Communication efficiency is one of the key bottlenecks in Federated Learning (FL). Compression techniques, such as sparsification and quantization, are used to reduce communication overhead. However, joint designs of ...
Communication efficiency is one of the key bottlenecks in Federated Learning (FL). Compression techniques, such as sparsification and quantization, are used to reduce communication overhead. However, joint designs of these techniques under communication constraints are not well-explored. This paper investigates the joint uplink compression problem in communication-constrained FL systems. We propose a Block-TopK sparsification scheme to reduce the proportion of bits used for locating entries of a sparsified vector. Considering the communication constraints, an optimization formulation is proposed to minimize the compression error. By solving the optimization problem, our joint compression method provides a better trade-off between sparsity budget and bit width. Numerical results demonstrate that our approach achieves 99.96% of baseline accuracy with only 1.56% of the baseline communication overhead when training ResNet-18 on CIFAR-10.
Image-based salient object detection (SOD) has been extensively explored in the past decades. However, SOD on 360◦ omnidirectional images is less studied owing to the lack of datasets with pixel-level annotations. Tow...
详细信息
This study investigated the positive effect of surface modification with ozone on the photocatalytic performance of anatase TiO2 with dominated（001） facets for toluene *** performance of photocatalyst was tested on ...
详细信息
This study investigated the positive effect of surface modification with ozone on the photocatalytic performance of anatase TiO2 with dominated（001） facets for toluene *** performance of photocatalyst was tested on a home-made volatile organic compounds degradation system. The ozone modification, toluene adsorption and degradation mechanism were established by a combination of various characterization methods, in situ diffuse reflectance infrared fourier transform spectroscopy, and density functional theory *** surface modification with ozone can significantly enhance the photocatalytic degradation performance for toluene. The abundant unsaturated coordinated 5 c-Ti sites on（001）facets act as the adsorption sites for ozone. The formed Ti–O bonds reacted with H2O to generate a large amount of isolated Ti5 c-OH which act as the adsorption sites for toluene,and thus significantly increase the adsorption capacity for toluene. The outstanding photocatalytic performance of ozone-modified TiO2 is due to its high adsorption ability for toluene and the abundant surface hydroxyl groups, which produce very reactive OH·radicals under irradiation. Furthermore, the O2 generated via ozone dissociation could combine with the photogenerated electrons to form superoxide radicals which are also conductive to the toluene degradation.
With the rapid progress of generative models, the current challenge in face forgery detection is how to effectively detect realistic manipulated faces from different unseen domains. Though previous studies show that p...
详细信息
3D anomaly detection has recently become a significant focus in computer vision. Several advanced methods have achieved satisfying anomaly detection performance. However, they typically concentrate on the external str...
详细信息
Document retrieval plays an essential role in many real-world applications especially when the data storage is outsourced. Due to the great advantages offered by cloud computing, clients tend to outsource their person...
详细信息
ISBN:
(数字)9798350381993
ISBN:
(纸本)9798350382006
Document retrieval plays an essential role in many real-world applications especially when the data storage is outsourced. Due to the great advantages offered by cloud computing, clients tend to outsource their personal data to remote servers maintained by external service providers. This raises serious privacy concerns about outsourced data because such providers are usually considered untrusted entities. The majority of previous schemes of document similarity search share the same limitation: they focus mainly on static collections. Dynamic searchable schemes (DSE) allow adding or removing documents at the expense of more leakage than static schemes. To thwart certain attacks, DSE schemes should support forward privacy property, which ensures that newly added documents cannot be related to previously issued search queries. We design and implement dynamic secure similarity search schemes with forward privacy for textual documents utilizing simhash method for hamming similarity. Our scheme provides an efficient search time and a sufficient level of privacy. To show the practicality of our proposed scheme, we performed excremental results with large document collections.
To the best of our knowledge, this paper is the first to apply IoT technologies to transform the popular Mahjong game into a Digital Mahjong system (DMS) for digitally performing cognitive assessments. People have sta...
详细信息
Implant prosthesis is the most appropriate treatment for dentition defect or dentition loss, which usually involves a surgical guide design process to decide the implant position. However, such design heavily relies o...
详细信息
Prototype-based clustering algorithms have garnered considerable attention in the field of machine learning due to their efficiency and interpretability. Nonetheless, these algorithms often face performance degradatio...
Prototype-based clustering algorithms have garnered considerable attention in the field of machine learning due to their efficiency and interpretability. Nonetheless, these algorithms often face performance degradation when confronted with high-dimensional or non-ellipsoidal data distributions. To surmount these challenges, this study introduces a novel clustering approach, dubbed Clustering with Adaptive Graph learning and Spectral Rotation (CAGSR). In CAGSR, the imposed spectral rotation operation mitigates the discrepancy between the membership matrix, which adheres to the notion of fuzzy clustering, and the spectral representations derived from an adaptive graph rather than a predefined one. This enables the generation of a comprehensive representation of the data across multiple spaces. Furthermore, the clustering and graph learning tasks are jointly optimized in a projected subspace, which can effectively reduce the adverse impact caused by irrelevant features in the original space. The proposed method seamlessly integrates fuzzy clustering, graph structure learning, and spectral rotation into a unified model, facilitating the detection of intrinsic structures. Experimental evaluations conducted on benchmark data sets substantiate the effectiveness of CAGSR when compared to related clustering approaches.
Deep Neural Networks (DNN) have achieved extraordinary success in many visual recognition tasks. Visual Transformer (ViT), which is derived from Natural Language Processing (NLP), has achieved state-of-the-art (SOTA) ...
Deep Neural Networks (DNN) have achieved extraordinary success in many visual recognition tasks. Visual Transformer (ViT), which is derived from Natural Language Processing (NLP), has achieved state-of-the-art (SOTA) results on many tasks due to its capability of capturing long-range dependencies in visual data. However, Existing ViT models are challenging to deploy on devices due to their massive computational consumption, huge memory overhead, and reliance on large datasets. In this work, we address these issues by replacing some computationally expensive and memory-intensive modules in ViT with standard Convolutional Neural Network (CNN) modules. Firstly, we propose an efficient Self-Attention module called SDG-Attention (SDGA) with linear space and time complexity, and an economical FeedForward Network (FFN) composed of group convolution and shuffle channel (SFFN). Then, we develop a lightweight CNN model with SDGA and SFFN, SDGFormer, which embraces several priors of ViT and is LayerNorm-Free. We evaluate SDGFormer on ImageNet-1K and Mini-ImageNet, and the SDGFormer-S achieves a comparable top-1 accuracy of 77.6% on ImageNet-1K with 9.1M parameters and 1.6 GFlops regimes. Moreover, our SDGFormer-T achieves SOTA performance on Mini-ImageNet with 83.3% accuracy, demonstrating good generalization on small datasets without extra data.
暂无评论