We propose a novel algorithm for distributed stochastic gradient descent (SGD) with compressed gradient communication in the parameter-server framework. Our gradient compression technique, named flattened one-bit stochastic gradient descent (FO-SGD), relies on two simple algorithmic ideas: (i) a one-bit quantization procedure leveraging the technique of dithering, and (ii) a randomized fast Walsh-Hadamard transform to flatten the stochastic gradient before quantization. As a result, the approximation of the true gradient in this scheme is biased, but it prevents commonly encountered algorithmic problems, such as exploding variance in the one-bit compression regime, deterioration of performance in the case of sparse gradients, and restrictive assumptions on the distribution of the stochastic gradients. In fact, we show SGD-like convergence guarantees under mild conditions. The compression technique can be used in both directions of worker-server communication, therefore admitting distributed optimization with full communication compression.

In the field of heterogeneous federated learning (FL), the key challenge is to efficiently and collaboratively train models across multiple clients with different data distributions, model structures, task objectives, computational capabilities, and communication resources. This diversity leads to significant heterogeneity, which increases the complexity of model training. In this paper, we first outline the basic concepts of heterogeneous federated learning and summarize the research challenges in federated learning in terms of five aspects: data, model, task, device, and communication. In addition, we explore how existing state-of-the-art approaches cope with the heterogeneity of federated learning, and categorize and review these approaches at three different levels: data-level, model-level, and architecture-level. Subsequently, the paper extensively discusses privacy-preserving strategies in heterogeneous federated learning environments. Finally, the paper discusses current open issues and directions for future research, aiming to promote the further development of heterogeneous federated learning.

Training neural networks requires optimizing a loss function that may be highly irregular, and in particular neither convex nor smooth. Popular training algorithms are based on stochastic gradient descent with momentum (SGDM), for which classical analysis applies only if the loss is either convex or smooth. We show that a very small modification to SGDM closes this gap: simply scale the update at each time point by an exponentially distributed random scalar. The resulting algorithm achieves optimal convergence guarantees. Intriguingly, this result is not derived by a specific analysis of SGDM: instead, it falls naturally out of a more general framework for converting online convex optimization algorithms to non-convex optimization algorithms.

Fine-tuning large-scale pre-trained models via transfer learning is an emerging important paradigm for a wide range of downstream tasks, with performance heavily reliant on extensive data. Federated learning (FL), as a distributed framework, provides a secure solution to train models on local datasets while safeguarding raw sensitive data. However, FL networks encounter high communication costs due to the massive parameters of large-scale pre-trained models, necessitating parameter-efficient methods. Notably, parameter efficient fine tuning, such as Low-Rank Adaptation (LoRA), has shown remarkable success in fine-tuning pre-trained models. However, prior research indicates that the fixed parameter budget may be prone to the overfitting or slower convergence. To address this challenge, we propose a Simulated Annealing-based Federated Learning with LoRA tuning (SA-FedLoRA) approach by reducing trainable parameters. Specifically, SA-FedLoRA comprises two stages: initiating and annealing. (1) In the initiating stage, we implement a parameter regularization approach during the early rounds of aggregation, aiming to mitigate client drift and accelerate the convergence for the subsequent tuning. (2) In the annealing stage, we allocate higher parameter budget during the early 'heating' phase and then gradually shrink the budget until the 'cooling' phase. This strategy not only facilitates convergence to the global optimum but also reduces communication costs. Experimental results demonstrate that SA-FedLoRA is an efficient FL, achieving superior performance to FedAvg and significantly reducing communication parameters by up to 93.62%.

Decentralized learning (DL) faces increased vulnerability to privacy breaches due to sophisticated attacks on machine learning (ML) models. Secure aggregation is a computationally efficient cryptographic technique that enables multiple parties to compute an aggregate of their private data while keeping their individual inputs concealed from each other and from any central aggregator. To enhance communication efficiency in DL, sparsification techniques are used, selectively sharing only the most crucial parameters or gradients in a model, thereby maintaining efficiency without notably compromising accuracy. However, applying secure aggregation to sparsified models in DL is challenging due to the transmission of disjoint parameter sets by distinct nodes, which can prevent masks from canceling out effectively. This paper introduces CESAR, a novel secure aggregation protocol for DL designed to be compatible with existing sparsification mechanisms. CESAR provably defends against honest-but-curious adversaries and can be formally adapted to counteract collusion between them. We provide a foundational understanding of the interaction between the sparsification carried out by the nodes and the proportion of the parameters shared under CESAR in both colluding and non-colluding environments, offering analytical insight into the working and applicability of the protocol. Experiments on a network with 48 nodes in a 3-regular topology show that with random subsampling, CESAR is always within 0.5% accuracy of decentralized parallel stochastic gradient descent (D-PSGD), while adding only 11% of data overhead. Moreover, it surpasses the accuracy on TopK by up to 0.3% on independent and identically distributed (IID) data.

Stochastic Gradient (SG) Markov Chain Monte Carlo algorithms (MCMC) are popular algorithms for Bayesian sampling in the presence of large datasets. However, they come with little theoretical guarantees and assessing their empirical performances is non-trivial. In such context, it is crucial to develop algorithms that are robust to the choice of hyperparameters and to gradients heterogeneity since, in practice, both the choice of step-size and behaviour of target gradients induce hard-to-control biases in the invariant distribution. In this work we introduce the stochastic gradient Barker dynamics (SGBD) algorithm, extending the recently developed Barker MCMC scheme, a robust alternative to Langevin-based sampling algorithms, to the stochastic gradient framework. We characterize the impact of stochastic gradients on the Barker transition mechanism and develop a bias-corrected version that, under suitable assumptions, eliminates the error due to the gradient noise in the proposal. We illustrate the performance on a number of high-dimensional examples, showing that SGBD is more robust to hyperparameter tuning and to irregular behavior of the target gradients compared to the popular stochastic gradient Langevin dynamics algorithm.

This paper studies how a stochastic gradient algorithm (SG) can be controlled to hide the estimate of the local stationary point from an eavesdropper. Such problems are of significant interest in distributed optimization settings like federated learning and inventory management. A learner queries a stochastic oracle and incentivizes the oracle to obtain noisy gradient measurements and perform SG. The oracle probabilistically returns either a noisy gradient of the function} or a non-informative measurement, depending on the oracle state and incentive. The learner's query and incentive are visible to an eavesdropper who wishes to estimate the stationary point. This paper formulates the problem of the learner performing covert optimization by dynamically incentivizing the stochastic oracle and obfuscating the eavesdropper as a finite-horizon Markov decision process (MDP). Using conditions for interval-dominance on the cost and transition probability structure, we show that the optimal policy for the MDP has a monotone threshold structure. We propose searching for the optimal stationary policy with the threshold structure using a stochastic approximation algorithm and a multi-armed bandit approach. The effectiveness of our methods is numerically demonstrated on a covert federated learning hate-speech classification task.

The conditional mean embedding (CME) encodes Markovian stochastic kernels through their actions on probability distributions embedded within the reproducing kernel Hilbert spaces (RKHS). The CME plays a key role in several well-known machine learning tasks such as reinforcement learning, analysis of dynamical systems, etc. We present an algorithm to learn the CME incrementally from data via an operator-valued stochastic gradient descent. As is well-known, function learning in RKHS suffers from scalability challenges from large data. We utilize a compression mechanism to counter the scalability challenge. The core contribution of this paper is a finite-sample performance guarantee on the last iterate of the online compressed operator learning algorithm with fast-mixing Markovian samples, when the target CME may not be contained in the hypothesis space. We illustrate the efficacy of our algorithm by applying it to the analysis of an example dynamical system.

In this work, we develop and analyze a Gradient Descent (GD) based solution, called Alternating GD and Minimization (AltGDmin), for efficiently solving the low rank matrix completion (LRMC) in a federated setting. LRMC involves recovering an $n \times q$ rank-$r$ matrix $\Xstar$ from a subset of its entries when $r \ll \min(n,q)$. Our theoretical guarantees (iteration and sample complexity bounds) imply that AltGDmin is the most communication-efficient solution in a federated setting, is one of the fastest, and has the second best sample complexity among all iterative solutions to LRMC. In addition, we also prove two important corollaries. (a) We provide a guarantee for AltGDmin for solving the noisy LRMC problem. (b) We show how our lemmas can be used to provide an improved sample complexity guarantee for AltMin, which is the fastest centralized solution.

Graph Neural Networks (GNNs) have emerged as potent models for graph learning. Distributing the training process across multiple computing nodes is the most promising solution to address the challenges of ever-growing real-world graphs. However, current adversarial attack methods on GNNs neglect the characteristics and applications of the distributed scenario, leading to suboptimal performance and inefficiency in attacking distributed GNN training. In this study, we introduce Disttack, the first framework of adversarial attacks for distributed GNN training that leverages the characteristics of frequent gradient updates in a distributed system. Specifically, Disttack corrupts distributed GNN training by injecting adversarial attacks into one single computing node. The attacked subgraphs are precisely perturbed to induce an abnormal gradient ascent in backpropagation, disrupting gradient synchronization between computing nodes and thus leading to a significant performance decline of the trained GNN. We evaluate Disttack on four large real-world graphs by attacking five widely adopted GNNs. Compared with the state-of-the-art attack method, experimental results demonstrate that Disttack amplifies the model accuracy degradation by 2.75 and achieves speedup by 17.33 on average while maintaining unnoticeability. Keywords: Graph Neural Network Distributed Training Adversarial Attack.