The task of stochastic human motion prediction has attracted significant attention in recent years due to its wide-ranging applications in robotics, animation, and human-computer interaction. While diffusion models have demonstrated promising progress in this domain, they remain hindered by two critical limitations: (1) slow inference speeds due to their reliance on iterative sampling, and (2) performance degradation resulting from suboptimal sample allocation during generation. To overcome these challenges, we propose SPARD (Single-step Inference with Adaptive Sampling in Residual Diffusion for Human Motion Prediction), a novel framework that achieves efficient single-step inference while maintaining high predictive accuracy. Furthermore, we introduce a novel adaptive noise predictor module that dynamically samples latent representations based on observed motion sequences, ensuring both accuracy and plausibility in generated motions. Extensive experiments on benchmark datasets demonstrate that SPARD significantly outperforms state-of-the-art methods in both inference efficiency and motion quality, achieving a 15× to 18× speedup in sampling time compared to conventional diffusion-based baselines while preserving generation quality.

Recently, the prominent performance of large language models (LLMs) has been largely driven by multi-task instruct-tuning. Unfortunately, this training paradigm suffers from a key issue, named cross-task interference, due to conflicting gradients over shared parameters among different tasks. Some previous methods mitigate this issue by isolating task-specific parameters, e.g., task-specific neuron selection and mixture-of-experts. In this paper, we empirically reveal that the cross-task interference still exists for the existing solutions because of many parameters also shared by different tasks, and accordingly, we propose a novel solution, namely Basic Abilities Decomposition for multi-task Instruct-Tuning (BADIT). Specifically, we empirically find that certain parameters are consistently co-activated, and that co-activated parameters naturally organize into base groups. This motivates us to analogize that LLMs encode several orthogonal basic abilities, and that any task can be represented as a linear combination of these abilities. Accordingly, we propose BADIT that decomposes LLM parameters into orthogonal high-singular-value LoRA experts representing basic abilities, and dynamically enforces their orthogonality during training via spherical clustering of rank-1 components. We conduct extensive experiments on the SuperNI benchmark with 6 LLMs, and empirical results demonstrate that BADIT can outperform SOTA methods and mitigate the degree of cross-task interference.

The rapid proliferation of location-based social media platforms has greatly accelerated the dissemination of geo-tagged information, but it has also facilitated the widespread propagation of localized rumors. Geo-textual Rumor Detection (GRD) has therefore become an important research topic in geoinformatics aimed at automatically identifying deceptive content tied to specific geographical contexts. However, most existing GRD methods rely on learning static patterns from offline datasets, which limits their ability to generalize to emergent local events characterized by rapidly evolving spatial-temporal information distributions. To better understand this limitation, we conduct preliminary analyses of model fitting behaviors during training and identify two critical issues: imbalanced fitting between real and fake classes, and low-rank feature representations caused by the model’s tendency to overfit to homogeneous real patterns. These phenomena directly lead to the severe loss of vital spatial information, which significantly constrains the model’s capacity to capture the diverse spatial and textual patterns inherent in localized rumors. To address these challenges, we propose a novel framework named Decomposing Orthogonal Spatial Subspaces for Emergent Geo-textual rumor detection (Doseg). Our approach decomposes model transformation matrices via singular value decomposition, explicitly separating linguistic semantic, geographical spatial information-aligned, and localized event-specific spatial components while enforcing orthogonality constraints to enhance spatial feature diversity. Extensive experiments on benchmark geo-textual datasets with strict spatial-temporal splits demonstrate that our method substantially improves detection performance and increases the number of dominant principal components in feature representations, leading to stronger generalization for emergent geo-textual rumor scenarios within the geospatial ecosystem.

Partial multi-label learning (PML) addresses weakly-supervised scenarios where each instance is associated with a candidate label set containing both ground-truth and noisy labels. Existing PML methods primarily focus on instance-level features or pairwise label correlations for disambiguation. Building on recent insights that exploiting pairwise label correlations improves disambiguation, we observe that high-order label correlations provide even stronger disambiguation evidence in partial settings because ground-truth labels with high-order correlations frequently co-occur, while noise labels produce inconsistent combinations that rarely repeat. To exploit this property, we propose REHC-PML (Label Confidence REcovery with High-order Label Correlations in Partial Multi-label Learning), which mines frequent high-order co-occurrences from candidate sets to identify global high-order label correlations, selects instance-relevant correlations via Gumbel-Softmax pruning, and propagates their evidence to constituent labels for confidence recovery. Self-training iteratively refines pseudo-labels and trains the classifier. Extensive experiments on both real-world and UCI datasets demonstrate the effectiveness of REHC-PML.

Diffusion-based methods have shown strong potential for stochastic human motion prediction, but existing ap proaches typically formulate the task as a conditional diffusion process initialized from Gaussian noise. This introduces a substantial distribution mismatch between the Gaussian prior and the conditional future-motion manifold, increasing the difficulty of reverse inference and often requiring many denoising steps. Although standard Brownian bridge diffusion provides a more condition-aware alternative, it fixes the endpoint to a deterministic target, resulting in a zero-variance terminal endpoint that may restrict its ability to model the stochastic and multimodal nature of future human motion. To address these issues, we propose HumanBBD, a generalized Brownian bridge diffusion framework for stochastic human motion prediction. HumanBBD explic itly constructs a bridge process between the observed-motion domain and the future-motion domain, thereby reducing initialization mismatch and providing stronger condition-aware generation. Moreover, by relaxing the zero-variance terminal endpoint into a distributional endpoint, HumanBBD improves the flexibility and expres siveness of bridge diffusion for modeling multimodal futures. Built upon this formulation, we further develop a unified spatio-temporal generation backbone that jointly captures long-range temporal dynamics and structured spatial dependencies among joints. Extensive experiments on HumanEva-I, Human3.6M, and AMASS demonstrate that HumanBBD achieves strong prediction performance while requiring only 5-10 reverse steps, substantially improving sampling efficiency compared with conventional diffusion-based methods. Comprehensive empiri cal analyses further validate the effectiveness, efficiency, and robustness of the proposed framework. Overall, HumanBBD provides an effective and efficient alternative to standard conditional diffusion for stochastic human motion prediction.

The proliferation of fake news has become a global risk. Although existing fake news detectors perform well in predefined domains such as politics, they struggle often in low-resource domains. Recent studies show that multi-domain fake news detection (MFND) can improve model performance. Building on this line of work, we further investigate the task of cross-domain fake news detection (CFND), which aims to identify emerging fake news in unseen domains. Specifically, our goal is to improve cross-domain generalization by mining general knowledge and implicit target-domain information from labeled data in source domains. To facilitate effective knowledge transfer, we first establish the relationships between news articles and domains by introducing a domain inference task that estimates the implicit domain distributions of news samples. Furthermore, we propose a mixture-of-experts based CFND model (named MEDAG), which integrates knowledge from multiple domain experts through a gating mechanism. The gate network is trained with guidance from dynamic implicit domain information and a hypernetwork. To demonstrate the effectiveness of MEDAG, we conduct extensive comparative experiments against both MFND and CFND baselines, along with comprehensive ablation studies. Experimental results on two real-world English and Chinese datasets show that MEDAG achieves an F1-score improvement of 1.38% on the English dataset and improvements of 1.74%, 1.02%, and 2.12% across three groups of the Chinese dataset on unseen news domains.

Time series foundation models (TSFMs) have recently achieved strong zero-shot forecasting performance through large-scale pretraining and retrieval-augmented prediction. However, our empirical analysis reveals a non-trivial limitation of retrieval-based forecasting: retrieval tends to induce more oscillatory predictions, improving performance on highly fluctuating series while degrading accuracy on smoother, trend-dominated ones. This suggests that retrieved information may be fused into prediction without explicitly distinguishing stable temporal structure from instance-specific variations, which can reduce robustness under distribution shifts. We propose a Retrieval-guided Invariant-Dynamic DEcomposition framework for time series forecasting. Rather than using retrieval as auxiliary predictive context, we leverage retrieved sequences as implicit samples from related environments to guide representation decomposition. Specifically, we first construct a retrieval-aware representation via attention-based aggregation, and then introduce a retrieval-guided routing mechanism to decompose it into an invariant component capturing stable shared structure and a dynamic component modeling context-dependent variations. These two components are forecast separately and fused for final prediction, enabling the model to preserve transferable patterns while remaining adaptive to evolving dynamics. We further design training objectives that encourage invariant learning and disentanglement, and provide theoretical insight showing that retrieval aggregation reduces variance and approximates invariant representation learning without explicit environment supervision. Extensive experiments demonstrate that our method consistently improves robustness under distribution shifts and outperforms existing TSFMs and retrieval-based baselines in zero-shot forecasting settings.

Vector-quantized (VQ) generative models have shown promising results in real-world image super-resolution (Real-ISR). However, existing methods typically rely on a monolithic latent space that entangles low-frequency structures with high-frequency textures. This entanglement forces a single codebook to capture a combinatorially complex set of structure-texture pairings, which constrains representational capacity and limits codebook utilization. To address this issue, we present HiTokSR, a hierarchical token prediction framework. Instead of using a single codebook, HiTokSR partitions the latent space along the channel dimension into frequency-aware groups, quantizing each with an independent sub-codebook. This coarse-to-fine design disentangles global structures from fine details, enhancing combinatorial expressiveness while circumventing the optimization instability of high-dimensional nearest-neighbor lookups. To further improve semantic consistency, our generator integrates priors from a vision foundation model via adaptive feature modulation, multi-scale class tokens, and a representation alignment loss. Additionally, we introduce an index-level perturbation strategy during decoder fine-tuning to bridge the train-test discrepancy in discrete token prediction. Extensive experiments on real-world benchmarks demonstrate that HiTokSR achieves state-of-the-art performance in both perceptual quality and reconstruction fidelity.

Multilabel text classification (MLTC) methods require enormous labeled training samples to ensure the model’s performance, which involves significant manual labor costs. An alternative to conducting MLTC is to only employ predefined representative words of classes, namely category words, as the weak supervision. In this article, we propose a novel weakly supervised (WS) learning MLTC framework consisting of two parts. First, we propose a novel category word selection method, namely category word selection with significance ranking and crowd-sourcing (Cws-src), which generates confident category words by manually selecting from the topically reranked words using a new TW-ITF weighting scheme, thereby effectively mitigating the noises in pseudolabels by filtering repetitive and less significant terms for each class, leading to improved classification performance. Subsequently, we propose a generic WS learning MLTC method, namely WS multilabel text classification with correlation-aware label propagation (Wmltc-clp), which estimates accurate pseudolabels by propagating them over a text correlation graph. To evaluate the proposed framework, we conduct extensive experiments on nine benchmark datasets, including five sentiment analysis datasets and four prevalent MLTC datasets. The results demonstrate that Cws-src can generate more confident category words and Wmltc-clp can achieve significant improvements over the WS learning baselines. The maximum performance gains of Wmltc-clp over the best WS learning baseline methods reach 0.096, 0.081, 0.075, and 0.02 on Micro- $F1$ , Macro- $F1$ , average precision (AP), and ranking loss (RL) across all benchmark datasets.

Learning from Label Proportions (LLP) aims to use bags of instances associated with the proportions of each label within the bag to learn an instance-level classifier. Proportion matching is a widely used strategy that aligns the average model outputs of all instances in a bag with the label proportions in order to induce the classifier. However, simply fitting the label proportion may cause over-smoothing problems and does not guarantee correct label prediction of individual instances, resulting in poor classification performance. In this paper, we propose a novel LLP approach that can mitigate the over-smoothing problems with theoretical guarantees. Rather than fitting the label proportions directly, we treat them as targets for an auxiliary proportional value classification task to induce the target classifier. Our approach only requires the incorporation of an aggregation function after the classification layer. We also introduce an efficient computational approach with a divide-and-conquer strategy. Extensive experiments on various image and text benchmark datasets demonstrate that our approach achieves superior performance against state-of-the-art LLP methods.

Time series forecasting (TSF) refers to a fundamental task of predicting future sequential data based on historical observations. One representative category of TSF methods is transformer-based approaches, which translate time series into token sequences (i.e., as raw texts) before applying well-established transformers. In this paper, we conducted extensive preliminary experiments to evaluate their stability against various noises, and we empirically observed an interesting phenomenon: applying noises with appropriate levels to training time series can promote the predictive accuracy of transformer-based TSF methods. We analyze this phenomenon from the perspective of token attention and find one basic reason is that applying noises with appropriate levels can lead to token attention weights wavy, which is more consistent to the basic characteristic of time series data. Motivated by this phenomenon and analysis, we apply perturbations and propose a sub-objective wrt, perturbations constraining token attention weights to be wavy. Accordingly, we propose a novel TSF method, namely Wave-Attention-aware TransformER (WATER). We empirically evaluate \baby across the commonly used benchmark datasets, and experimental results indicate that it consistently outperforms the existing TSF methods.

Multi-Label Learning (MLL) refers to inducing multi-label prediction models from the precisely labeled training dataset. However, in many real-world scenarios, e.g ., crowdsourcing annotations, the training datasets are often only partially valid, where each training instance is associated with a candidate label set, covering ground-truth labels but also with irrelevant ones. Naturally, learning with such datasets, formally referred to as Partial Multi-label Learning (PML), involves many noisy supervised signals, hence imposing a significant challenge to the prediction model induction. To meet this challenge, we purify the noisy supervised signals by formulating the latent label distribution, i.e ., the probability of a candidate label being a ground-truth one, and then jointly learn it with the prediction model by minimizing their regularized Wasserstein distance, i.e ., a robust distance for distributions as well as involving label correlations. Therefore, we propose a novel PML method, namely Wasserstein Partial Multi-Label Learning with dual Label Correlation Perspectives ( Wpml 3 cp ), solved by the gradient descent with an augmented Lagrange multiplier technique. To further enhance the robustness of Wpml 3 cp against exceptionally high ratios of irrelevant labels, we extend it with a Dual-branch Competitive Cleansing mechanism, leading to Wpml 3 cp -D. Besides, we also analyze the generalization error bound and time complexity of Wpml 3 cp and Wpml 3 cp -D. The extensive experiments are constructed by comparing Wpml 3 cp and Wpml 3 cp -D with existing PML baselines across synthetic and real-world datasets, and empirical results demonstrate that Wpml 3 cp and Wpml 3 cp -D can outperform the PML baselines in various noisy levels.

Model-based diagnosis (MBD) with multiple pseudo-normal observations enables the effective detection of latent faults even when some actual observations are consistent with the system’s expected observations. Current state-of-the-art algorithms overlook this diagnostically rich scenario and incorporate too many components, which frequently leads to stack overflow errors in large-scale experiments. To address these challenges, we introduce our algorithm DKIPNO (Diagnosis with Key node and IterativeDFS for Pseudo-Normal Observations) in this paper, which mainly focuses on MBD with multiple pseudo-normal observations and incorporates two original proposed ideas. Firstly, we introduce the novel concept of the ‘key node’. By virtually flipping the outputs of key nodes and comparing flipped observations with the system’s expected observations, we effectively identify more functionally normal components and significantly reduce the number of potentially faulty components in the diagnostic process. Secondly, we present the IterativeDFS method, an original iterative traversal technique that can prevent stack overflow errors commonly encountered with the recursive method used in previous algorithms. We evaluate the performance of DKIPNO using the ISCAS’85 benchmark in our experiment. Experimental results demonstrate that DKIPNO not only significantly reduces the time required for diagnosis but also prevents stack overflow errors, thereby outperforming other state-of-the-art algorithms.

Semi-Supervised Learning (SSL) aims to improve the learning performance of supervised learning with a large number of unlabeled samples. The existing SSL methods such as FixMatch and FlexMatch select unlabeled samples with high-confident pseudo-labels and make consistency constraints between their weak and strong augmentations. Unfortunately, they cannot be applied Semi-Supervised Regression (SSR) because regression predictions can not reflect the confidence of pseudo-labels. To solve this, a recent SSR method RankUp incorporates an auxiliary ranking task by leveraging sample pairs with high-confident pseudo-ranks. In this paper, we upgrade Rankup to a novel SSR method, namely Semi-Supervised Regression by Ranking Close Unlabeled Samples (SSR-RCUS). Its basic idea is reconstructing closed mixup augmented samples with high-confident pseudo-ranks under a monotonicity assumption, and then applying them to the auxiliary ranking task to improve regression performance. We conduct extensive experiments to evaluate the performance of SSR-RCUS on benchmark datasets, and empirical results demonstrate that SSR-RCUS can outperform the existing baselines in various settings, especially when labeled data are scarce.

Semi-Supervised Text Classification (SSTC) mainly works under the spirit of self-training. They initialize the deep classifier by training over labeled texts; and then alternatively predict unlabeled texts as their pseudo-labels and train the deep classifier over the mixture of labeled and pseudo-labeled texts. Naturally, their performance is largely affected by the accuracy of pseudo-labels for unlabeled texts. Unfortunately, they often suffer from low accuracy because of the margin bias problem caused by the large difference between representation distributions of labels in SSTC. To alleviate this problem, we apply the angular margin loss, and perform several Gaussian linear transformations to achieve balanced label angle variances, i.e., the variance of label angles of texts within the same label. More accuracy of predicted pseudo-labels can be achieved by constraining all label angle variances balanced, where they are estimated over both labeled and pseudo-labeled texts during self-training loops. With this insight, we propose a novel SSTC method, namely Semi-Supervised Text Classification with Balanced Deep representation Distributions (S2TC-BDD). We implement both multi-class classification and multi-label classification versions of S2TC-BDD by introducing some pseudo-labeling tricks and regularization terms. To evaluate S2 TC-BDD, we compare it against the state-of-the-art SSTC methods. Empirical results demonstrate the effectiveness of S2 TC-BDD, especially when the labeled texts are scarce.

Learning from Label Proportions (LLP) is a weakly supervised problem in which the training data comprise bags, that is, groups of instances, each annotated only with bag-level class label proportions, and the objective is to learn a classifier that predicts instance-level labels. This setting is widely applicable when privacy constraints limit access to instance-level annotations or when fine-grained labeling is costly or impractical. In this work, we introduce a method that leverages Dual proportion Constraints (LLP-DC) during training, enforcing them at both the bag and instance levels. Specifically, the bag-level training aligns the mean prediction with the given proportion, and the instance-level training aligns hard pseudo-labels that satisfy the proportion constraint, where a minimum-cost maximum-flow algorithm is used to generate hard pseudo-labels. Extensive experimental results across various benchmark datasets empirically validate that LLP-DC consistently improves over previous LLP methods across datasets and bag sizes. The code is publicly available at https://github.com/TianhaoMa5/CV PR2026_Findings_LLP_DC.

Data augmentation is an intuitive solution to increase the diversity of training instances in the machine learning community. Mixup is acknowledged as an effective and efficient mix-based data augmentation method, following a linear alignment assumption that the linear interpolations of features align the corresponding linear interpolations of labels. Unfortunately, this assumption can be violated in many complex scenarios, resulting in augmented instances with noisy labels, especially for regression problems. To solve this problem, we propose an easy-to-implement mixup method, namely DEnosing MIXUP (DE-mixup), which iteratively corrects the noisy response targets by leveraging an auxiliary noise estimation task with mixup deep features. Additionally, we suggest an efficient optimization method with alternating direction method of multipliers. We compare DE-mixup with the existing mixup variants and other prevalent data augmentation methods across benchmark regression datasets. Empirical results indicate the effectiveness of DE-mixup under the in-distribution and out-of-distribution cases.

The rapid spread of misinformation on social media platforms has become a formidable challenge. To mitigate its proliferation, Misinformation Detection (MD) has emerged as a critical research topic. Traditional MD approaches based on small models typically perform binary classification through a black-box process. Recently, the rise of Large Language Models (LLMs) has enabled explainable MD, where models generate rationales that explain their decisions, thereby enhancing transparency. Existing explainable MD methods primarily focus on crafting sophisticated prompts to elicit rationales from off-the-shelf LLMs. In this work, we propose a pipeline to fine-tune a dedicated LLM specifically for explainable MD. Our pipeline begins by collecting large-scale fact-checked articles, and then uses multiple strong LLMs to produce veracity predictions and rationales. To ensure high-quality training data, we leverage a filtering strategy that selects only the correct instances for fine-tuning. While this pipeline is intuitive and prevalent, our experiments reveal that naive filtering based solely on label correctness is insufficient in practice and suffers from two critical limitations: (1) Coarse-grained labels cause insufficient rationales: Rationales filtered solely based on binary labels are insufficient to adequately support their decisions; (2) Over-verification behavior causes unnecessary rationales: Stronger LLMs tend to exhibit over-verification behavior, producing excessively verbose and unnecessary rationales. To address these issues, we introduce LONSREX, a novel data synthesis pipeline to Locate Necessary and Sufficient Rationales for Explainable MD. Specifically, we propose a metric that quantifies the contribution of each verification step to the final prediction, thereby evaluating its necessity and sufficiency. Experimental results demonstrate the effectiveness of LONSREX.

clustering nowadays has proven to significantly surpass the classical clustering method, so it has been widely used in diverse applications. One current branch of deep clustering methods enhances the primary task through auxiliary tasks, among which the most prevalent is over-clustering, i.e., jointly training clustering with different numbers of clusters in a multi-task manner. However, existing approaches typically treat these auxiliary tasks in isolation and neglect the inherent correlations among their cluster assignments. In this paper, we interpret the cluster assignment memberships of samples generated by all clustering tasks as correlated pseudo-labels. Motivated by this observation, we propose to explicitly exploit such correlation knowledge to improve clustering performance. To achieve this, we can formulate the collection of samples with pseudo-labels as a pseudo-multi-label learning problem, and solve it by employing any off-the-shelf multi-label learning methods which enable to capture correlations between pseudo-labels. Based on this idea, beyond the clustering tasks, we propose a correlation learning auxiliary task, namely Self-supervised Multi-Label Learning (SMLL); and we then specify a novel deep clustering method with SMLL, namely DCSL3. We conduct several experiments to examine the performance of DCSL3 on benchmark datasets. Empirical results demonstrate the superiority of DCSL3 over the existing deep clustering baseline methods.

Offline reinforcement learning (RL) is a data-driven paradigm that learns policies from static datasets without real-time interacting with the environment. However, action data collected from real-world are often incomplete due to issues such as sensor failures or communication disruptions, which can significantly impair the performance of offline RL. We focus on the dimension-specific missing action data problem (DSMADP) and utilize such expensive yet incomplete action data to enhance offline RL. Inspired by the coordinated nature of joint movements in physical systems, we propose that intrinsic correlations exist across dimensions within each action example—referred to as intra-example inter-dimension correlations. Based on this insight, we propose an effective MLP-based CORrelation-driven IMPutation model named CORIMP. It models the correlations by learning mappings from observed to missing action dimensions, which then guides the imputation of missing values using available data. Theoretically, we bound CORIMP’s imputation error and its downstream impact on offline RL performance. Experimental results on variants of missing D4RL datasets demonstrate the effectiveness of our method. Notably, with the TD3BC algorithm, the CORIMP-imputed dataset achieves 95.15% of the Halfcheetah-medium-expert dataset performance (oracle). It provides an average improvement of 99.12% over zero-filled datasets with missing ratios from 0.1 to 0.9 across two dimensions.