Multimodal learning frequently faces two coupled challenges: modality imbalance, where dominant modalities suppress others during training, and modality conflict, where opposing gradient directions hinder optimization. Existing methods typically address these issues in isolation, yet they are intrinsically correlated and most fundamentally reflected in the gradient space—severe imbalance may obscure conflicts, while suppressing conflict may homogenize features and worsen imbalance, affecting fusion performance. To jointly address this coupled challenge, we propose Reconcile Gradient Modulation (RGM), a unified framework that adaptively adjusts gradient magnitude and direction for harmony multimodal learning. The core of RGM is SynOrth Grad, which minimizes Dirichlet energy to perform minimal-gradient surgery. It enhances cooperation synergy when modalities are aligned and enforces orthogonality to preserve uniqueness in conflict situations, thus promoting stable and balanced learning. To guide this modulation, we propose Cumulative Gradient Energy (CGE) as a convergence-guaranteed measure of modality-wise progress, and construct a Balance-nonConflict Plane (BCP) for real-time diagnosis and control of training dynamics. Experiments on diverse benchmarks validate our effectiveness and generalizability, consistently outperforming counterparts that are designed to handle multimodal imbalance or conflict independently.
论文检索
输入标题、作者或关键词，从 100,903 篇学术成果中精准定位
Talon: Breaking the Synchronization Barrier in Speculative Decoding with Hybrid Model-based and Retrieve-based Drafting
PDF ↗Large Language Models face fundamental deployment challenges due to the computational demands of auto-regressive token-by-token generation. While speculative decoding has emerged as a promising acceleration technique through its draft-then-verify framework, current implementations suffer from two critical limitations: (1) mutual waiting problem caused by sequential dependencies between draft generation and verification phases, and (2) constrained token acceptance rates where retrieval-based drafting methods under-perform in general domains while models-based drafting approaches show reduced efficacy in knowledge-intensive scenarios. To address these challenges, we propose Talon, a novel parallel inference architecture featuring two key innovations: (1) **a novel asynchronous execution paradigm** that decouples draft generation from verification, effectively eliminating synchronization bottlenecks, and (2) **an adaptive hybrid drafting strategy** that dynamically combines model-based and retrieval-based approaches to improve token acceptance rates across diverse domains. Extensive evaluations across standard benchmarks (MT-Bench, HumanEval, GSM8K, Alpaca, CNN/DM) demonstrate Talon's exceptional performance, achieving 4.04x–6.52x acceleration across multiple model families including Vicuna, Deepseek, and LLaMA series. These results represent a significant advancement over existing speculative decoding methods (EAGLE 1-3, Hydra, Medusa, Lookahead, SPS, and PLD), establishing a new paradigm for speculative decoding.
Existing feature engineering methods based on large language models (LLMs) have not yet been applied to multi-label learning tasks. They lack the ability to model complex label dependencies and are not specifically adapted to the characteristics of multi-label tasks. To address the above issues, we propose Feature Engineering Automation for Multi-Label Learning (FEAML), an automated feature engineering method for multi-label classification which leverages the code generation capabilities of LLMs. By utilizing metadata and label co-occurrence matrices, LLMs are guided to understand the relationships between data features and task objectives, based on which high-quality features are generated.The newly generated features are evaluated in terms of model accuracy to assess their effectiveness, while Pearson correlation coefficients are used to detect redundancy. FEAML further incorporates the evaluation results as feedback to drive LLMs to continuously optimize code generation in subsequent iterations. By integrating LLMs with a feedback mechanism, FEAML realizes an efficient, interpretable and self-improving feature engineering paradigm. Empirical results on various multi-label datasets demonstrate that our FEAML outperforms other feature engineering methods.
Large Language Models (LLMs) hold significant potential for enhancing healthcare applications, yet their deployment is hindered by high computational and memory demands. Model compression techniques offer solutions to reduce these demands, but their impact on medical LLMs remains underexplored. In this paper, we introduce CMedBench, the first comprehensive benchmark for evaluating compressed LLMs in medical contexts. CMedBench assesses five core dimensions: Medical Knowledge Ability, Medical Application Ability, Trustworthiness Maintenance, Compression Cross Combination, and Computational Efficiency. Through extensive empirical studies, we analyze the trade-offs between model efficiency and clinical performance across diverse models, datasets, and compression strategies. Our findings highlight critical limitations in current evaluation practices and provide a robust framework for aligning compression strategies with medical requirements. CMedBench serves as a vital resource for researchers and practitioners, guiding the development of efficient, trustworthy, and clinically effective LLMs for healthcare applications.
DiAPR: Dimensionally-Allocated Prototype Refinement for Non-Exemplar Class Incremental Learning
PDF ↗Non-Exemplar Class Incremental Learning (NECIL) strives to preserve classification performance in an evolving data stream without revisiting old-class exemplars. Current methods mitigate catastrophic forgetting by replaying and augmenting historical prototypes as surrogates for old classes. However, they treat prototypes as holistic representations for global-level augmentations, which overlook dimensional semantic disparity and old-new class relationships, failing to maintain old-class discriminability and adaptability to the evolving feature space. To address this challenge, we propose Dimensionally-Allocated Prototype Refinement (DiAPR), a granular framework that progressively refines prototypes to exhibit class separability in the new feature space through three modules. Specifically, Distribution-aware Pairing (DAP) captures old-new class semantic consistency to guide Granular Semantic Allocation (GSA) in dimension-wise conflation, while Cross-Dimensional Transition (CDT) enhances cross-dimensional dependencies. The resulting prototypes sharpen classifier decision boundaries. Moreover, CDT inherently enables softened feature alignment, thereby yielding a more compatible feature space. Extensive experiments demonstrate DiAPR’s superiority, with improvements over SOTA by 2.35%, 0.70%, 0.96% on three CIFAR-100 settings, 1.03%, 0.54%, 0.40% on Tiny-ImageNet, and 0.60% on ImageNet-Subset.
Radiology report generation from longitudinal medical data is critical for assessing disease progression and automating diagnostic workflows. While recent methods incorporate longitudinal information, they primarily rely on multimodal feature fusion, with limited capacity for explicit disease evolution modeling and temporal reasoning. To address this, we propose MARE, an end-to-end framework that formulates longitudinal radiology report generation as a multimodal analogical reasoning task. Inspired by the Abduction–Mapping–Induction paradigm, MARE models latent relational structures underlying disease evolution by aligning lesion-level visual features across time and mapping them to the textual domain for temporally coherent and clinically meaningful report generation. To mitigate the spatial misalignment caused by patient positioning or imaging variation, we introduce an Adaptive Region Alignment (ARA) module for robust temporal correspondence. Additionally, we design Dual Evolution Consistency (DEC) losses to regularize analogical reasoning by enforcing temporal coherence in both visual and textual evolution paths. Extensive experiments on the Longitudinal-MIMIC dataset demonstrate that MARE significantly outperforms state-of-the-art baselines across both natural language generation and clinical effectiveness metrics, highlighting the value of structured analogical reasoning for disease evolution-aware report generation.
Learning Whom to Align With: Progressive Anomaly Combination Detection for Partially View-Aligned Clustering
PDF ↗Partially View-aligned Clustering (PVC) addresses the challenge of partial view alignment in multi-view learning by leveraging complementary and consistent information. While existing PVC methods show promise, most rely on distance-based strategies that are sensitive to view-specific details and noise, limiting their robustness. In this work, we propose a novel view alignment strategy that reformulates the alignment task as an anomaly detection problem. Rather than learning a view-alignment matrix that enforces strict one-to-one correspondences across views, we adopt a progressive approach to identify well-aligned samples. Specifically, we sample subsets of data by generating random view combinations from unaligned samples and propose an anomaly combination detection module to evaluate the alignment consistency of these combinations. In addition, our progressive training framework alternates between updating model parameters and selecting high-confidence view combinations for subsequent optimization. By reformulating view alignment as an anomaly detection task, our approach provides a more robust and effective solution to partial view alignment. Experiments on benchmark datasets demonstrate that our method outperforms state-of-the-art approaches in the PVC problem.
Fairness-Aware Design for Contextual Experiments: Guaranteeing Reliability and Equity in Heterogeneous Subgroups
PDF ↗Experimental design is critical for evidence-based decision-making in healthcare, marketing, and public policy. However, designing efficient experiments across heterogeneous subgroups presents significant challenges. Existing methods often optimize for statistical power or overall sample efficiency, overlooking crucial fairness considerations across these different subgroups. To address this gap, we introduce a Fairness-Aware Contextual Track-and-Stop Design (F-CTSD) algorithm. The proposed F-CTSD algorithm provides statistical guarantees on subgroup fairness while minimizing required sample sizes. We quantify the fairness-efficiency trade-off and derive the sample complexity bound for the proposed F-CTSD algorithm under its fairness constraints. We further theoretically prove that the proposed F-CTSD algorithm consistently produces accurate treatment effect estimates even under fairness requirements, enhancing statistical reliability. Numerical experiments show that the proposed F-CTSD algorithm outperforms existing methods, achieving higher sample efficiency while reducing subgroup fairness violations by 4.95%.
Quantization is a pivotal technique for enhancing communication efficiency in Federated Learning (FL). Traditional quantization methods often set uniform intervals, may fail to adequately characterize non-uniform data distributions, thus leading to substantial estimation errors and degrated model performance. Non-uniform quantization can better solve the problem. However, when applied to FL, it would bring additional communication overheads for the alignment of parameter distributions among distributed models. To address this issue, we propose Bisection Interval Quantization (BIQ), a novel non-uniform quantization framework for FL with great communication efficiency. In particular, BIQ works by optimizing the interval selection through recursive bisection among distributed clients without extra parameter communication. For scenarios involving amounts of boundary inputs, we further design Weighted Bisection Interval Quantization (WBIQ), which incorporates maximum likelihood estimation to refine boundary value reconstruction to enhance the estimation quality of boundary inputs. Our theoretical analysis rigorously establishes, for the first time under biased quantization conditions, that both BIQ and WBIQ achieve tighter error bounds and enhanced stability. Extensive experiments validate that both BIQ and WBIQ significantly accelerate the convergence of FL model training when compared to the state-of-the-art quantizers under both convex and non-convex settings.
Training general agents to follow complex instructions (tasks) in intricate environments (levels) remains a core challenge in reinforcement learning. Random sampling of task-level pairs often produces unsolvable combinations, highlighting the need to co-design tasks and levels. While unsupervised environment design (UED) has proven effective at automatically designing level curricula, prior work has only considered a fixed task. We present ATLAS (Aligning Tasks and Levels for Autocurricula of Specifications), a novel method that generates joint autocurricula over tasks and levels. Our approach builds upon UED to automatically produce solvable yet challenging task-level pairs for policy training. To evaluate ATLAS and drive progress in the field, we introduce an evaluation suite that models tasks as reward machines in Minigrid levels. Experiments demonstrate that ATLAS vastly outperforms random sampling approaches, particularly when sampling solvable pairs is unlikely. We further show that mutations leveraging the structure of both tasks and levels accelerate convergence to performant policies.
Without manual annotations, unsupervised cross-modal hashing (UCMH) aims to achieve efficient clustering and retrieval by leveraging data interrelationships. However, the retrieval accuracy is constrained by two main aspects: 1) insufficient exploration of data relationships; 2) existing knowledge mining strategies are not well aligned with the architectural properties of multilayer perceptrons. Through summary and error analysis, the human brain is able to achieve fast learning through experience and minimal data. Inspired by this cognitive process, we propose a novel Error Notebook strategy, named ENHash, to more effectively capture similarity information between multi-modal data for fine-grained unsupervised clustering. Firstly, simulating the human process of summarizing experiences, ENHash gradually integrates the information from each batch into a global clustering representation. Secondly, drawing upon human error analysis capabilities, ENHash utilizes the summarized experiences to identify and record incorrectly predicted hash codes. Finally, by leveraging the knowledge derived from this analysis, ENHash guides the hash function to learn fine-grained patterns from the errors. To the best of our knowledge, ENHash represents the first attempt at integrating cognitively-inspired mechanisms into fine-grained UCMH optimization paradigms. We evaluate the proposed ENHash against eight state-of-the-art methods on three widely used datasets and one fine-grained cross-modal dataset. Experimental results show that ENHash achieves substantial improvements over existing approaches.
Training large neural network models requires extensive computational resources, often distributed across several nodes and accelerators. Recent findings suggest that it may be sufficient to only exchange the fast moving components of the gradients, while accumulating momentum locally (Decoupled Momentum, or DeMo). However, DeMo assumes that models fit on a single accelerator. We relax this assumption and introduce FlexDeMo, whereby nodes fully shard model parameters locally between different accelerators, while inter-node communication is reduced by synchronizing only fast-moving components instead of the full gradients -- resulting in a hybrid sharded data parallel training strategy. We further introduce a framework, denoted as DeToNATION, that generalizes DeMo, FlexDeMo, and other popular distributed training schemes such as DiLoCo -- introducing new variations of replication schemes and challenging choices made in DeMo. Our results across language and vision domains show that FlexDeMo attains similar validation loss as hybrid sharded data parallel training employing AdamW and full gradient synchronization, while being substantially faster. FlexDeMo is thus a promising distributed training scheme for the largest machine learning models.
In supervised learning with distributional inputs in the two-stage sampling setup, relevant to applications like learning-based medical screening or causal learning, the inputs (which are probability distributions) are not accessible in the learning phase, but only samples thereof. This problem is particularly amenable to kernel-based learning methods, where the distributions or samples are first embedded into a Hilbert space, often using kernel mean embeddings (KMEs), and then a standard kernel method like Support Vector Machines (SVMs) is applied, using a kernel defined on the embedding Hilbert space. In this work, we contribute to the theoretical analysis of this latter approach, with a particular focus on classification with distributional inputs using SVMs. We establish a new oracle inequality and derive consistency and learning rate results. Furthermore, for SVMs using the hinge loss and Gaussian kernels, we formulate a novel variant of an established noise assumption from the binary classification literature, under which we can establish learning rates. Finally, some of our technical tools like a new feature space for Gaussian kernels on Hilbert spaces are of independent interest.
Efficient Multimodal Large Language Models (MLLMs) compress vision tokens to reduce resource consumption, but the loss of visual information can degrade comprehension capabilities. Although some priors introduce Knowledge Distillation to enhance student models, they overlook the fundamental differences in fine-grained vision comprehension caused by unbalanced vision tokens between the efficient student and vanilla teacher. In this paper, we propose EM-KD, a novel paradigm that enhances the Efficient MLLMs with Knowledge Distillation. To overcome the challenge of unbalanced vision tokens, we first calculate the Manhattan distance between the vision logits of teacher and student, and then align them in the spatial dimension with the Hungarian matching algorithm. After alignment, EM-KD introduces two distillation strategies: 1) Vision-Language Affinity Distillation (VLAD) and 2) Vision Semantic Distillation (VSD). Specifically, VLAD calculates the affinity matrix between text tokens and aligned vision tokens, and minimizes the smooth L1 distance of the student and the teacher affinity matrices. Considering the semantic richness of vision logits in the final layer, VSD employs the reverse KL divergence to measure the discrete probability distributions of the aligned vision logits over the vocabulary space. Comprehensive evaluation on diverse benchmarks demonstrates that EM-KD trained model outperforms prior Efficient MLLMs on both accuracy and efficiency with a large margin, validating its effectiveness. Compared with previous distillation methods, which are equipped with our proposed vision token matching strategy for fair comparison, EM-KD also achieves better performance.
Traditional Discriminant analysis (DA) is one of the classical supervised learning algorithms to reduce the dimensionality of data with Gaussian assumption. Since the unique class mean in traditional DA is intractable to estimate the non-Gaussian distrbution of data, some existing DA algorithms based on the clustering criterion focus on learning multiple means in each class so as to address the non-Gaussian issue. The clustering-based DA inevitably involved the constraint optimization problem to learn multiple means, which may lead to the locally optimal solution. To address these issues, inspired by the smooth approximation theory and the concept of Kolmogorov mean, this paper explores an unconstraint function with asymptotic property as an alternative proxy to clustering-based DA algorithms. Thus the derived DA algorithm, i.e., adaptive and asymptotic mean-based subclass discriminant analysis (AASDA), which not only leverages multiple means to represent different subclasses in same class but also adaptively and asymptotically learns the similar mean for each sample in the learned optimal subspace via the gradient-based optimizer. The asymptotic analysis of unconstraint function, the gradient analysis and convergence guarantee of proposed criterion verify the effectiveness of AASDA algorithm. Its merits are thoroughly assessed on a suite of synthetic and real world data experiments.
Behavior Tokens Speak Louder: Disentangled Explainable Recommendation with Behavior Vocabulary
PDF ↗Recent advances in explainable recommendation have explored the integration of language models to analyze natural language rationales for user–item interactions. Despite their potential, existing methods often rely on ID-based representations that obscure semantic meaning and impose structural constraints on language models, thereby limiting their applicability in open-ended scenarios. These challenges are intensified by the complex nature of real-world interactions, where diverse user intents are entangled and collaborative signals rarely align with linguistic semantics. To overcome these limitations, we propose BEAT, a unified and transferable framework that tokenizes user and item behaviors into discrete, interpretable sequences. We construct a behavior vocabulary via a vector-quantized autoencoding process that disentangles macro-level interests and micro-level intentions from graph-based representations. We then introduce multi-level semantic supervision to bridge the gap between behavioral signals and language space. A semantic alignment regularization mechanism is designed to embed behavior tokens directly into the input space of frozen language models. Experiments on three public datasets show that BEAT improves zero-shot recommendation performance while generating coherent and informative explanations. Further analysis demonstrates that our behavior tokens capture fine-grained semantics and offer a plug-and-play interface for integrating complex behavior patterns into large language models.
Federated Multi-View Clustering has gained increasing attention for its ability to discover complementary clustering structures of distributed multi-view data while preserving data privacy. However, real-world clients often only have access to partial views, and the view incompleteness poses great challenges to federated multi-view feature fusion to exploit consistent and complementary information. Moreover, efficiency is highly expected in federated scenarios due to the limited resources of each client. To alleviate these issues, we propose Federated Incomplete Multi-View Clustering with Tensorized Low-Rank Constraint (FIMVC-TLRC), which incorporates anchors to improve efficiency and is able to address prevalent view incompleteness issue in federated scenarios. FIMVC-TLRC aligns the local anchor graphs and employs a tensorized low-rank constraint based on the tensor Schatten p-norm to enforce the consistency of the data representations learned by each client. Besides, a federated optimization framework is developed to jointly optimize the construction and alignment of anchor graphs, thus enabling collaborative and privacy-preserving training. Experimental results on multiple datasets demonstrate its effectiveness.
WIET: Harmonizing Group-aware Model Weighting and Worker Allocation for Ensemble Temporal Prediction MaaS
PDF ↗Ensemble Temporal Prediction Model-as-a-Service (ETP-MaaS) has become crucial in fields like financial modeling and cloud monitoring. Existing solutions fail to co-optimally address a two-fold challenge of dynamic collaboration and heterogeneity, treating models as independent entities and employing simplistic worker allocation rules. However, at the model level, data volatility means that optimal performance requires identifying and weighting constantly shifting subgroups of base models, not just individual ones; at the system level, these model groups must be efficiently mapped to a pool of heterogeneous and dynamically available workers. To this end, we introduce WIET, an efficient ETP-MaaS system that co-optimizes model weighting and worker allocation. For adaptive weighting, WIET identifies evolving group behaviors among base models and propose a novel group temporal locality-enhanced weighting method. Additionally, WIET develops an efficient, multi-dimensional worker allocation method powered by hybrid heuristic optimization, effectively reducing bottlenecks and resource waste. Experiments show WIET consistently outperforms state-of-the-art methods in terms of accuracy, latency, and resource usage across various workloads and tasks.
Conventional time-series forecasting methods typically aim to minimize overall prediction error, without accounting for the varying importance of different forecast ranges in downstream applications. We propose a training methodology that enables forecasting models to adapt their focus to application-specific regions of interest at inference time, without retraining. The approach partitions the prediction space into fine-grained segments during training, which are dynamically reweighted and aggregated to emphasize the target range specified by the application. Unlike prior methods that predefine these ranges, our framework supports flexible, on-demand adjustments. Experiments on standard benchmarks and a newly collected wireless communication dataset demonstrate that our method not only improves forecast accuracy within regions of interest but also yields measurable gains in downstream task performance. These results highlight the potential for closer integration between predictive modeling and decision-making in real-world systems.
To Align or Not to Align: Strategic Multimodal Representation Alignment for Optimal Performance
PDF ↗Multimodal learning often relies on aligning representations across modalities to enable effective information integration—an approach traditionally assumed to be universally beneficial. However, prior research has primarily taken an observational approach, examining naturally occurring alignment in multimodal data and exploring its correlation with model performance, without systematically studying the direct effects of explicitly enforced alignment between representations of different modalities. In this work, we investigate how explicit alignment influences both model performance and representation alignment under different modality-specific information structures. Specifically, we introduce a controllable contrastive learning module that enables precise manipulation of alignment strength during training, allowing us to explore when explicit alignment improves or hinders performance. Our results on synthetic and real datasets under different data characteristics show that the impact of explicit alignment on the performance of unimodal models is related to the characteristics of the data: the optimal level of alignment depends on the amount of redundancy between the different modalities. We can find an optimal alignment strength that balances modality-specific signals and shared redundancy in the mixed information distributions. This work can help practitioners on when and how to enforce alignment for optimal unimodal encoder performance.