Naen Xu, Jiayi Sheng, Changjiang Li, Chunyi Zhou, Yuyuan Li, Tianyu Du, Jun Wang, Zhihui Fu, Jinbao Li, Shouling Ji. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Rui Yin, Tianxu Han, Naen Xu, Changjiang Li, Ping He, Chunyi Zhou, Jun Wang, Zhihui Fu, Tianyu Du, Jinbao Li, Shouling Ji. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Federated graph learning (FGL) has emerged as a promising approach to enable collaborative training of graph models while preserving data privacy. However, current FGL methods overlook the out-of-distribution (OOD) shifts that occur in real-world scenarios. The distribution shifts between training and testing datasets in each client impact the FGL performance. To address this issue, we propose federated graph OOD generalization framework FedGOG, which includes two modules, i.e., diffusion data exploration (DDE) and latent embedding decorrelation (LED). In DDE, all clients jointly train score models to accurately estimate the global graph data distribution and sufficiently explore sample space using score-based graph diffusion with conditional generation. In LED, each client models a global invariant GNN and a personalized spurious GNN. LED aims to decorrelate spuriousness from invariant relationships by minimizing the mutual information between two categories of latent embeddings from different GNN models. Extensive experiments on six benchmark datasets demonstrate the superiority of FedGOG.

Personalized recommender systems are critical for enhancing user engagement across a range of digital platforms. However, conventional approaches rely heavily on centralized data collection, raising significant privacy concerns. Federated recommender systems (PFRS) address these concerns by decentralizing model training, ensuring user data privacy. Despite the progress, existing methods still struggle with capturing the multi-faceted nature of user and transferring global knowledge effectively. In this work, we propose FedMUR, a novel federated recommendation framework that models user representation as a Gaussian mixture distribution, capturing users' multi-faceted characteristics. Each Gaussian component corresponds to a distinct interest facet, with adaptive mixture weights representing the user's preference intensity toward each facet. To facilitate knowledge transfer, FedMUR constructs global consistent prototypes that encode shared behavioral trends across users via popularity-weighted optimal transport. These prototypes enhance local models by injecting global shared patterns into personalized representation learning. Extensive experiments across several real-world datasets demonstrate that FedMUR significantly outperforms existing state-of-the-art federated recommendation systems.

Diffusion models have demonstrated superior performance in image generation tasks, thus becoming the mainstream model for generative visual tasks. Diffusion models need to execute multiple timesteps sequentially, resulting in a dramatic increase in workload. Existing accelerators leverage the data similarity between adjacent timesteps and perform mixed-precision differential quantization to accelerate diffusion models. However, merging differential values with raw inputs in each layer of each timestep to ensure computational correctness requires significant memory access for loading raw inputs, which creates a heavy memory burden. Moreover, mixed-precision computations may lead to low hardware utilization if not well designed. Unlike these works, we propose MHDiff, a tailored framework that identifies the focal pixels at the first layer and finetunes them to fit all layers, then represents focal pixels with high-precision while using low-precision for others, thereby accelerating diffusion models while minimizing memory burden. To improve hardware utilization, MHDiff employs a packing module that merges low-precision values into high-precision values to create full high-precision matrices and designs a processing element (PE) array to efficiently process the packed matrices. Extensive experiment results demonstrate that MHDiff can achieve satisfactory performance with negligible quality loss.

Vision Transformers (ViTs) have demonstrated remarkable performance in computer vision tasks by effectively extracting global features. However, their self-attention mechanism suffers from quadratic time and memory complexity as image resolution or video duration increases, leading to inefficiency on GPUs. To accelerate ViTs, existing works mainly focus on pruning tokens based on value-level sparsity. However, they miss the chance to achieve peak performance as they overlook the bit-level sparsity. Instead, we propose Inter-token Bit-sparsity Awareness (IBA) algorithm to accelerate ViTs by exploring bit-sparsity from similar tokens. Next, we implement IBA on GPUs that synergize CUDA and Tensor Cores by addressing two issues: firstly, the bandwidth congestion of the Register File hinders the parallel ability of CUDA and Tensor Cores. Secondly, due to the varying exponent of floating-point vectors, it is hard to accelerate bitsparse matrix multiplication and accumulation (MMA) in Tensor Core through fixed-point-based bit-level circuits. Therefore, we present SynGPU, an algorithm-hardware co-design framework, to accelerate ViTs. SynGPU enhances data reuse by a novel data mapping to enable full parallelism of CUDA and Tensor Cores. Moreover, it introduces Bit-Serial Tensor Core (BSTC) that supports fixed- and floating-point MMA by combining the fixedpoint Bit-Serial Dot Product (BSDP) and exponent alignment techniques. Extensive experiments show that SynGPU achieves an average of $2.15 \times \sim 3.95 \times$ speedup and $2.49 \times \sim 3.81 \times$ compute density over A100 GPU.

The advent of edge intelligence and escalating concerns for data privacy protection have sparked a surge of interest in device-cloud collaborative computing. Large-scale device deployments to validate prototype solutions are often prohibitively expensive and practically challenging, resulting in a pronounced demand for simulation tools that can emulate real-world scenarios. However, existing simulators predominantly rely solely on high-performance servers to emulate edge computing devices, overlooking (1) the discrepancies between virtual computing units and actual heterogeneous computing devices and (2) the simulation of device behaviors in real-world environments. In this paper, we propose a high-fidelity device simulation platform, called SimDC, which uses a hybrid heterogeneous resource and integrates high-performance servers and physical mobile phones. Utilizing this platform, developers can simulate numerous devices for functional testing cost-effectively and capture precise operational responses from varied real devices. To simulate real behaviors of heterogeneous devices, we offer a configurable device behavior traffic controller that dispatches results on devices to the cloud using a user-defined operation strategy. Comprehensive experiments on the public dataset show the effectiveness of our simulation platform and its great potential for application.1

Federated Recommendation (FedRec) enables joint training across a large number of clients without centralizing user interaction data. However, existing FedRec methods overlook two key challenges, i.e. (1) sufficiently explore the global item embedding space, and (2) effectively achieve local and global collaboration. The former is caused by client sparsity, which leads to suboptimal item embeddings and subsequently impacts the global item embedding in both the dimension and sample views. The latter arises from the lack of modeling the relative importance of local and global contributions to personalized user preferences. To address the above challenges, we propose FedIAR which contains two modules, i.e., item embedding dual-view exploration and adaptive local-global fusion. The first module enhances the global item embedding by reducing redundancy in the dimension view and capturing latent item relationships in the sample view, improving representational capacity. The second module enables the adaptive fusion of local and global item embeddings based on the user preference representation, achieving personalized optimum for recommendation. Extensive experiments on six datasets demonstrate the effectiveness of FedIAR in improving federated recommendation performance.

Federated graph learning involves training graph neural networks distributively on local graphs and aggregating model parameters in a central server. However, existing methods fail to effectively capture and leverage the inherent global structures, hindering local structural modeling. To address this, we propose Federated Graph Factorization (FedGF), which enhances structural knowledge via privacy-preserving graph factorization. Specifically, FedGF includes three modules, i.e., global structure reconstruction (GSR), local structure exploration (LSE), and global-local structure alignment (GLSA). Firstly, GSR factorizes client graphs into a series of learnable graph atoms and conducts reconstruction to capture the globally shared structure. Then, LSE explores the local structure, mining potential but unrevealed connections within client subgraphs. GLSA further aligns the global and local structure to alternatively refine the graph atoms and GNN model, enhancing the overall structural modeling. Extensive experiments on six datasets consistently validate the effectiveness of \modelname.

Cross-Domain Recommendation (CDR) has been widely investi- gated for solving long-standing data sparsity problem via knowl- edge sharing across domains. In this paper, we focus on the Multi- Modal Cross-Domain Recommendation (MMCDR) problem where different items have multi-modal information while few users are overlapped across domains. MMCDR is particularly challenging in two aspects: fully exploiting diverse multi-modal information within each domain and leveraging useful knowledge transfer across domains. However, previous methods fail to cluster items with similar characteristics while filtering out inherit noises within different modalities, hurdling the model performance. What is worse, conventional CDR models primarily rely on overlapped users for domain adaptation, making them ill-equipped to handle scenarios where the majority of users are non-overlapped. To fill this gap, we propose Joint Similarity Item Exploration and Overlapped User Guidance (SIEOUG) for solving the MMCDR problem. SIEOUG first proposes similarity item exploration module, which not only obtains pair-wise and group-wise item-item graph knowledge, but also reduces irrelevant noise for multi-modal modeling. Then SIEOUG proposes user-item collaborative filtering module to aggregate user/item embeddings with the attention mechanism for collaborative filtering. Finally SIEOUG proposes overlapped user guidance module with optimal user matching for knowledge sharing across domains. Our empirical study on Amazon dataset with several different tasks demonstrates that SIEOUG significantly outperforms the state-of-the-art models under the MMCDR setting.

Federated learning breaks through the barrier of data owners by allowing them to collaboratively train a federated machine learning model without compromising the privacy of their own data. However, Federation Learning also faces the threat of poisoning attacks, especially from the client model updates, which may impair the accuracy of the global model. To defend against the poisoning attacks, previous work aims to identify the malicious updates in high dimensional spaces. However, we find that the distances in high dimensional spaces cannot identify the changes in a small subset of dimensions, and the small changes may affect the global models severely. Based on this finding, we propose an untargeted poisoning attack under the federated learning setting via the partial perturbations on a small subset of the carefully selected model parameters, and present two attack object selection strategies. We experimentally demonstrate that the proposed attack scheme achieves high attack success rate on five state-of-the-art defense schemes. Furthermore, the proposed attack scheme remains effective at low malicious client ratios and still circumvents three defense schemes with a malicious client ratio as low as 2%.

This paper presents a transductive multicomponent video segmentation algorithm, which is capable of segmenting the predefined object of interest in the frames of a video sequence. To ensure temporal consistency, a temporal coherent parametric min-cut algorithm is developed to generate segmentation hypotheses based on visual cues and motion cues. Furthermore, each hypothesis is evaluated by an energy function from foreground resemblance, foreground/background divergence, boundary strength, and visual saliency. In particular, the state-of-the-art R-convolutional neural network descriptor is leveraged to encode the visual appearance of the foreground object. Finally, the optimal segmentation of the frame can be attained by assembling the segmentation hypotheses through the Monte Carlo approximation. In particular, multiple foreground components are built to capture the variances of the foreground object in shapes and poses. To group the frames into different components, a tree-structured graphical model named temporal tree is designed, where visually similar and temporally coherent frames are arranged in branches. The temporal tree can be constructed by iteratively adding frames to the active nodes by probabilistic clustering. In addition, each component, consisting of frames in the same branch, is characterized by a support vector machine classifier, which is learned in a transductive fashion by jointly maximizing the margin over the labeled frames and the unlabeled frames. As the frames from the same video sequence follow the same distribution, the transductive classifiers achieve stronger generalization capability than inductive ones. Experimental results on the public benchmarks demonstrate the effectiveness of the proposed method in comparison with other state-of-the-art supervised and unsupervised video segmentation methods.

This paper proposes a novel multi-component video co-segmentation approach to simultaneously separate the foreground from the background in the video frames. To capture the variance of appearance of the foreground object, a multi-component foreground model is developed. Each component of the model characterizes a specific viewpoint/pose/appearance of the foreground object. To learn the parameters of the multi-component model, a transductive learning algorithm is leveraged to “transfer” the information of the labeled frames to the unlabeled frames in a tree-structured model, namely, temporal tree. Each branch of the temporal tree consists of the exemplars of a foreground component, and a transductive support vector regressor is capable of being trained. Experiments show that the proposed method outperforms quite a few state-of-the-art video segmentation algorithms in public benchmark.

Cosegmentation has achieved great success in exploiting inter-image segmentation consistency to segment a group of images simultaneously. To enforces non-local temporal coherence across all the frames by high-order object-level appearance/semantic correspondence with a compensation to the short-time window motion coherence cue, this paper cosegments the video frames together with a novel interframe segmentation consistency term. A direct application of existing cosegmentation algorithms to video frames encounters the following challenges: the high correlation of adjacent frames which makes the segmentation ambiguous and a large number of video frames which makes the computation expensive. To tackle them, we formulate the cosegmentation in a transductive learning framework to iteratively learn the inter-frame consistency term from all the video frames. The proposed algorithm is evaluated on the standard SegTrack dataset and promising results are obtained.