论文检索

Aojun Lu, Tao Feng, Hangjie Yuan, Chunhui Ding, Yanan Sun

Continual Learning (CL) seeks to enable neural networks to incrementally acquire new knowledge (plasticity) while retaining existing knowledge (stability). Although pre-trained models (PTMs) have provided a strong foundation for CL, existing approaches face a fundamental challenge in balancing these two competing objectives. Current methods typically address stability by freezing the PTM backbone, which severely limits the model's plasticity, particularly when incoming data distribution diverges largely from the pre-training data. Alternatively, sequentially fine-tuning the entire PTM can adapt to new knowledge but often leads to catastrophic forgetting, highlighting the critical stability-plasticity trade-off in PTM-based CL. To address this limitation, we propose Adapting PTMs before the core CL process (ACL), a novel framework that introduces a plug-and-play adaptation phase prior to learning each new task. During this phase, ACL refines the PTM backbone by aligning embeddings with their original class prototypes while distancing them from irrelevant classes. This mechanism theoretically and empirically demonstrates desirable balance between stability and plasticity, significantly improving CL performance across benchmarks and integrated methods.

Zhaowei Liu, Yihao Jiang, Rufei Gao, Jinglei Liu, Dong Yang

Graph Neural Networks (GNNs) have demonstrated superior performance in processing centralized graph-structured data. However, real-world privacy and security concerns hinder data centralization and shareing, leading to severe data isolation (data silos). While Federated Learning (FL) offers a distributed solution to mitigate these obstacles, existing Federated Graph Neural Network (FedGNN) frameworks struggle to effectively address data heterogeneity. To address this, this paper proposes DA-DFGAS, a federated graph neural architecture search algorithm. Specifically, DA-DFGAS facilitates model personalization via a directed tree topology and path constraint mechanisms, while simultaneously employing a joint self-attention mechanism based on predicted probability distributions to capture distributional variations across multiple clients. Furthermore, it integrates a bi-level global-local objective optimization strategy to ensure global model consistency while preserving local adaptability. Experimental results on multiple datasets demonstrate that DA-DFGAS outperforms state-of-the-art methods, achieving 0.5–3.0% accuracy improvements over centralized baselines and 0.5–5.0% over federated counterparts.

Dohun Lim, Minji Kim, Jaewoon Lim, Sungchan Kim

We propose BRIC, a novel test-time adaptation (TTA) framework that enables long-term human motion generation by resolving execution discrepancies between diffusion-based kinematic motion planners and reinforcement learning-based physics controllers. While diffusion models can generate diverse and expressive motions conditioned on text and scene context, they often produce physically implausible outputs, leading to execution drift during simulation. To address this, BRIC dynamically adapts the physics controller to noisy motion plans at test time, while preserving pre-trained skills via a loss function that mitigates catastrophic forgetting. In addition, BRIC introduces a lightweight test-time guidance mechanism that steers the diffusion model in the signal space without updating its parameters. By combining both adaptation strategies, BRIC ensures consistent and physically plausible long-term executions across diverse environments in an effective and efficient manner. We validate the effectiveness of BRIC on a variety of long-term tasks, including motion composition, obstacle avoidance, and human-scene interaction, achieving state-of-the-art performance across all tasks.

You Li, Guannan Zhao, Yuhao Ju, Yunqi He, Jie Gu, Hai Zhou

We introduce LLA, an effective intellectual property (IP) protection scheme for generative AI models. LLA leverages the synergy between hardware and software to defend against various supply chain threats, including model theft, model corruption, and information leakage. On the software side, it embeds key bits into neurons that can trigger outliers to degrade performance and applies invariance transformations to obscure the key values. On the hardware side, it integrates a lightweight locking module into the AI accelerator while maintaining compatibility with various dataflow patterns and toolchains. An accelerator with a pre-stored secret key acts as a license to access the model services provided by the IP owner. The evaluation results show that LLA can withstand a broad range of oracle-guided key optimization attacks, while incurring a minimal computational overhead of less than 0.1% for 7,168 key bits.

Lujun Li, Qiyuan Zhu, Jiacheng Wang, Xiaoyu Qin, Wei Li, Hao Gu, Sirui Han, Yike Guo

Mixture of Experts (MoE) LLMs face significant obstacles due to their massive parameter scale, which imposes memory, storage, and deployment challenges. Although recent expert merging methods aim to achieve greater efficiency by consolidating several experts, they are fundamentally hindered by parameter conflicts arising from expert specialization. In this paper, we present Sub-MoE, a novel MoE compression framework via Subspace Expert Merging. Our key insight is to perform joint Singular Value Decomposition (SVD) on concatenated expert weights, reducing conflicting parameters by extracting shared U-matrices while enabling effective merging of the expert-specific V components. Specifically, Sub-MoE consists of two innovative stages: (1) Adaptive Expert Clustering, which groups functionally coherent experts via K-means clustering based on cosine similarity of expert outputs; and (2) Subspace Expert Merging, which first performs Experts Union Decomposition to derive the shared U-matrix across experts in the same group, then applies frequency-based merging for individual V-matrices, and completes expert reconstruction using the merged V-matrix. In this way, we align and fuse experts in a shared subspace. Additionally, the framework can be extended with intra-expert compression for further inference optimization. Extensive experiments on Mixtral, DeepSeek, and Qwen-1.5/3 MoE LLMs demonstrate that our Sub-MoE significantly outperforms existing expert pruning and merging methods. Notably, our Sub-MoE maintains 96%/86% of original performance with 25%/50% expert reduction on Mixtral-8×7B in zero-shot benchmarks.

Lin Li, Yan Wang, Zhuopeng Wang

The Mixture-of-Experts (MoE) architecture has emerged as a promising paradigm for scaling large language models (LLMs) by activating only a sparse subset of experts per input. However, its massive parameter size remains a major obstacle to efficient deployment. Existing pruning methods often ignore two key aspects: the intricate structural dependencies among experts and the heterogeneous importance of different layers. To tackle these issues, we propose C-GNN-PRUNE, a unified and structure-aware compression framework tailored for MoE models. Our method introduces an EntropyGuided Allocation Module that dynamically assigns pruning budgets by leveraging expert activation entropy, enabling adaptive handling of inter-layer heterogeneity. To preserve structural collaboration patterns, we construct an expert interaction graph that fuses functional similarity and routing behavior, and employ a GNN-Based Embedding Module to learn structure-aware expert representations. These embeddings, along with co-activation patterns, are fed into a Community Detection Module to identify expert clusters for structured pruning. Finally, an Activation-Aware Selection Module retains the most critical experts in each community, balancing sparsity and expressiveness. Experiments on multiple open-source MoE models demonstrate that C-GNN-PRUNE consistently outperforms prior methods under various pruning ratios, achieving better trade-offs between compression and accuracy. This framework provides a modular and effective solution for structure-preserving compression of large-scale MoE models.

Jiacheng Li, Songhe Feng

Test-time adaptation (TTA) enables online model adaptation using only unlabeled test data, aiming to bridge the gap between source and target distributions. However, in multimodal scenarios, varying degrees of distribution shift across different modalities give rise to a complex coupling effect of unimodal shallow feature shift and cross-modal high-level semantic misalignment, posing a major obstacle to extending existing TTA methods to the multimodal field. To address this challenge, we propose a novel multimodal test-time adaptation (MMTTA) framework, termed as Bridging Modalities via Progressive Re-alignment (BriMPR). BriMPR, consisting of two progressively enhanced modules, tackles the coupling effect with a divide-and-conquer strategy. Specifically, we first decompose MMTTA into multiple unimodal feature alignment sub-problems. By leveraging the strong function approximation ability of prompt tuning, we calibrate the unimodal global feature distributions to their respective source distributions, so as to achieve the initial semantic re-alignment across modalities. Subsequently, we assign the credible pseudo-labels to combinations of masked and complete modalities, and introduce inter-modal instance-wise contrastive learning to further enhance the information interaction among modalities and refine the alignment. Extensive experiments on MMTTA tasks, including both corruption-based and real-world domain shift benchmarks, demonstrate the superiority of our method.

Tianyu Guo, Shanwei Zhao, Shiai Zhu, Chenguang Ma

Deploying Vision-Language Models (VLMs) on edge devices (e.g., smartphones and robots) is crucial for enabling low-latency and privacy-preserving intelligent applications. Given the resource constraints of these devices, quantization offers a promising solution by improving memory efficiency and reducing bandwidth requirements, thereby facilitating the deployment of VLMs. However, existing research has rarely explored aggressive quantization on VLMs, particularly for the models ranging from 1B to 2B parameters, which are more suitable for resource-constrained edge devices. In this paper, we propose SPEED-Q, a novel Staged Processing with EnhancEd Distillation framework for VLM low-bit weight-only quantization that systematically addresses the following two critical obstacles: (1) significant discrepancies in quantization sensitivity between vision (ViT) and language (LLM) components in VLMs; (2) training instability arising from the reduced numerical precision inherent in low-bit quantization. In SPEED-Q, a staged sensitivity adaptive mechanism is introduced to effectively harmonize performance across different modalities. We further propose a distillation-enhanced quantization strategy to stabilize the training process and reduce data dependence. Together, SPEED-Q enables accurate, stable, and data-efficient quantization of complex VLMs. SPEED-Q is the first framework tailored for quantizing entire small-scale billion-parameter VLMs to low bits. Extensive experiments across multiple benchmarks demonstrate that SPEED-Q achieves up to 6x higher accuracy than existing quantization methods under 2-bit settings and consistently outperforms prior on-device VLMs under both 2-bit and 4-bit settings.

Mingrong Gong, Chaoqi Chen, Luyao Tang, Yuxi Wang, Sergio Escalera

4D point cloud segmentation is crucial for autonomous driving with continuous LiDAR streams. While test-time adaptation (TTA) is the standard approach for handling dynamic environments, current methods suffer from catastrophic error accumulation due to over-reliance on pseudo-labels. Active learning could provide reliable annotations for critical samples, but combining it with TTA faces severe challenges: realtime processing requirements and expensive 3D labeling costs. In this paper, we propose ATTA-4DSeg, the first framework to achieve efficient active test-time adaptation for 4D point cloud segmentation under extreme budget constraints. Our key insight is a self-reinforcing loop: oracle annotations refine adaptation prototypes, which then guide the selection of subsequent high-value samples from regions with severe distribution shifts, maximizing each annotation’s impact. Specifically, we propose three key innovations: (1) dual-prototype comparison that precisely localizes distribution shift boundaries to narrow annotation scope, (2) Class-Inverse Budget Allocation (CIBA) ensuring balanced adaptation across all categories, coupled with hybrid uncertainty scoring combining voxel-level geometry and point-wise variance for optimal sample selection, and (3) a refinement strategy leveraging sparse oracle annotations to improve predictions on unlabeled points, maximizing annotation utility. Extensive experiments show ATTA-4DSeg improves mIoU by 18.87%, 19.92%, and 3.6% on three domain adaptation benchmarks using only 1% annotation budget. Our method operates 2.28× faster than state-of-the-art methods. Remarkably, our approach reaches 90% of fully-supervised performance using only 5% annotation budget.

Christian Fiedler

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.

Clément Elliker, Jesse Read, Sonia Vanier, Albert Bifet

Reliable prediction of train delays is essential for enhancing the robustness and efficiency of railway transportation systems. In this work, we reframe delay forecasting as a stochastic simulation task, modeling state-transition dynamics through imitation learning. We introduce Drift-Corrected Imitation Learning (DCIL), a novel self-supervised algorithm that extends DAgger by incorporating distance-based drift correction, thereby mitigating covariate shift during rollouts without requiring access to an external oracle or adversarial schemes. Our approach synthesizes the dynamical fidelity of event-driven models with the representational capacity of data-driven methods, enabling uncertainty-aware forecasting via Monte Carlo simulation. We evaluate DCIL using a comprehensive real-world dataset from \textsc{Infrabel}, the Belgian railway infrastructure manager, which encompasses over three million train movements. Our results, focused on predictions up to 30 minutes ahead, demonstrate superior predictive performance of DCIL over traditional regression models and behavioral cloning on deep learning architectures, highlighting its effectiveness in capturing the sequential and uncertain nature of delay propagation in large-scale networks.

Henggang Deng, Yuchao Tang, Wenjie Fu, Huandong Wang, Kun Chen, Tao Jiang

In real-world time-series modelling, graph structures are widely adopted because they explicitly encode node topology and capture complex network dynamics. In practice, however, a complete graph is often partitioned across multiple parties; each party can access only its local sub-graph and, owing to privacy regulations, cannot share topology or data, creating pervasive data silos. Federated Graph Learning (FGL) offers a privacy-preserving collaborative-learning paradigm, yet current methods still face two key challenges: (1) the graph topology itself contains sensitive structural information, which can lead to privacy leakage if directly shared during FGL; (2) cross-party edges are crucial for accurate modeling, yet exploiting them without compromising privacy remains a significant challenge. To overcome these obstacles, we propose FedSkeleton, a privacy-preserving framework for time-series prediction that comprises a Skeleton Construction Module and a Dual-stream Forecasting Module, enabling global dependency capture without revealing the topology. Extensive experiments show that FedSkeleton consistently outperforms existing baselines and even surpasses models trained in a centralized setting with full-graph access in certain cases. In addition, we conduct comprehensive security analysis, communication-cost evaluation and scalability experiments, demonstrating that FedSkeleton effectively resists common attacks, keeps communication overhead manageable, and remains robust with respect to key hyper-parameters and the number of participating parties.

Zihao Xu, Junchen Ding, Yiling Lou, Kun Zhang, Dong Gong, Yuekang Li

Large Language Models (LLMs) have achieved significant progress in language understanding and reasoning. Evaluating and analyzing their logical reasoning abilities has therefore become essential. However, existing datasets and benchmarks are often limited to overly simplistic, unnatural, or contextually constrained examples. In response to the growing demand, we introduce SMARTYPAT-BENCH, a challenging, naturally expressed, and systematically labeled benchmark derived from real-world high-quality Reddit posts containing subtle logical fallacies. Unlike existing datasets and benchmarks, it provides more detailed annotations of logical fallacies and features more diverse data. To further scale up the study and address the limitations of manual data collection and labeling, such as fallacy-type imbalance and labor-intensive annotation, we introduce SMARTYPAT, an automated framework powered by logic programming-based oracles. SMARTYPAT utilizes Prolog rules to systematically generate logically fallacious statements, which are then refined into fluent natural language sentences by LLMs, ensuring precise fallacy rep- resentation. Extensive evaluation demonstrates that SMARTYPAT produces fallacies comparable in subtlety and quality to human-generated content and significantly outperforms baseline methods. Finally, experiments reveal insights into LLM capabilities, highlighting that while excessive reasoning steps hinder fallacy detection accuracy, structured reasoning enhances fallacy categorization performance.

Enjun Du, Siyi Liu, Yongqi Zhang

Knowledge graph reasoning in the fully-inductive setting—where both entities and relations at test time are unseen during training—remains an open challenge. In this work, we introduce GraphOracle, a novel framework that achieves robust fully-inductive reasoning by transforming each knowledge graph into a Relation-Dependency Graph (RDG). The RDG encodes directed precedence links between relations, capturing essential compositional patterns while drastically reducing graph density. Conditioned on a query relation, a multi-head attention mechanism propagates information over the RDG to produce context-aware relation embeddings. These embeddings then guide a second GNN to perform inductive message passing over the original knowledge graph, enabling prediction on entirely new entities and relations. Comprehensive experiments on 60 benchmarks demonstrate that GraphOracle outperforms prior methods by up to 25% in fully-inductive and 28% in cross-domain scenarios. Our analysis further confirms that the compact RDG structure and attention-based propagation are key to efficient and accurate generalization

Zhuoyuan Yu, Yuxing Long, Zihan Yang, Chengyan Zeng, Hongwei Fan, Jiyao Zhang, Hao Dong

Existing vision-and-language navigation models often deviate from the correct trajectory when executing instructions. However, these models lack effective error correction capability, hindering their recovery from errors. To address this challenge, we propose Self-correction Flywheel, a novel post-training paradigm. Instead of considering the model’s error trajectories on the training set as a drawback, our paradigm emphasizes their significance as a valuable data source. We have developed a method to identify deviations in these error trajectories and devised innovative techniques to automatically generate self-correction data for perception and action. These self-correction data serve as fuel to power the model’s continued training. The brilliance of our paradigm is revealed when we re-evaluate the model on the training set, uncovering new error trajectories. At this time, the self-correction flywheel begins to spin. Through multiple flywheel iterations, we progressively enhance our monocular RGB-based VLA navigation model CorrectNav. Experiments on R2R-CE and RxR-CE benchmarks show CorrectNav achieves new state-of-the-art success rates of 65.1% and 69.3%, surpassing prior best VLA navigation models by 8.2% and 16.4%. Real robot tests in various indoor and outdoor environments demonstrate \method's superior capability of error correction, dynamic obstacle avoidance, and long instruction following.

Tzu-Jung Lin, Jia-Fong Yeh, Hung-Ting Su, Chung-Yi Lin, Yi-Ting Chen, Winston H. Hsu

In open-vocabulary mobile manipulation (OVMM), task success often hinges on the selection of an appropriate base placement for the robot. Existing approaches typically navigate to proximity-based regions without considering affordances, resulting in frequent manipulation failures. We propose Affordance-Guided Coarse-to-Fine Exploration, a zero-shot framework for base placement that integrates semantic understanding from vision-language models (VLMs) with geometric feasibility through an iterative optimization process. Our method constructs cross-modal representations, namely Affordance RGB and Obstacle Map+, to align semantics with spatial context. This enables reasoning that extends beyond the egocentric limitations of RGB perception. To ensure interaction is guided by task-relevant affordances, we leverage coarse semantic priors from VLMs to guide the search toward task-relevant regions and refine placements with geometric constraints, thereby reducing the risk of convergence to local optima. Evaluated on five diverse open-vocabulary mobile manipulation tasks, our system achieves an 85% success rate, significantly outperforming classical geometric planners and VLM-based methods. This demonstrates the promise of affordance-aware and multimodal reasoning for generalizable, instruction-conditioned planning in OVMM.

Dun Dai, Ze Lu, Xunhua Dai, Quan Quan

Ultra-low altitude UAVs (below 120 meters) are gaining importance in the booming low-altitude economy, where GNSS signals are often unreliable or unavailable. Vision-based localization emerges as a promising alternative; however, existing benchmarks are not designed for ultra-low flight and typically adopt pinhole cameras with limited field of view, making them less effective in handling occlusions and repetitive textures near the ground. To address these limitations, we introduce the first panoramic UAV localization dataset tailored for ultra-low altitude scenarios. Built on a four-fisheye-camera system in the high-fidelity RflySim platform, our dataset captures diverse conditions — including day/night cycles, extreme weather, and dynamic obstacles — and contains over hundreds of thousands of frames. It is further enhanced with real-world UAV panoramic data to narrow the sim-to-real gap and will be continuously updated for broader applicability. Comprehensive experiments confirm the effectiveness and transferability of our dataset, establishing it as a robust benchmark for future research in vision-based UAV localization.

Paul Gölz, Jan Maly, Ulrike Schmidt-Kraepelin, Markus Utke, Philipp C. Verpoort

In citizens' assemblies, a group of constituents is randomly selected to weigh in on policy issues. We study a two-stage sampling problem faced by practitioners in countries such as Germany, in which constituents' contact information is stored at a municipal level. As a result, practitioners can only select constituents from a bounded number of cities ex post, while ensuring equal selection probability for constituents ex ante. We develop several algorithms for this problem. Although minimizing the number of contacted cities is NP-hard, we provide a pseudo-polynomial time algorithm and an additive 1-approximation, both based on separation oracles for a linear programming formulation. Recognizing that practical objectives go beyond minimizing city count, we further introduce a simple and more interpretable greedy algorithm, which additionally satisfies an ex-post monotonicity property and achieves an additive 2-approximation. Finally, we explore a notion of ex-post proportionality, for which we propose two practical algorithms: an optimal algorithm based on column generation and integer linear programming and a simple heuristic creating particularly transparent distributions. We evaluate these algorithms on data from Germany, and plan to deploy them in cooperation with a leading nonprofit organization in this space.

Adam Dziwoki, Rostislav Horcik

This paper investigates the impact of perturbations on the best-response-based algorithms approximating Nash equilibria in zero-sum games, namely Double Oracle and Fictitious Play. More precisely, we assume that the oracle computing the best responses perturbs the utilities before selecting the best response. We show that using such an oracle reduces the number of iterations for both algorithms. For some cases, suitable perturbations ensure the expected number of iterations is logarithmic. Although the utility perturbation is computationally demanding as it requires iterating through all pure strategies, we demonstrate that one can efficiently perturb the utilities in games where pure strategies have further inner structure.

Jianghong Ma, Changran He, Dezhao Yang, Tianjun Wei, Haijun Zhang, Xiaofeng Zhang

The sparsity of user–item interactions remains a fundamental obstacle in collaborative filtering, limiting the ability of Graph Neural Network (GNN)-based recommender systems to capture high-order user relationships without incurring over-smoothing and computational overhead. Existing social recommendation approaches mitigate this by incorporating social networks, yet most rely on explicit ties and fail to construct informative links in their absence. Meanwhile, contrastive learning (CL) has shown promise in improving representation quality, but current view generation strategies, augmentation-based for robustness and nonaugmentation-based for semantic fidelity, are seldom combined, leaving their complementary potential underexplored. We propose Social Generating with Multiview-guided Tuning (SGMT), a unified framework that addresses both challenges. First, an interest-aware social generation mechanism constructs synthetic user–user links from shared interaction patterns, theoretically shown to compress collaborative paths and uncover latent high-order relations. Second, we present two complementary CL modules, Noise-augmented View and Semantic-explored View, which we theoretically prove to preferentially enhance uniformity and alignment, respectively, two fundamental objectives in CL. Experiments on three real-world datasets show that SGMT outperforms state-of-the-art baselines, validating both the theoretical analysis and the practical efficacy of our model.