Federated learning (FL), as a representative distributed learning paradigm, has achieved remarkable success. However, most existing FL studies assume that each client holds correctly labeled data, whereas in reality, noisy labels are ubiquitous on the client side. To mitigate the adverse impact of noisy labels on FL performance, we propose a robust FL method with double denoising. Specifically, we first perform primary denoising based on the idea of cross-prediction, where two global models are trained and mutually used on each client to identify and filter out mislabeled samples. Next, after fine-tuning the models, we develop secondary denoising by detecting and removing residual noisy samples through clustering based on loss values. In addition, we design a noise-tolerant local training strategy that dynamically assesses the influence of noisy data on local updates and applies differentiated update rules to prevent overfitting. Finally, experimental results on three benchmark datasets, including the real-world noisy dataset Clothing1M, demonstrate that our method effectively removes label noise, delivering improved performance of the global model.

Estimating the 6-DoF posture of parts in assembly-based modeling is a critical task in the fields of computer graphics, computer vision and robotics. A typical scenario involves enabling a machine agent to automatically assemble IKEA furniture using the provided parts. This paper presents HiFormer, a novel Hierarchical Transformer with Box-packed Positional Encoding, designed for highly automatic 3D part assembly. Our method addresses three important issues commonly encountered in 3D part assembly: 1) How to mitigate the overfitting problem associated with Transformer-based feature learning for 3D point clouds? 2) How to effectively model the relationships between the intragroup and intergroup parts? 3) How to compute positional encoding and integrate it into the Transformer for parts with diverse geometric forms in the coarse-to-fine assembly process? These challenges are tackled through three key contributions: 1) a multi-task 3D Swin Transformer with a two-stage training strategy for feature extraction, 2) a novel hierarchical Transformer for capturing part relationships at flattening, intragroup, and intergroup levels, and 3) an innovative box-packed positional encoding that enhances the Transformer by incorporating query, key, and value information derived from relative box positions. On the PartNet benchmark, our method outperforms the state-of-the-art PWH-MP model on three representative categories-Chair, Table, and Lamp-, achieving average improvements of 2.84% in Part Accuracy (PA) and 3.72% in Connection Accuracy (CA) for diversity modeling (with noise), and 3.55% in PA and 3.21% in CA for deterministic modeling (without noise).

As an extension of machine unlearning in distributed scenarios, federated unlearning gains significant attention. However, federated unlearning remains challenging, as many studies require additional resources, such as auxiliary dataset or storage, to achieve high-quality models. These requirements incur extra costs and are often difficult to satisfy in practical applications. To address these issues, we propose a flexible client-level federated unlearning algorithm with prototypes, called FedUP. Specifically, our algorithm consists of two components: prototype-based unlearning and model recovering. First, we design a prototype-based unlearning strategy that uses prototypes of the erased client to guide the unlearning process, and maximizes the prototype loss between the remaining and erased clients to unlearn the information. It does not rely on historical storage updates or additional standard datasets, making the unlearning process more streamlined. To mitigate performance degradation from the unlearning process, we develop a brief model recovering approach guided by global prototypes to swiftly and efficiently restore models' accuracy on the remaining datasets. Unlike other unlearning algorithms, our approach exchanges prototypes instead of model parameters, significantly reducing communication overhead. Finally, we empirically evaluate the proposed algorithm from multiple perspectives on two datasets, demonstrating that our algorithm can achieve high-quality unlearned models with minimal communication cost.

In data-rich everything connected world, the rapid and green data processing is essential, especially for some delay-sensitive and computation-intensive tasks. Motivated by these requirements, an energy and delay co-aware fog computation offloading mechanism is conceived in this paper. Specifically, we formulate a weighted sum minimization problem of task completion time and energy consumption at the local fog for achieving efficient task computation. Further, a deep learning-based joint offloading decision and resource allocation (DL-JODRA) algorithm is developed to address such problem by jointly optimizing offloading action, local CPU, bandwidth and external CPU occupation ratios. The optimal offloading decision based comprehensive optimization consideration of network resources further improves the network efficiency. Finally, the extensive simulation results demonstrate that the proposed DL-JODRA can achieve optimal offloading decision with low computation resource requirement and gain significant reduction on network costs (i.e., delay and energy) comparing with benchmark methods.

Multi-view deep neural network is perhaps the most successful approach in 3D shape classification. However, the fusion of multi-view features based on max or average pooling lacks a view selection mechanism, limiting its application in, e.g., multi-view active object recognition by a robot. This paper presents VERAM, a view-enhanced recurrent attention model capable of actively selecting a sequence of views for highly accurate 3D shape classification. VERAM addresses an important issue commonly found in existing attention-based models, i.e., the unbalanced training of the subnetworks corresponding to next view estimation and shape classification. The classification subnetwork is easily overfitted while the view estimation one is usually poorly trained, leading to a suboptimal classification performance. This is surmounted by three essential view-enhancement strategies: 1) enhancing the information flow of gradient backpropagation for the view estimation subnetwork, 2) devising a highly informative reward function for the reinforcement training of view estimation and 3) formulating a novel loss function that explicitly circumvents view duplication. Taking grayscale image as input and AlexNet as CNN architecture, VERAM with 9 views achieves instance-level and class-level accuracy of 95.5 and 95.3 percent on ModelNet10, 93.7 and 92.1 percent on ModelNet40, both are the state-of-the-art performance under the same number of views.

This paper proposes a scalable method for organizing the collection of motion capture data for overview and exploration, and it mainly addresses three core problems, including data abstraction, neighborhood construction and data visualization. To alleviate the contradiction between limited visual space and the ever-increasing size of real-word datasets, hierarchical affinity propagation (HAP) is adopt to perform data abstraction on low-level pose features to generate multi-layers of data aggregations in consistent with coarse to fine abstraction levels of human cognition. To construct a meaningful neighborhood for user choosing a browsing path and positioning themselves, quartet analysis-based phylogenetic tree is created upon high-level pose features to produce more reliable neighbors for different aggregations of the specific abstraction level. To provide a convenient interactive environment for user navigation, a phylogenetic tree-centric visualization strategy in three-dimensional space is present. Experimental results on HDM05 motion capture dataset verify the effectiveness of the proposed method.