23 May 2023 in IEEE Transactions on Neural Networks and Learning Systems
Federated learning is an emerging learning paradigm where multiple clients collaboratively train a machine learning model in a privacy-preserving manner. Personalized federated learning extends this paradigm to overcome heterogeneity across clients by learning personalized models. Recently, there have been some initial attempts to apply transformers to federated learning. However, the impacts of federated learning algorithms on self-attention have not yet been studied. In this article, we investigate this relationship and reveal that federated averaging (FedAvg) algorithms actually have a negative impact on self-attention in cases of data heterogeneity, which limits the capabilities of the transformer model in federated learning settings. To address this issue, we propose FedTP, a novel transformer-based federated learning framework that learns personalized self-attention for each client while aggregating the other parameters among the clients. Instead of using a vanilla personalization mechanism that maintains personalized self-attention layers of each client locally, we develop a learn-to-personalize mechanism to further encourage the cooperation among clients and to increase the scalability and generalization of FedTP. Specifically, we achieve this by learning a hypernetwork on the server that outputs the personalized projection matrices of self-attention layers to generate clientwise queries, keys, and values. Furthermore, we present the generalization bound for FedTP with the learn-to-personalize mechanism. Extensive experiments verify that FedTP with the learn-to-personalize mechanism yields state-of-the-art performance in the non-IID scenarios. Our code is available online https://github.com/zhyczy/FedTP.
17 May 2021 in IEEE Transactions on Geoscience and Remote Sensing
Polarimetric synthetic aperture radar (PolSAR) image classification is one of the hottest issues in remote sensing, where studies on pixel-level information and relationship are of great significance. In this article, graph convolutional network (GCN) is employed to accomplish this pixel-level task benefiting from its excellent capability in structure exploration and information propagation between different pixels. To reduce the communication burden between various PolSAR pixels and high computational cost for the whole PolSAR image, an adaptive GCN (AdapGCN) consisting of pixel-centered subgraphs is proposed in this article. In the AdapGCN, a data-adaptive kernel and a spatial-adaptive kernel are introduced to, respectively, model data structure and spatial structure for PolSAR image. Moreover, a multiscale learning structure is integrated to further explore complicated relations between pixels. Extensive comparative evaluations validate the superiority of our new AdapGCN model for PolSAR image classification over a wide range of state-of-the-art methods on three challenging benchmarks.
4 January 2023 in Computer Vision and Image Understanding
27 November 2012 in IET Communications
Algebraic soft-decision decoding (ASD) of Reed–Solomon (RS) codes can obtain significant coding gain over the hard-decision decoding with polynomial complexity. Compared with other ASD algorithms, the low-complexity chase (LCC) decoding testing 2η test vectors has less computation complexity with similar or better coding gain. To reduce the latency of the interpolation, one major step of the LCC decoding, multiple interpolators can be applied and the pipelined architecture is usually adopted to make the throughput of the decoder higher. However, for application specific integrated circuit (ASIC) implementation and practical applications, the area and power consumption of the pipelined decoder are not preferable. This study proposes a modified serial LCC decoder architecture which reduces the power consumption and hardware requirement while keeping the shortest critical path as one adder, one multiplexer and one multiplier. As major contributions of this study, a novel re-encoder and erasure decoder block and an improved way to compute syndromes are proposed to reduce the whole latency of the decoder. For a (458, 410) RS code over Galois Field (GF)(210) with η=8, the hardware requirement and power consumption of the proposed LCC decoder are reduced by 23 and 25% than those of the common pipelined counterparts, respectively.