffusion), providing a comprehensive and detailed analysis. Our open-source code can be accessed at \href{https://github.com/SKDDJ/LoLDU}{https://github.com/SKDDJ/LoLDU}.

cs.CV↗

arXiv2024

SVFit: Parameter-Efficient Fine-Tuning of Large Pre-Trained Models Using Singular Values

Large pre-trained models (LPMs) have demonstrated exceptional performance in diverse natural language processing and computer vision tasks. However, fully fine-tuning these models poses substantial memory challenges, particularly in resource-constrained environments. Parameter-efficient fine-tuning (PEFT) methods, such as LoRA, mitigate this issue by adjusting only a small subset of parameters. Nevertheless, these methods typically employ random initialization for low-rank matrices, which can lead to inefficiencies in gradient descent and diminished generalizability due to suboptimal starting points. To address these limitations, we propose SVFit, a novel PEFT approach that leverages singular value decomposition (SVD) to initialize low-rank matrices using critical singular values as trainable parameters. Specifically, SVFit performs SVD on the pre-trained weight matrix to obtain the best rank-r approximation matrix, emphasizing the most critical singular values that capture over 99% of the matrix's information. These top-r singular values are then used as trainable parameters to scale the fundamental subspaces of the matrix, facilitating rapid domain adaptation. Extensive experiments across various pre-trained models in natural language understanding, text-to-image generation, and image classification tasks reveal that SVFit outperforms LoRA while requiring 16 times fewer trainable parameters.

cs.LG↗

arXiv2024

Penalized spline estimation of principal components for sparse functional data: rates of convergence

This paper gives a comprehensive treatment of the convergence rates of penalized spline estimators for simultaneously estimating several leading principal component functions, when the functional data is sparsely observed. The penalized spline estimators are defined as the solution of a penalized empirical risk minimization problem, where the loss function belongs to a general class of loss functions motivated by the matrix Bregman divergence, and the penalty term is the integrated squared derivative. The theory reveals that the asymptotic behavior of penalized spline estimators depends on the interesting interplay between several factors, i.e., the smoothness of the unknown functions, the spline degree, the spline knot number, the penalty order, and the penalty parameter. The theory also classifies the asymptotic behavior into seven scenarios and characterizes whether and how the minimax optimal rates of convergence are achievable in each scenario.

math.ST↗

arXiv2024

Channelling Multimodality Through a Unimodalizing Transport: Warp-U Sampler and Stochastic Bridge Sampling Estimator

Monte Carlo integration is a powerful tool for scientific and statistical computation, but faces significant challenges when the integrand is a multi-modal distribution, even when the mode locations are known. This work introduces novel Monte Carlo sampling and integration estimation strategies for the multi-modal context by leveraging a generalized version of the stochastic Warp-U transformation Wang et al. [2022]. We propose two flexible classes of Warp-U transformations, one based on a general location-scale-skew mixture model and a second using neural ordinary differential equations. We develop an efficient sampling strategy called Warp-U sampling, which applies a Warp-U transformation to map a multi-modal density into a uni-modal one, then inverts the transformation with injected stochasticity. In high dimensions, our approach relies on information about the mode locations, but requires minimal tuning and demonstrates better mixing properties than conventional methods with identical mode information. To improve normalizing constant estimation once samples are obtained, we propose a stochastic Warp-U bridge sampling estimator, which we demonstrate has higher asymptotic precision per CPU second compared to the original approach proposed by Wang et al. [2022]. We also establish the ergodicity of our sampling algorithm. The effectiveness and current limitations of our methods are illustrated through simulation studies and an application to exoplanet detection.

stat.ME↗

arXiv2023

Functional PCA with Covariate Dependent Mean and Covariance Structure

Incorporating covariates into functional principal component analysis (PCA) can substantially improve the representation efficiency of the principal components and predictive performance. However, many existing functional PCA methods do not make use of covariates, and those that do often have high computational cost or make overly simplistic assumptions that are violated in practice. In this article, we propose a new framework, called Covariate Dependent Functional Principal Component Analysis (CD-FPCA), in which both the mean and covariance structure depend on covariates. We propose a corresponding estimation algorithm, which makes use of spline basis representations and roughness penalties, and is substantially more computationally efficient than competing approaches of adequate estimation and prediction accuracy. A key aspect of our work is our novel approach for modeling the covariance function and ensuring that it is symmetric positive semi-definite. We demonstrate the advantages of our methodology through a simulation study and an astronomical data analysis.

stat.ME↗

arXiv2023

A Unified Analysis of Multi-task Functional Linear Regression Models with Manifold Constraint and Composite Quadratic Penalty

This work studies the multi-task functional linear regression models where both the covariates and the unknown regression coefficients (called slope functions) are curves. For slope function estimation, we employ penalized splines to balance bias, variance, and computational complexity. The power of multi-task learning is brought in by imposing additional structures over the slope functions. We propose a general model with double regularization over the spline coefficient matrix: i) a matrix manifold constraint, and ii) a composite penalty as a summation of quadratic terms. Many multi-task learning approaches can be treated as special cases of this proposed model, such as a reduced-rank model and a graph Laplacian regularized model. We show the composite penalty induces a specific norm, which helps to quantify the manifold curvature and determine the corresponding proper subset in the manifold tangent space. The complexity of tangent space subset is then bridged to the complexity of geodesic neighbor via generic chaining. A unified convergence upper bound is obtained and specifically applied to the reduced-rank model and the graph Laplacian regularized model. The phase transition behaviors for the estimators are examined as we vary the configurations of model parameters.

math.ST↗

arXiv2023

Bayesian Nonlinear Tensor Regression with Functional Fused Elastic Net Prior

Tensor regression methods have been widely used to predict a scalar response from covariates in the form of a multiway array. In many applications, the regions of tensor covariates used for prediction are often spatially connected with unknown shapes and discontinuous jumps on the boundaries. Moreover, the relationship between the response and the tensor covariates can be nonlinear. In this article, we develop a nonlinear Bayesian tensor additive regression model to accommodate such spatial structure. A functional fused elastic net prior is proposed over the additive component functions to comprehensively model the nonlinearity and spatial smoothness, detect the discontinuous jumps, and simultaneously identify the active regions. The great flexibility and interpretability of the proposed method against the alternatives are demonstrated by a simulation study and an analysis on facial feature data.

stat.ME↗

arXiv2022

Spline Estimation of Functional Principal Components via Manifold Conjugate Gradient Algorithm

Functional principal component analysis has become the most important dimension reduction technique in functional data analysis. Based on B-spline approximation, functional principal components (FPCs) can be efficiently estimated by the expectation-maximization (EM) and the geometric restricted maximum likelihood (REML) algorithms under the strong assumption of Gaussianity on the principal component scores and observational errors. When computing the solution, the EM algorithm does not exploit the underlying geometric manifold structure, while the performance of REML is known to be unstable. In this article, we propose a conjugate gradient algorithm over the product manifold to estimate FPCs. This algorithm exploits the manifold geometry structure of the overall parameter space, thus improving its search efficiency and estimation accuracy. In addition, a distribution-free interpretation of the loss function is provided from the viewpoint of matrix Bregman divergence, which explains why the proposed method works well under general distribution settings. We also show that a roughness penalization can be easily incorporated into our algorithm with a potentially better fit. The appealing numerical performance of the proposed method is demonstrated by simulation studies and the analysis of a Type Ia supernova light curve dataset.

stat.ME↗

arXiv2022

Functional Principal Subspace Sampling for Large Scale Functional Data Analysis

Functional data analysis (FDA) methods have computational and theoretical appeals for some high dimensional data, but lack the scalability to modern large sample datasets. To tackle the challenge, we develop randomized algorithms for two important FDA methods: functional principal component analysis (FPCA) and functional linear regression (FLR) with scalar response. The two methods are connected as they both rely on the accurate estimation of functional principal subspace. The proposed algorithms draw subsamples from the large dataset at hand and apply FPCA or FLR over the subsamples to reduce the computational cost. To effectively preserve subspace information in the subsamples, we propose a functional principal subspace sampling probability, which removes the eigenvalue scale effect inside the functional principal subspace and properly weights the residual. Based on the operator perturbation analysis, we show the proposed probability has precise control over the first order error of the subspace projection operator and can be interpreted as an importance sampling for functional subspace estimation. Moreover, concentration bounds for the proposed algorithms are established to reflect the low intrinsic dimension nature of functional data in an infinite dimensional space. The effectiveness of the proposed algorithms is demonstrated upon synthetic and real datasets.

stat.CO↗

arXiv2022

Validating Synthetic Galaxy Catalogs for Dark Energy Science in the LSST Era

Large simulation efforts are required to provide synthetic galaxy catalogs for ongoing and upcoming cosmology surveys. These extragalactic catalogs are being used for many diverse purposes covering a wide range of scientific topics. In order to be useful, they must offer realistically complex information about the galaxies they contain. Hence, it is critical to implement a rigorous validation procedure that ensures that the simulated galaxy properties faithfully capture observations and delivers an assessment of the level of realism attained by the catalog. We present here a suite of validation tests that have been developed by the Rubin Observatory Legacy Survey of Space and Time (LSST) Dark Energy Science Collaboration (DESC). We discuss how the inclusion of each test is driven by the scientific targets for static ground-based dark energy science and by the availability of suitable validation data. The validation criteria that are used to assess the performance of a catalog are flexible and depend on the science goals. We illustrate the utility of this suite by showing examples for the validation of cosmoDC2, the extragalactic catalog recently released for the LSST DESC second Data Challenge.

astro-ph.CO↗

arXiv2021

A Reproducing Kernel Hilbert Space Approach to Functional Calibration of Computer Models

This paper develops a frequentist solution to the functional calibration problem, where the value of a calibration parameter in a computer model is allowed to vary with the value of control variables in the physical system. The need of functional calibration is motivated by engineering applications where using a constant calibration parameter results in a significant mismatch between outputs from the computer model and the physical experiment. Reproducing kernel Hilbert spaces (RKHS) are used to model the optimal calibration function, defined as the functional relationship between the calibration parameter and control variables that gives the best prediction. This optimal calibration function is estimated through penalized least squares with an RKHS-norm penalty and using physical data. An uncertainty quantification procedure is also developed for such estimates. Theoretical guarantees of the proposed method are provided in terms of prediction consistency and consistency of estimating the optimal calibration function. The proposed method is tested using both real and synthetic data and exhibits more robust performance in prediction and uncertainty quantification than the existing parametric functional calibration method and a state-of-art Bayesian method.

stat.ME↗

arXiv2021

A novel stellar spectrum denoising method based on deep Bayesian modeling

Spectrum denoising is an important procedure for large-scale spectroscopical surveys. This work proposes a novel stellar spectrum denoising method based on deep Bayesian modeling. The construction of our model includes a prior distribution for each stellar subclass, a spectrum generator and a flow-based noise model. Our method takes into account the noise correlation structure, and it is not susceptible to strong sky emission lines and cosmic rays. Moreover, it is able to naturally handle spectra with missing flux values without ad-hoc imputation. The proposed method is evaluated on real stellar spectra from the Sloan Digital Sky Survey (SDSS) with a comprehensive list of common stellar subclasses and compared to the standard denoising auto-encoder. Our denoising method demonstrates superior performance to the standard denoising auto-encoder, in respect of denoising quality and missing flux imputation. It may be potentially helpful in improving the accuracy of the classification and physical parameter measurement of stars when applying our method during data preprocessing.

astro-ph.IM↗