This paper presents Bayesian-inspired Space-Time Superpixels (BIST): a fast, state-of-the-art method to compute space-time superpixels. BIST is a novel extension of a single-image Bayesian method named BASS, and it is inspired by hill-climbing to a local mode of a Dirichlet-Process Gaussian Mixture Model (DP-GMM). The method is only Bayesian-inspired, rather than actually Bayesian, because it includes heuristic modifications to the theoretically correct sampler. Similar to existing methods, BIST can adapt the number of superpixels to an individual frame using split-merge steps. A key novelty is a new temporal coherence term in the split step, which reduces the chance of splitting propagated superpixels. This term enforces temporal coherence in propagated regions, but allows for unconstrained adaptation in disoccluded regions. A hyperparameter determines the strength of this new term, which does not require special tuning to return consistent results across a dataset of videos. The wall-clock runtime of BIST is over twice as fast as BASS and over 30 times faster than the next fastest space-time superpixel method with open-source code.
论文检索
输入标题、作者或关键词，从 901 篇学术成果中精准定位
The widespread availability of off-the-shelf machine learning models poses a challenge: which model, of the many available candidates, should be chosen for a given data analysis task? This question of model selection is traditionally answered by collecting and annotating a validation dataset---a costly and time-intensive process. We propose a method for active model selection, using predictions from candidate models to prioritize the labeling of test data points that efficiently differentiate the best candidate. Our method, CODA, performs consensus-driven active model selection by modeling relationships between classifiers, categories, and data points within a probabilistic framework. The framework uses the consensus and disagreement between models in the candidate pool to guide the label acquisition process, and Bayesian inference to update beliefs about which model is best as more information is collected. We validate our approach by curating a collection of 26 benchmark tasks capturing a range of model selection scenarios.CODA outperforms existing methods for active model selection significantly, reducing the annotation effort required to discover the best model by upwards of 70% compared to the previous state-of-the-art. We will make our code and data public. Code and data are available at https://github.com/justinkay/coda.
Image restoration aims to recover high-quality images from degraded observations. When the degradation process is known, the recovery problem can be formulated as an inverse problem, and in a Bayesian context, the goal is to sample a clean reconstruction given the degraded observation. Recently, modern pretrained diffusion models have been used for image restoration by modifying their sampling procedure to account for the degradation process. However, these methods often rely on certain approximations that can lead to significant errors and compromised sample quality. In this paper, we propose a simple modification to existing diffusion-based restoration methods that exploits the frequency structure of the reverse diffusion process. Specifically, our approach, denoted as Frequency Guided Posterior Sampling (FGPS), introduces a time-varying low-pass filter in the frequency domain of the measurements, progressively incorporating higher frequencies during the restoration process. We provide the first rigorous analysis of the approximation error of FGPS for linear inverse problems under distributional assumptions on the space of natural images, demonstrating cases where previous works can fail dramatically. On real-world data, we develop an adaptive curriculum for our method's frequency schedule based on the underlying data distribution. FGPS significantly improves performance on challenging image restoration tasks including motion deblurring and image dehazing.
Recent advances in vision-language models (VLMs) have significantly improved performance in embodied tasks such as goal decomposition and visual comprehension. However, providing accurate rewards for robotic manipulation without fine-tuning VLMs remains challenging due to the absence of domain-specific robotic knowledge in pre-trained datasets and high computational costs that hinder real-time applicability. To address this, we propose T2-VLM, a novel training-free, temporally consistent framework that generates accurate rewards through tracking the status changes in VLM-derived subgoals. Specifically, our method first queries the VLM to establish spatially aware subgoals and an initial completion estimate before each round of interaction. We then employ a Bayesian tracking algorithm to update the goal completion status dynamically, using subgoal hidden states to generate structured rewards for reinforcement learning (RL) agents. This approach enhances long-horizon decision-making and improves failure recovery capabilities with RL. Extensive experiments indicate that T2-VLM achieves state-of-the-art performance in two robot manipulation benchmarks, demonstrating superior reward accuracy with reduced computation consumption. We believe our approach not only advances reward generation techniques but also contributes to the broader field of embodied AI. Project website: https://t2-vlm.github.io/.
ViT-EnsembleAttack: Augmenting Ensemble Models for Stronger Adversarial Transferability in Vision Transformers
PDF ↗Ensemble-based attacks have been proven to be effective in enhancing adversarial transferability by aggregating the output of models with various architectures. However, existing research primarily focuses on refining ensemble weights or optimizing the ensemble path, overlooking the exploration of ensemble models to enhance the transferability of adversarial attacks. To address this gap, we propose applying adversarial augmentation to the surrogate models, aiming to boost overall generalization of ensemble models and reduce the risk of adversarial overfitting. Meanwhile, observing that ensemble Vision Transformers (ViTs) gain less attention, we propose ViT-EnsembleAttack based on the idea of model adversarial augmentation, the first ensemble-based attack method tailored for ViTs to the best of our knowledge. Our approach generates augmented models for each surrogate ViT using three strategies: Multi-head dropping, Attention score scaling, and MLP feature mixing, with the associated parameters optimized by Bayesian optimization. These adversarially augmented models are ensembled to generate adversarial examples. Furthermore, we introduce Automatic Reweighting and Step Size Enlargement modules to boost transferability. Extensive experiments demonstrate that ViT-EnsembleAttack significantly enhances the adversarial transferability of ensemble-based attacks on ViTs, outperforming existing methods by a substantial margin. Code is available at https://github.com/Trustworthy-AI-Group/TransferAttack.
Multi-Modal Multi-Task Unified Embedding Model (M3T-UEM): A Task-Adaptive Representation Learning Framework
PDF ↗We present Multi-Modal Multi-Task Unified Embedding Model (M3T-UEM), a framework that advances vision-language matching and retrieval by leveraging a large language model (LLM) backbone. While concurrent LLM-based approaches like VLM2VEC, MM-Embed, NV-Embed, and MM-GEM have demonstrated impressive capabilities in multi-modal and multi-task scenarios, our work introduces novel mechanisms for task-adaptive learning and embedding extraction that further enhance the potential of LLM-based retrieval systems. Our key technical contribution lies in the development of a task-aware contrastive learning framework with an automated Bayesian weighing mechanism. This approach provides a principled way to balance multiple tasks during training, departing from conventional contrastive learning strategies. We further enhance the framework through a multiple-token summarization strategy and an auxiliary language modeling objective, which together significantly improve retrieval performance.Comprehensive experiments on M-BEIR and ICinW benchmarks demonstrate M3T-UEM's effectiveness, showing competitive or superior performance compared to both traditional encoder-based methods and recent LLM-based approaches. Furthermore, we demonstrate particular strengths in handling compositional conceptual changes and multilingual scenarios owing to the incorporation of an LLM backbone where the method drastically outperforms CLIP in zero-shot settings, often by orders of magnitude.
Multimodal large language models (MLLMs) are increasingly used to evaluate text-to-image (TTI) generation systems, providing automated judgments based on visual and textual context. However, these "judge" models often suffer from biases, overconfidence, and inconsistent performance across diverse image domains. While prompt ensembling has shown promise for mitigating these issues in unimodal, text-only settings, our experiments reveal that standard ensembling methods fail to generalize effectively for TTI tasks. To address these limitations, we propose a new multimodal-aware method called **M**ultimodal **M**ixture-of-**B**ayesian Prompt Ensembles (MMB). Our approach uses a Bayesian prompt ensemble approach augmented by image clustering, allowing the judge to dynamically assign prompt weights based on the visual characteristics of each sample. We show that MMB improves accuracy in pairwise preference judgments and greatly enhances calibration, making it easier to gauge the judge's true uncertainty. In evaluations on two TTI benchmarks, HPSv2 and MJBench, MMB outperforms existing baselines in alignment with human annotations and calibration across varied image content. Our findings highlight the importance of multimodal-specific strategies for judge calibration and suggest a promising path forward for reliable large-scale TTI evaluation.
ContraGS: Codebook-Condensed and Trainable Gaussian Splatting for Fast, Memory-Efficient Reconstruction
PDF ↗3D Gaussian Splatting (3DGS) is a state-of-art technique to model real-world scenes with high quality and real-time rendering.Typically, a higher quality representation can be achieved by using a large number of 3D Gaussians. However, using large 3D Gaussian counts significantly increases the GPU device memory for storing model parameters. A large model thus requires powerful GPUs with high memory capacities for training and has slower training/rendering latencies due to the inefficiencies of memory access and data movement. In this work, we introduce ContraGS, a method to enable training directly on compressed 3DGS representations without reducing the Gaussian Counts, and thus with a little loss in model quality. ContraGS leverages codebooks to compactly store a set of Gaussian parameter vectors throughout the training process, thereby significantly reducing memory consumption. While codebooks have been demonstrated to be highly effective at compressing fully trained 3DGS models, directly training using codebook representations is an unsolved challenge. ContraGS solves the problem of learning non-differentiable parameters in codebook-compressed representations by posing parameter estimation as a Bayesian inference problem. To this end, ContraGS provides a framework that effectively uses MCMC sampling to sample over a posterior distribution of these compressed representations. With ContraGS, we demonstrate that ContraGS significantly reduces the peak memory during training (on average 3.49X) and accelerated training and rendering 1.36Xand 1.88X on average, respectively), while retraining close to state-of-art quality.
Diffusion models have shown remarkable promise for image restoration by leveraging powerful priors. Prominent methods typically frame the restoration problem within a Bayesian inference framework, which iteratively combines a denoising step with a likelihood guidance step. However, the interactions between these two components in the generation process remain underexplored. In this paper, we analyze the underlying gradient dynamics of these components and identify significant instabilities. Specifically, we demonstrate conflicts between the prior and likelihood gradient directions, alongside temporal fluctuations in the likelihood gradient itself. We show that these instabilities disrupt the generative process and compromise restoration performance. To address these issues, we propose Stabilized Progressive Gradient Diffusion (SPGD), a novel gradient management technique. SPGD integrates two synergistic components: (1) a progressive likelihood warm-up strategy to mitigate gradient conflicts; and (2) adaptive directional momentum (ADM) smoothing to reduce fluctuations in the likelihood gradient. Extensive experiments across diverse restoration tasks demonstrate that SPGD significantly enhances generation stability, leading to state-of-the-art performance in quantitative metrics and visually superior results. Code is available at https://github.com/74587887/SPGD.
The field of cross-modal retrieval aims to construct a shared representation space for samples from multiple modalities, typically within the vision and language domains. Deep hashing, with its high computational efficiency and low storage costs, has emerged as a central focus in this field and has garnered significant attention in recent research. However, current hash retrieval, concentrating on deterministic methods, struggles to effectively capture semantically ambiguous correspondences between cross-modal samples, where heterogeneous data have complex-semantic many-to-many relationships in the latent space. To address this limitation, we propose a novel Deep Probabilistic Binary Embedding (DPBE) framework, designed to generate discriminative, modality-invariant hash codes that facilitate accurate and reliable cross-modal retrieval. In contrast to contemporary probabilistic methods, we focus on optimizing hash networks to learn more accurate binary embeddings by using the learning mode of probabilistic embeddings. We introduce the first Bayesian encoder for hash learning, which employs Laplace Approximation to model a distribution over network weights. Extensive experimental results demonstrate that our approach not only outperforms deterministic methods in retrieval performance but also provides uncertainty estimates, enhancing the interpretability of the embeddings. The corresponding code is available at https://github.com/QinLab-WFU/DPBE.
Large Vision-Language Models (LVLMs) usually generate texts which satisfy context coherence but don't match the visual input. Such a hallucination issue hinders LVLMs' applicability in the real world. The key to solving hallucination in LVLM is to make the text generation rely more on the visual content. Most previous works choose to enhance/adjust the features/output of a specific modality (i.e., visual or textual) to alleviate hallucinations in LVLM, which do not explicitly or systematically enhance the visual reliance. In this paper, we comprehensively investigate the factors that may degenerate the visual reliance in text generation of LVLM from a Bayesian perspective. We propose to mitigate hallucination in LVLM from three aspects. Firstly, we observe that not all visual tokens are informative in generating meaningful texts. We propose to evaluate and remove redundant visual tokens to avoid their disturbance. Secondly, LVLM may encode inappropriate prior information, making it lean toward generating unexpected words. We propose a simple, yet effective way to rectify the prior from a Bayesian perspective. Thirdly, we observe that starting from certain steps, the posterior of next-token prediction conditioned on visual tokens may collapse to a prior distribution which does not depend on any informative visual tokens at all. Thus, we propose to stop further text generation to avoid hallucination. Extensive experiments on three benchmarks, including POPE, CHAIR, and MME, demonstrate that our method can consistently mitigate the hallucination issue of LVLM and performs favorably against previous state-of-the-arts. Codes are available at https://github.com/NeilHnxTcc/EVRB.
Medical image restoration tasks aim to recover high-quality images from degraded observations, exhibiting emergent desires in many clinical scenarios, such as low-dose CT image denoising, MRI super-resolution, and MRI artifact removal. Despite the success achieved by existing deep learning-based restoration methods with sophisticated modules, they struggle with rendering computationally-efficient reconstruction results. Moreover, they usually ignore the reliability of the restoration results, which is much more urgent in medical systems. To alleviate these issues, we present LRformer, a Lightweight Transformer-based method via Reliability-guided learning in the frequency domain. Specifically, inspired by the uncertainty quantification in Bayesian neural networks (BNNs), we develop a Reliable Lesion-Semantic Prior Producer (RLPP). RLPP leverages Monte Carlo (MC) estimators with stochastic sampling operations to generate sufficiently-reliable priors by performing multiple inferences on the foundational medical image segmentation model, MedSAM. Additionally, instead of directly incorporating the priors in the spatial domain, we decompose the cross-attention (CA) mechanism into real symmetric and imaginary anti-symmetric parts via fast Fourier transform (FFT), resulting in the design of the Guided Frequency Cross-Attention (GFCA) solver. By leveraging the conjugated symmetric property of FFT, GFCA reduces the computational complexity of naive CA by nearly half. Extensive experimental results in various tasks demonstrate the superiority of the proposed LRformer in both effectiveness and efficiency.
Unleashing the Potential of Multi-Channel Fusion in Retrieval for Personalized Recommendations
PDF ↗Recommender systems (RS) are pivotal in managing information overload in modern digital services. A key challenge in RS is efficiently processing vast item pools to deliver highly personalized recommendations under strict latency constraints. Multi-stage cascade ranking addresses this by employing computationally efficient retrieval methods to cover diverse user interests, followed by more precise ranking models to refine the results. In the retrieval stage, multi-channel retrieval is often used to generate distinct item subsets from different candidate generators, leveraging the complementary strengths of these methods to maximize coverage. However, forwarding all retrieved items overwhelms downstream rankers, necessitating truncation. Despite advancements in individual retrieval methods, multi-channel fusion, the process of efficiently merging multi-channel retrieval results, remains underexplored. We are the first to identify and systematically investigate multi-channel fusion in the retrieval stage. Current industry practices often rely on heuristic approaches and manual designs, which often lead to suboptimal performance. Moreover, traditional gradient-based methods like SGD are unsuitable for this task due to the non-differentiable nature of the selection process. In this paper, we explore advanced channel fusion strategies by assigning systematically optimized weights to each channel. We utilize black-box optimization techniques, including the Cross Entropy Method and Bayesian Optimization for global weight optimization, alongside policy gradient-based approaches for personalized merging. Our methods enhance both personalization and flexibility, achieving significant performance improvements across multiple datasets and yielding substantial gains in real-world deployments, offering a scalable solution for optimizing multi-channel fusion in retrieval.
Hyperparameter optimization is crucial for maximizing machine learning model performance, yet most existing algorithms are designed for batch or offline scenarios and assume static data distributions. Such assumptions fall short in data stream settings, where models must adapt to evolving inputs in real time. To address these limitations, we propose the Bayesian Stream Tuner (BST), a novel framework for online hyperparameter optimization in non-stationary data streams. BST maintains a dynamic set of candidate hyperparameter configurations and periodically refines them using an incremental Bayesian model, which estimates configuration performance based on recent data statistics and hyperparameter values. This systematic exploration and refinement strategy allows BST to detect and respond to concept drift by resetting its adaptation mechanisms whenever necessary, ensuring strong performance under changing distributions. Our theoretical analysis establishes sublinear regret bounds for BST in dynamic environments, and extensive experiments on classification and regression tasks demonstrate that BST consistently outperforms state-of-the-art online hyperparameter optimization methods in both predictive accuracy and adaptability, making it a powerful solution for real-time hyperparameter tuning in evolving data streams.
Selecting the optimal combination of a machine learning (ML) algorithm and its hyper-parameters is crucial for the development of high-performance ML systems. However, since the combination of ML algorithms and hyper-parameters is enormous, the exhaustive validation requires a significant amount of time. Many existing studies use Bayesian optimization (BO) for accelerating the search. On the other hand, a significant difficulty is that, in general, there exists a different hyper-parameter space for each one of candidate ML algorithms. BO-based approaches typically build a surrogate model independently for each hyper-parameter space, by which sufficient observations are required for all candidate ML algorithms. In this study, our proposed method embeds different hyper-parameter spaces into a shared latent space, in which a surrogate multi-task model for BO is estimated. This approach can share information of observations from different ML algorithms by which efficient optimization is expected with a smaller number of total observations. We further propose the pre-training of the latent space embedding with an adversarial regularization, and a ranking model for selecting an effective pre-trained embedding for a given target dataset. Our empirical study demonstrates effectiveness of the proposed method through datasets from OpenML.
Federated Learning (FL) enables multiple clients to collaboratively develop a global model while maintaining data privacy. However, online FL deployment faces challenges due to distribution shifts and evolving test samples. Personalized Federated Learning (PFL) tailors the global model to individual client distributions, but struggles with Out-Of-Distribution (OOD) samples during testing, leading to performance degradation. In real-world scenarios, balancing personalization and generalization during online testing is crucial and existing methods primarily focus on training-phase generalization. To address the test-time trade-off, we introduce a new scenario: Test-time Generalization for Internal and External Distributions in Federated Learning (TGFL), which evaluates adaptability under Internal Distribution (IND) and External Distribution (EXD). We propose BTFL, a Bayesian-based test-time generalization method for TGFL, which balances generalization and personalization at the sample level during testing. BTFL employs a two-head architecture to store local and global knowledge, interpolating predictions via a dual-Bayesian framework that considers both historical test data and current sample characteristics with theoretical guarantee and faster speed. Our experiments demonstrate that BTFL achieves improved performance across various datasets and models with less time cost. The source codes are made publicly available at https://github.com/ZhouYuCS/BTFL.
This work addresses a key limitation in current federated learning approaches, which predominantly focus on homogeneous tasks, neglecting the task diversity on local devices. We propose a principled integration of multi-task learning using multi-output Gaussian processes (MOGP) at the local level and federated learning at the global level. MOGP handles correlated classification and regression tasks, offering a Bayesian non-parametric approach that naturally quantifies uncertainty. The central server aggregates the posteriors from local devices, updating a global MOGP prior redistributed for training local models until convergence. Challenges in performing posterior inference on local devices are addressed through the Polya-Gamma augmentation technique and mean-field variational inference, enhancing computational efficiency and convergence rate. Experimental results on both synthetic and real data demonstrate superior predictive performance, OOD detection, uncertainty calibration and convergence rate, highlighting the method's potential in diverse applications. Our code is publicly available at https://github.com/JunliangLv/task_diversity_BFL.
Clustering high-dimensional directional data (i.e., L2 normalized vectors) presents significant challenges due to the intricate spherical representations of latent embeddings and the limitations of classical (non-deep) clustering techniques. Moreover, dynamically inferring the number of clusters remains a fundamental issue in existing deep clustering methods, especially those involving complex model-selection criteria. This paper addresses these challenges by introducing a novel deep nonparametric clustering framework that employs hyperspherical latent embeddings within a Variational Autoencoder architecture, enhanced by an infinite Von Mises-Fisher Mixture Model as a dynamic prior. This approach enables automatic adaptation of cluster numbers during training, eliminating the need for predefined clusters and traditional model selection processes. Our scalable architecture effectively integrates In-vMFMM with hyperspherical embeddings to tackle the complexities of directional data. Utilizing a joint training strategy, our method alternates between updating neural network parameters and adjusting mixture model priors via nonparametric variational Bayes. Empirical evaluations on benchmark datasets, including complex ImageNet-50, demonstrate that our approach significantly outperforms state-of-the-art deep nonparametric clustering methods. It also robustly estimates the number of clusters, showcasing its effectiveness and versatility in handling high-dimensional directional data.
Conditional Independent Test in the Presence of Measurement Error with Causal Structure Learning
PDF ↗Testing conditional independence is a critical task, particularly in causal discovery and learning in Bayesian networks. However, in many real-world scenarios, variables are often measured with errors, such as those introduced by insufficient measurement accuracy, complicating the testing process. This paper focuses on testing conditional independence in the linear non-Gaussian measurement error model, under the condition that measurement error noise follows a Gaussian distribution. By leveraging high-order cumulants, we derive rank constraints on the cumulant matrix and establish their role in effectively assessing conditional independence, even in the presence of measurement errors. Based on these theoretical results, we leverage the rank constraints of the cumulant matrix as a tool for conditional independence testing and incorporate it into the PC algorithm, resulting in the PC-ME algorithm — a method designed to learn causal structures from observed data while accounting for measurement errors. Experimental results demonstrate that the proposed method outperforms existing approaches, particularly in cases other methods encounter difficulties.
Causal structure learning (CSL) plays a pivotal role in causality and is often formulated as an optimization problem within score-and-search methods. Under the assumption of an infinite dataset and a predefined distribution, several well-established and consistent score functions have been shown to be both optimal and reliable for identifying ground-truth causal graphs. However, in practice, these idealized assumptions are often infeasible, which can result in CSL algorithms learning suboptimal structures. In this paper, we introduce L-SFE, a framework designed to automatically discover effective score functions by exploring the "score function space". L-SFE addresses this task from a bi-level optimization perspective. First, it leverages a Large Language Model (LLM) to interpret the characteristics of score functions and generate the corresponding code implementations. Next, L-SFE employs evolutionary algorithms along with carefully designed operators, to search for solutions with higher fitness. Additionally, we take the BIC as example and prove the consistency of the generated score functions. Experimental evaluations, conducted on discrete, continuous, and real datasets, demonstrate the high stability, generality and effectiveness of L-SFE.