Adapting Large Multimodal Models (LMMs) to real-world scenarios poses the dual challenges of learning from sequential data streams while handling frequent modality incompleteness, a task known as Continual Missing Modality Learning (CMML). However, existing works on CMML have predominantly relied on prompt tuning, a technique that struggles with this task due to cross-task interference between its learnable prompts in their shared embedding space. A naive application of Low-Rank Adaptation (LoRA) with modality-shared module will also suffer modality interference from competing gradients. To this end, we propose DeLo, the first framework to leverage a novel dual-decomposed low-rank expert architecture for CMML. Specifically, this architecture resolves modality interference through decomposed LoRA expert, dynamically composing LoRA update matrix with rank-one factors from disentangled modality-specific factor pools. Embedded within a task-partitioned framework that structurally prevents catastrophic forgetting, this expert system is supported by two key mechanisms: a Cross-Modal Guided Routing strategy to handle incomplete data and a Task-Key Memory for efficient, task-agnostic inference. Extensive experiments on established CMML benchmarks demonstrate that our method significantly outperforms state-of-the-art approaches. This highlights the value of a principled, architecturally-aware LoRA design for real-world multimodal challenges.
论文检索
输入标题、作者或关键词，从 11,272 篇学术成果中精准定位
Adversarial attacks pose a major challenge to distributed learning systems, prompting the development of numerous robust learning methods. However, most existing approaches suffer from the curse of dimensionality, i.e. the error increases with the number of model parameters. In this paper, we make a progress towards high dimensional problems, under arbitrary number of Byzantine attackers. The cornerstone of our design is a direct high dimensional semi-verified mean estimation method. The idea is to identify a subspace with large variance. The components of the mean value perpendicular to this subspace are estimated using corrupted gradient vectors uploaded from worker machines, while the components within this subspace are estimated using auxiliary dataset. As a result, a combination of large corrupted dataset and small clean dataset yields significantly better performance than using them separately. We then apply this method as the aggregator for distributed learning problems. The theoretical analysis shows that compared with existing solutions, our method gets rid of sqrt{d} dependence on the dimensionality, and achieves minimax optimal statistical rates. Numerical results validate our theory as well as the effectiveness of the proposed method.
Macro placement is a crucial subproblem of chip design, focusing on determining the locations of numerous macros while minimizing multiple metrics. In recent years, reinforcement learning (RL) has gained traction as a favorable technique to improve placement performance. However, existing RL-based placers ignore the orientation of macros, resulting in the state space constrained to two-dimensional discrete coordinates and greatly restricting the exploration opportunities. To address this issue, we propose a novel macro placement method, RSPlace, which guides the bidirectional expansion of the global search tree to offer the RL agent more exploration opportunities, incorporating rotation into the RL-based macro placement solution for the first time. RSPlace intelligently determines the optimal rotation angle to maximize placement benefits by leveraging rotation sensing and placement perturbations. Extensive experiments demonstrate that taking the macro orientation into account substantially broadens the feasible locations and effectively reduces the half-perimeter wirelength (HPWL), thus ensuring that our approach significantly improves the optimization effect compared to the state-of-the-art method.
Conformal Prediction (CP) is a popular method for uncertainty quantification that converts a pretrained model's point prediction into a prediction set, with the set size reflecting the model's confidence. Although existing CP methods are guaranteed to achieve marginal coverage, they often exhibit imbalanced coverage across classes under long-tailed label distributions, tending to over cover the head classes at the expense of under covering the remaining tail classes. This under coverage is particularly concerning, as it undermines the reliability of the prediction sets for minority classes, even with coverage ensured on average. In this paper, we propose the Tail-Aware Conformal Prediction (TACP) method to mitigate the under coverage of the tail classes by utilizing the long-tailed structure and narrowing the head-tail coverage gap. Theoretical analysis shows that it consistently achieves a smaller head-tail coverage gap than standard methods. To further improve coverage balance across all classes, we introduce an extension of TACP: soft TACP (sTACP) via a reweighting mechanism. The proposed framework can be combined with various non-conformity scores, and experiments on multiple long-tailed benchmark datasets demonstrate the effectiveness of our methods.
One-shot federated learning (OSFL) reduces the communication cost and privacy risks of iterative federated learning by constructing a global model with a single round of communication. However, most existing methods struggle to achieve robust performance on real-world domains such as medical imaging, or are inefficient when handling non-IID (Independent and Identically Distributed) data. To address these limitations, we introduce FALCON, a novel framework that enhances the effectiveness of OSFL over non-IID image data. The core idea of FALCON is to leverage the feature-aware hierarchical token sequences generation and knowledge distillation into OSFL. First, each client leverages a pretrained visual encoder with hierarchical scale encoding to compress images into hierarchical token sequences, which capture multi-scale semantics. Second, a multi-scale autoregressive transformer generator is used to model the distribution of these token sequences and generate the synthetic sequences. Third, clients upload the synthetic sequences along with the local classifier trained on the real token sequences to the server. Finally, the server incorporates knowledge distillation into global training to reduce reliance on precise distribution modeling. Experiments on medical and natural image datasets validate the effectiveness of FALCON in diverse non-IID scenarios, outperforming the best OSFL baselines by 9.58\% in average accuracy.
Views Attention Fusion of Granular-ball Fuzzy Representations Split for Improved Multi-view Clustering
PDF ↗Multi-View Clustering (MVC) is a pivotal multi-view learning paradigm widely adopted across various fields. Despite recent advances, existing methods primarily focus on enhancing the performance of fused multi-view representation, often neglecting the issue of Representation Degradation (RD) arising from discrepancies in the intrinsic quality of different views. To address the limitations, we propose a novel Granular-ball Fuzzy Split and Attention Fusion (GFSAF) learning, which leverages the nature of granular-ball to extract mutual and complementary representation separately. Meanwhile, the proposed method introduces an attention variant for fused representations to mitigate the RD issue. GFSAF mainly consists of two training stages: Split-Extract Stage and Views-Fusion Stage. Specifically, we design a novel Granular-ball Fuzzy Contrastive Learning to extract mutual representation, and introduce Noise Stripping Loss to reduce the influence of noise for complementary representation. Then, a novel multi-head Cross Views Attention is proposed to employ attention mechanism from multi-view perspectives for comprehensive fused representations. Experimental results on eight databases demonstrate that our GFSAF achieves superior performance compared to several state-of-the-art MVC methods.
SVGL: Scale-Variable Graph Learning in Model Space for Multivariate Time Series Classification
PDF ↗Multivariate time series classification (MTSC) has broad applications in numerous domains. Existing MTSC methods typically focus on either temporal dynamics or variable interactions of the data, often overlooking cross-scale couplings among different variables. To bridge this gap, we propose Scale-Variable Graph Learning (SVGL), a novel framework that effectively captures data-inherent scale-variable interactions for MTSC. SVGL begins with spectral analysis to adaptively identify key periodic scales for each variable. A period-aware reservoir computing network is then incorporated to fit the variable at these scales, encoding the sequential and periodic dynamics into multi-scale dynamic representations. Subsequently, we construct a scale-variable graph to model interactions of the encoded temporal dynamics, where nodes represent scale-variable pairs and edges denote their correlations. After sparsely initializing the graph via nearest neighbors, a parallel graph learning architecture is integrated in SVGL, combining global graph convolutional and sample-specific graph attention to aggregate effective features for classification. Extensive experiments on 30 UEA datasets demonstrate that SVGL outperforms state-of-the-art baselines in accuracy and maintains low training overhead.
Energy-guided Dual Domain-invariant Prompting Framework with Fourier Regularization for Generalized Few-Shot Medical Segmentation
PDF ↗Precise segmentation of organ and tissue lesions is essential for clinical diagnosis and treatment. Despite the progress of deep learning and foundation segmentation models, their domain generalization capability remains limited particularly when dealing with cross-domain scenarios or unseen data, leading to significant performance degradation. Current medical SAM-based generalization methods face two primary challenges: First, existing prompt-tuning strategies inadequately capture key domain-invariant features; Second, the reliance on fully labeled source domain data is unrealistic in clinical practice. To address these challenges, we propose a novel Dual domain-Invariant Prompt Optimization (DIPO) enhanced by energy-guided augmentation and frequency consistency regularization for few-shot medical image segmentation generalization. Our approach introduces a multi-band momentum enhancement strategy to dynamically augment source data by leveraging diverse frequency bands of the Fourier amplitude spectrum. Furthermore, we integrate multiscale geometric representation-based non-subsampled shearlet transform and text prompts to strengthen the extraction of shape- and texture-related domain-invariant features. Finally, we employ frequency consistency regularization to refine model robustness using predictions from unlabeled data. Experimental results in prostate and fundus datasets demonstrate that our method significantly outperforms current state-of-the-art methods.
Graph Neural Networks (GNNs) have demonstrated strong performance across various tasks by leveraging the structural information inherent in graph-structured data. To address the challenge of edge heterophily, where connected nodes may have dissimilar labels or features, two main families of GNNs have emerged: Mixture-of-Experts (MoE) based spatial GNNs and frequency filtering based spectral GNNs. While MoE-based spatial GNNs intuitively assign experts to different hops without solid theoretical grounding, spectral GNNs are based on principled insights from graph signal processing but often rely on manually designed filters and global operators, limiting their scalability and adaptability. In this work, we identify an inherent connection between these two families by showing that the eigengraph components in spectral methods can be treated as experts within an MoE framework. Building on this insight, we propose MORGAN, a novel spectral GNN that integrates Mixture-of-Experts into the spectral domain. MORGAN performs eigen-decomposition of the graph Laplacian, partitions the spectrum into multiple frequency bands, and assigns a dedicated expert network to each band. A learnable gating function dynamically combines these experts based on the spectral characteristics of the input. To support scalable and inductive learning, we further develop MORGAN(L), which incorporates subgraph sampling to enable localized spectral filtering without requiring full access to the graph Laplacian. Extensive experiments on 16 real-world benchmark datasets show that MORGAN achieves competitive or superior performance compared to state-of-the-art baselines, particularly in inductive node classification under heterophilic settings.
We present a novel framework for online learning in Stackelberg general-sum games, where two agents, the leader and follower, engage in sequential turn-based interactions. At the core of this approach is a learned diffeomorphism that maps the joint action space to a smooth spherical Riemannian manifold, referred to as the Stackelberg manifold. This mapping, facilitated by neural normalizing flows, ensures the formation of tractable isoplanar subspaces, enabling efficient techniques for online learning. Leveraging the linearity of the agents' reward functions on the Stackelberg manifold, our construct allows the application of linear bandit algorithms. We then provide a rigorous theoretical basis for regret minimization on the learned manifold and establish bounds on the simple regret for learning Stackelberg equilibrium. This integration of manifold learning into game theory uncovers a previously unrecognized potential for neural normalizing flows as an effective tool for multi-agent learning. We present empirical results demonstrating the effectiveness of our approach compared to standard baselines, with applications spanning domains such as cybersecurity and economic supply chain optimization.
Selecting a subset of promising candidates from a large pool is crucial across various scientific and real-world applications. Conformal selection offers a distribution-free and model-agnostic framework for candidate selection with uncertainty quantification. While effective in offline settings, its application to online scenarios, where data arrives sequentially, poses challenges. Notably, conformal selection permits the deselection of previously selected candidates, which is incompatible with applications requiring irreversible selection decisions. This limitation is particularly evident in resource-intensive sequential processes, such as drug discovery, where advancing a compound to subsequent stages renders reversal impractical. To address this issue, we extend conformal selection to an online Accept-to-Reject Changes (ARC) procedure: non-selected data points can be reconsidered for selection later, and once a candidate is selected, the decision is irreversible. Specifically, we propose a novel conformal selection method, Online Conformal Selection with Accept-to-Reject Changes (dubbed OCS-ARC), which incorporates online Benjamini–Hochberg procedure into the candidate selection process. We provide theoretical guarantees that OCS-ARC controls the false discovery rate (FDR) at or below the nominal level at any timestep under both i.i.d. and exchangeable data assumptions. Additionally, we theoretically show that our approach naturally extends to multivariate response settings. Extensive experiments on synthetic and real-world datasets demonstrate that OCS-ARC significantly improves selection power over the baseline while maintaining valid FDR control across all examined timesteps.
Gaussian process regression (GPR) is a popular nonparametric Bayesian method that provides predictive uncertainty estimates and is widely used in safety-critical applications. While prior research has introduced various uncertainty bounds, most existing approaches require access to specific input features, and rely on posterior mean and variance estimates or the tuning of hyperparameters. These limitations hinder robustness and fail to capture the model’s global behavior in expectation. To address these limitations, we propose a chaining-based framework for estimating upper and lower bounds on the expected extreme values over unseen data, without requiring access to specific input features. We provide kernel-specific refinements for commonly used kernels such as RBF and Matérn, in which our bounds are tighter than generic constructions. We further improve numerical tightness by avoiding analytical relaxations. In addition to global estimation, we also develop a novel method for local uncertainty quantification at specified inputs. This approach leverages chaining geometry through partition diameters, adapting to local structures without relying on posterior variance scaling. Our experimental results validate the theoretical findings and demonstrate that our method outperforms existing approaches on both synthetic and real-world datasets.
FedAdamW: A Communication-Efficient Optimizer with Convergence and Generalization Guarantees for Federated Large Models
PDF ↗AdamW has become one of the most effective optimizers for training large-scale models. We have also observed its effectiveness in the context of federated learning (FL). However, directly applying AdamW in federated learning settings poses significant challenges: (1) due to data heterogeneity, AdamW often yields high variance in the second-moment estimate v; (2) the local overfitting of AdamW may cause client drift; and (3) Reinitializing moment estimates (v, m) at each round slows down convergence. To address these challenges, we propose the first Federated AdamW algorithm, called FedAdamW, for training and fine-tuning various large models. FedAdamW aligns local updates with the global update using both a local correction mechanism and decoupled weight decay to mitigate local overfitting. FedAdamW efficiently aggregates the mean of the second-moment estimates to reduce their variance and reinitialize them. Theoretically, we prove that FedAdamW achieves a linear speedup convergence rate of O（p（L∆σ2l ）/（SKRε2） + （L∆）/R） without heterogeneity assumption, where S is the number of participating clients per round, K is the number of local iterations, and R is the total number of communication rounds. We also employ PAC-Bayesian generalization analysis to explain the effectiveness of decoupled weight decay in local training. Empirically, we validate the effectiveness of FedAdamW on language and vision Transformer models. Compared to several baselines, FedAdamW significantly reduces communication rounds and improves test accuracy.
Knowledge distillation (KD) has proven highly effective for compressing large models and enhancing the performance of smaller ones. However, its effectiveness diminishes in cross-modal scenarios, such as vision-to-language distillation, where inconsistencies in representation across modalities lead to difficult knowledge transfer. To address this challenge, we propose frequency-decoupled cross-modal knowledge distillation, a method designed to decouple and balance knowledge transfer across modalities by leveraging frequency-domain features. We observed that low-frequency features exhibit high consistency across different modalities, whereas high-frequency features demonstrate extremely low cross-modal similarity. Accordingly, we apply distinct losses to these features: enforcing strong alignment in the low-frequency domain and introducing relaxed alignment for high-frequency features. We also propose a scale consistency loss to address distributional shifts between modalities, and employ a shared classifier to unify feature spaces. Extensive experiments across multiple benchmark datasets show our method substantially outperforms traditional KD and state-of-the-art cross-modal KD approaches.
Node-level federated graph clustering allows multiple unlabeled subgraph holders to collaboratively train on node-level tasks without sharing private information. Existing methods usually assume that the node attributes are complete and have achieved promising progress. However, in the Federated Graph Learning (FGL) scenarios, this assumption is overly strict due to failures in data collection devices. Consequently, most existing FGL frameworks struggle to extract useful features from attribute-incomplete graphs for clustering, yet the issue remains underexplored. To bridge this gap, we propose a causally-aware attribute completion for Incomplete Federated Graph Clustering (IFedGC), which constructs a reliable global causal structure that incorporates clustering-friendly information to guide attribute completion for each subgraph. Specifically, in the attribute completion step, we first construct the causal structure to extract the causal relationships between initialized features, and then upload them to the server. Subsequently, we integrate multiple uploaded causal structures into a global causal one to achieve cross-client attribute completion. Moreover, to support reliable clustering, we first collect the high-confidence cluster centroids from each subgraph using a Graph Neural Network (GNN) model and subsequently aggregate these centroids on the server. The above two steps are seamlessly integrated into a unified FGL framework to obtain a clustering-oriented causal structure, which is sent back to the client to promote high-quality attribute completion for better clustering. Extensive results on five benchmark datasets demonstrate the effectiveness and superiority of IFedGC against its competitors.
In the federated clustering task, structural heterogeneity across clients inevitably impedes effective multi-source information sharing. To solve this issue, Personalized Federated Learning (PFL) has emerged as a potentially effective solution for image and text clustering. Unlike Euclidean data, graph-structured data exhibits diverse and fragile local patterns, which widely exist in real-world scenarios. Multi-graph data analysis in the federated learning setting is challenging and important, yet remains underexplored. This motivates us to propose a novel PERsonalized Federated graph-lEvel Clustering neTwork (PERFECT), which generates a specialized aggregation strategy for each client by uploading key model parameters and representative samples without sharing private information. Specifically, for each client, we first reconstruct privacy-preserving representative samples in a min-max optimization manner and then upload these samples to the server for subsequent personalized parameter aggregation. On the server, we first extract graph-level embeddings from the uploaded data, and then estimate affinities among multiple learned embeddings to formulate a personalized aggregation strategy for each client. Subsequently, to help each local model better identify the cluster boundaries, we utilize clustering-wise gradient to update the key components in the personalized model parameters from the server. Extensive experimental results have demonstrated the effectiveness and superiority of PERFECT over its competitors.
Beyond Observations: Reconstruction Error-Guided Irregularly Sampled Time Series Representation Learning
PDF ↗Irregularly sampled time series (ISTS), characterized by non-uniform time intervals with natural missingness, are prevalent in real-world applications. Existing approaches for ISTS modeling primarily rely on observed values to impute unobserved ones or infer latent dynamics. However, these methods overlook a critical source of learning signal: the reconstruction error inherently produced during model training. Such error implicitly reflects how well a model captures the underlying data structure and can serve as an informative proxy for unobserved values. To exploit this insight, we propose iTimER, a simple yet effective self-supervised pre-training framework for ISTS representation learning. iTimER models the distribution of reconstruction errors over observed values and generates pseudo-observations for unobserved timestamps through a mixup strategy between sampled errors and the last available observations. This transforms unobserved timestamps into noise-aware training targets, enabling meaningful reconstruction signals. A Wasserstein metric aligns reconstruction error distributions between observed and pseudo-observed regions, while a contrastive learning objective enhances the discriminability of learned representations. Extensive experiments on classification, interpolation, and forecasting tasks demonstrate that iTimER consistently outperforms state-of-the-art methods under the ISTS setting.
Fine-grained image-text alignment is a pivotal challenge in multimodal learning, underpinning key applications such as visual question answering, image captioning, and vision-language navigation. Unlike global alignment, fine-grained alignment requires precise correspondence between localized visual regions and textual tokens, often hindered by noisy attention mechanisms and oversimplified modeling of cross-modal relationships. In this work, we identify two fundamental limitations of existing approaches: the lack of robust intra-modal mechanisms to assess the significance of visual and textual tokens, leading to poor generalization in complex scenes; and the absence of fine-grained uncertainty modeling, which fails to capture the one-to-many and many-to-one nature of region-word correspondences. To address these issues, we propose a unified approach that incorporates significance-aware and granularity-aware modeling and region-level uncertainty modeling. Our method leverages modality-specific biases to identify salient features without relying on brittle cross-modal attention, and represents region features as a mixture of Gaussian distributions to capture fine-grained uncertainty. Extensive experiments on Flickr30K and MS-COCO demonstrate that our approach achieves state-of-the-art performance across various backbone architectures, significantly enhancing the robustness and interpretability of fine-grained image-text alignment.
Mixture-of-Experts (MoE) architectures have recently become a more prevalent choice for large language models (LLMs) than dense architectures due to their superior performance. However, billions of parameters bring MoE LLMs a huge cost for deployment and inference. To address these issues, knowledge distillation (KD) has become a widely adopted technique to compress LLMs. Existing KD methods for LLMs can be divided into dense-to-dense and moe-to-dense distillation. Dense-to-dense distillation transfers knowledge between single dense LLMs, while moe-to-dense distillation attempts to transfer knowledge between the MoE LLMs and the dense LLMs. However, the architectural mismatch prevents the student from fully absorbing knowledge when distilling MoE LLMs. To address this limitation, we investigate a new distillation setting, moe-to-moe, which aims to fully leverage expert knowledge of teachers and enable the student to absorb it more effectively. Compared to dense-to-dense and moe-to-dense, moe-to-moe suffers from two imbalance issues. First, expert-coverage deficiency reflects an imbalanced knowledge transfer of teacher experts: traditional distillation utilizes only the few experts activated by the teacher router. Second, routing imbalance appears when the student routing distribution drifts from the teacher, which makes it difficult for students to learn how to distribute different experts. To overcome these issues, we propose a novel distillation framework for moe-to-moe, Balanced Distillation (B-Distill), which equally spreads teacher expertise across student experts while regularizing the student router toward teacher-consistent balance. First, to mitigate expert-coverage deficiency, we introduce Monte Carlo exploration, which stochastically perturbs router probabilities so every teacher and student expert is sampled without enlarging the search space. Second, to correct routing imbalance and avert load collapse, we propose an entropy-aware router distillation mechanism that aligns the student router with the teacher while curbing over-concentration. Experiments show that B-Distill outperforms baselines by up to 6.6% in Rouge-L.
Can Pseudo-Label Be More Reliable? A Simple yet Effective Topology-Aware Graph Self-Training Method
PDF ↗Graph Neural Networks (GNNs) have demonstrated impressive success across a range of graph-based tasks. However, their performance in node classification typically relies on enough high-quality labeled data which are difficult to obtain in practice. Self-training emerges as a promising solution to tackle the issue of label scarcity. Most existing studies in this direction mainly rely on classification scores to explore high-confidence unlabeled samples. Nevertheless, these methods often lead to false positive samples, which hinders the capability of GNNs. To this end, we propose a simple yet effective Topology-Aware Graph Self-Training (TA-GST) method. Specifically, we first explore the origin of false positives in pseudo-labeled samples. We then design a topology-aware scoring method, which considers both the classification score and connectivity pattern to enhance the reliability of pseudo-labeled samples. Besides, we depart TA-GST from the traditional teacher-student pattern and simplify it in an end-to-end manner. Extensive experiments on seven real-world datasets demonstrate the effectiveness of our method.