Federated Learning (FL) is a nascent decentralized learning framework under which a massive collection of heterogeneous clients collaboratively train a model without revealing their local data. Scarce communication, privacy leakage, and Byzantine attacks are the key bottlenecks of system scalability. In this paper, we focus on communication-efficient distributed (stochastic) gradient descent for non-convex optimization, a driving force of FL. We propose two algorithms, named {\em Adaptive Stochastic Sign SGD (Ada-StoSign)} and {\em $\beta$-Stochastic Sign SGD ($\beta$-StoSign)}, each of which compresses the local gradients into bit vectors. To handle unbounded gradients, Ada-StoSign uses a novel norm tracking function that adaptively adjusts a coarse estimation on the $\ell_{\infty}$ of the local gradients - a key parameter used in gradient compression. We show that Ada-StoSign converges in expectation with a rate $O(\log T/\sqrt{T} + 1/\sqrt{M})$, where $M$ is the number of clients. To the best of our knowledge, when $M$ is sufficiently large, Ada-StoSign outperforms the state-of-the-art sign-based method whose convergence rate is $O(T^{-1/4})$. Under bounded gradient assumption, $\beta$-StoSign achieves quantifiable Byzantine resilience and privacy assurances, and works with partial client participation and mini-batch gradients which could be unbounded. We corroborate and complement our theories by experiments on MNIST and CIFAR-10 datasets.
相關內容
This paper focuses on decentralized stochastic optimization in the presence of Byzantine attacks. During the optimization process, an unknown number of malfunctioning or malicious workers, termed as Byzantine workers, disobey the algorithmic protocol and send arbitrarily wrong messages to their neighbors. Even though various Byzantine-resilient algorithms have been developed for distributed stochastic optimization with a central server, we show that there are two major issues in the existing robust aggregation rules when being applied to the decentralized scenario: disagreement and non-doubly stochastic virtual mixing matrix. This paper provides comprehensive analysis that discloses the negative effects of these two issues, and gives guidelines of designing favorable Byzantine-resilient decentralized stochastic optimization algorithms. Under these guidelines, we propose iterative outlier scissor (IOS), an iterative filtering-based robust aggregation rule with provable performance guarantees. Numerical experiments demonstrate the effectiveness of IOS. The code of simulation implementation is available at github.com/Zhaoxian-Wu/IOS.
The rapid development of parallel and distributed computing paradigms has brought about great revolution in computing. Thanks to the intrinsic parallelism of evolutionary computation (EC), it is natural to implement EC on parallel and distributed computing systems. On the one hand, the computing power provided by parallel computing systems can significantly improve the efficiency and scalability of EC. On the other hand, data are collected and processed in a distributed manner, which brings a novel development direction and new challenges to EC. In this paper, we intend to give a systematic review on distributed EC (DEC). First, a new taxonomy for DEC is proposed from top design mechanism to bottom implementation mechanism. Based on this taxonomy, existing studies on DEC are reviewed in terms of purpose, parallel structure of the algorithm, parallel model for implementation, and the implementation environment. Second, we clarify two major purposes of DEC, i.e., improving efficiency through parallel processing for centralized optimization and cooperating distributed individuals/sub-populations with partial information to perform distributed optimization. Third, noting that the latter purpose of DEC is an emerging and attractive trend for EC with the booming of spatially distributed paradigms, this paper gives a systematic definition of the distributed optimization and classifies it into dimension distributed-, data distributed-, and objective distributed-optimization problems. Formal formulations for these problems are provided and various DEC studies on these problems are reviewed. We also discuss challenges and potential research directions, aiming to enlighten the design of DEC and pave the way for future developments.
Federated learning (FL) facilitates edge devices to cooperatively train a global shared model while maintaining the training data locally and privately. However, a common but impractical assumption in FL is that the participating edge devices possess the same required resources and share identical global model architecture. In this study, we propose a novel FL method called Federated Intermediate Layers Learning (FedIN), supporting heterogeneous models without utilizing any public dataset. The training models in FedIN are divided into three parts, including an extractor, the intermediate layers, and a classifier. The model architectures of the extractor and classifier are the same in all devices to maintain the consistency of the intermediate layer features, while the architectures of the intermediate layers can vary for heterogeneous devices according to their resource capacities. To exploit the knowledge from features, we propose IN training, training the intermediate layers in line with the features from other clients. Additionally, we formulate and solve a convex optimization problem to mitigate the gradient divergence problem induced by the conflicts between the IN training and the local training. The experiment results show that FedIN achieves the best performance in the heterogeneous model environment compared with the state-of-the-art algorithms. Furthermore, our ablation study demonstrates the effectiveness of IN training and the solution to the convex optimization problem.
Tiny machine learning (TinyML) is a rapidly growing field aiming to democratize machine learning (ML) for resource-constrained microcontrollers (MCUs). Given the pervasiveness of these tiny devices, it is inherent to ask whether TinyML applications can benefit from aggregating their knowledge. Federated learning (FL) enables decentralized agents to jointly learn a global model without sharing sensitive local data. However, a common global model may not work for all devices due to the complexity of the actual deployment environment and the heterogeneity of the data available on each device. In addition, the deployment of TinyML hardware has significant computational and communication constraints, which traditional ML fails to address. Considering these challenges, we propose TinyReptile, a simple but efficient algorithm inspired by meta-learning and online learning, to collaboratively learn a solid initialization for a neural network (NN) across tiny devices that can be quickly adapted to a new device with respect to its data. We demonstrate TinyReptile on Raspberry Pi 4 and Cortex-M4 MCU with only 256-KB RAM. The evaluations on various TinyML use cases confirm a resource reduction and training time saving by at least two factors compared with baseline algorithms with comparable performance.
For distributed graph processing on massive graphs, a graph is partitioned into multiple equally-sized parts which are distributed among machines in a compute cluster. In the last decade, many partitioning algorithms have been developed which differ from each other with respect to the partitioning quality, the run-time of the partitioning and the type of graph for which they work best. The plethora of graph partitioning algorithms makes it a challenging task to select a partitioner for a given scenario. Different studies exist that provide qualitative insights into the characteristics of graph partitioning algorithms that support a selection. However, in order to enable automatic selection, a quantitative prediction of the partitioning quality, the partitioning run-time and the run-time of subsequent graph processing jobs is needed. In this paper, we propose a machine learning-based approach to provide such a quantitative prediction for different types of edge partitioning algorithms and graph processing workloads. We show that training based on generated graphs achieves high accuracy, which can be further improved when using real-world data. Based on the predictions, the automatic selection reduces the end-to-end run-time on average by 11.1% compared to a random selection, by 17.4% compared to selecting the partitioner that yields the lowest cut size, and by 29.1% compared to the worst strategy, respectively. Furthermore, in 35.7% of the cases, the best strategy was selected.
In the LOCAL model, low-diameter decomposition is a useful tool in designing algorithms, as it allows us to shift from the general graph setting to the low-diameter graph setting, where brute-force information gathering can be done efficiently. Recently, Chang and Su [PODC 2022] showed that any high-conductance network excluding a fixed minor contains a high-degree vertex, so the entire graph topology can be gathered to one vertex efficiently in the CONGEST model using expander routing. Therefore, in networks excluding a fixed minor, many problems that can be solved efficiently in LOCAL via low-diameter decomposition can also be solved efficiently in CONGEST via expander decomposition. In this work, we show improved decomposition and routing algorithms for networks excluding a fixed minor in the CONGEST model. Our algorithms cost $\text{poly}(\log n, 1/\epsilon)$ rounds deterministically. For bounded-degree graphs, our algorithms finish in $O(\epsilon^{-1}\log n) + \epsilon^{-O(1)}$ rounds. Our algorithms have a wide range of applications, including the following results in CONGEST. 1. A $(1-\epsilon)$-approximate maximum independent set in a network excluding a fixed minor can be computed deterministically in $O(\epsilon^{-1}\log^\ast n) + \epsilon^{-O(1)}$ rounds, nearly matching the $\Omega(\epsilon^{-1}\log^\ast n)$ lower bound of Lenzen and Wattenhofer [DISC 2008]. 2. Property testing of any additive minor-closed property can be done deterministically in $O(\log n)$ rounds if $\epsilon$ is a constant or $O(\epsilon^{-1}\log n) + \epsilon^{-O(1)}$ rounds if the maximum degree $\Delta$ is a constant, nearly matching the $\Omega(\epsilon^{-1}\log n)$ lower bound of Levi, Medina, and Ron [PODC 2018].
This paper proposes a cell-free massive multiple-input multiple-output (CF-mMIMO) architecture with joint list-based detection with soft interference cancelation (soft-IC) and access points (APs) selection. In particular, we derive a new closed-form expression for the minimum mean-square error receive filter while taking the uplink transmit powers and APs selection into account. This is achieved by optimizing the receive combining vector by minimizing the mean square error between the detected symbol estimate and transmitted symbol, after canceling the multi-user interference (MUI). By using low-density parity check (LDPC) codes, an iterative detection and decoding (IDD) scheme based on a message passing is devised. In order to perform joint detection at the central processing unit (CPU), the access points locally estimate the channel and send their received sample data to the CPU via the front haul links. In order to enhance the system's bit error rate performance, the detected symbols are iteratively exchanged between the joint detector and the LDPC decoder in log likelihood ratio form. Furthermore, we draw insights into the derived detector as the number of IDD iterations increase. Finally, the proposed list detector is compared with existing detection techniques.
This paper reconsiders end-to-end learning approaches to the Optimal Power Flow (OPF). Existing methods, which learn the input/output mapping of the OPF, suffer from scalability issues due to the high dimensionality of the output space. This paper first shows that the space of optimal solutions can be significantly compressed using principal component analysis (PCA). It then proposes Compact Learning, a new method that learns in a subspace of the principal components before translating the vectors into the original output space. This compression reduces the number of trainable parameters substantially, improving scalability and effectiveness. Compact Learning is evaluated on a variety of test cases from the PGLib with up to 30,000 buses. The paper also shows that the output of Compact Learning can be used to warm-start an exact AC solver to restore feasibility, while bringing significant speed-ups.
Federated Learning (FL) is a decentralized machine-learning paradigm, in which a global server iteratively averages the model parameters of local users without accessing their data. User heterogeneity has imposed significant challenges to FL, which can incur drifted global models that are slow to converge. Knowledge Distillation has recently emerged to tackle this issue, by refining the server model using aggregated knowledge from heterogeneous users, other than directly averaging their model parameters. This approach, however, depends on a proxy dataset, making it impractical unless such a prerequisite is satisfied. Moreover, the ensemble knowledge is not fully utilized to guide local model learning, which may in turn affect the quality of the aggregated model. Inspired by the prior art, we propose a data-free knowledge distillation} approach to address heterogeneous FL, where the server learns a lightweight generator to ensemble user information in a data-free manner, which is then broadcasted to users, regulating local training using the learned knowledge as an inductive bias. Empirical studies powered by theoretical implications show that, our approach facilitates FL with better generalization performance using fewer communication rounds, compared with the state-of-the-art.
Federated learning is a new distributed machine learning framework, where a bunch of heterogeneous clients collaboratively train a model without sharing training data. In this work, we consider a practical and ubiquitous issue in federated learning: intermittent client availability, where the set of eligible clients may change during the training process. Such an intermittent client availability model would significantly deteriorate the performance of the classical Federated Averaging algorithm (FedAvg for short). We propose a simple distributed non-convex optimization algorithm, called Federated Latest Averaging (FedLaAvg for short), which leverages the latest gradients of all clients, even when the clients are not available, to jointly update the global model in each iteration. Our theoretical analysis shows that FedLaAvg attains the convergence rate of $O(1/(N^{1/4} T^{1/2}))$, achieving a sublinear speedup with respect to the total number of clients. We implement and evaluate FedLaAvg with the CIFAR-10 dataset. The evaluation results demonstrate that FedLaAvg indeed reaches a sublinear speedup and achieves 4.23% higher test accuracy than FedAvg.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191