Search for: All records

Dealing with data heterogeneity is a key challenge in the theoretical analysis of federated learning (FL) algorithms. In the literature, gradient divergence is often used as the sole metric for data heterogeneity. However, we observe that the gradient divergence cannot fully characterize the impact of the data heterogeneity in Federated Averaging (FedAvg) even for the quadratic objective functions. This limitation leads to an overestimate of the communication complexity. Motivated by this observation, we propose a new analysis framework based on the difference between the minima of the global objective function and the minima of the local objective functions. Using the new framework, we derive a tighter convergence upper bound for heterogeneous quadratic objective functions. The theoretical results reveal new insights into the impact of the data heterogeneity on the convergence of FedAvg and provide a deeper understanding of the two-stage learning rates. Experimental results using non-IID data partitions validate the theoretical findings. 

In this work, we develop a two time-scale deep learning approach for beamforming and phase shift (BF-PS) design in time-varying RIS-aided networks. In contrast to most existing works that assume perfect CSI for BF-PS design, we take into account the cost of channel estimation and utilize Long Short-Term Memory (LSTM) networks to design BF-PS from limited samples of estimated channel CSI. An LSTM channel extrapolator is designed first to generate high resolution estimates of the cascaded BS-RIS-user channel from sampled signals acquired at a slow time scale. Subsequently, the outputs of the channel extrapolator are fed into an LSTM-based two stage neural network for the joint design of BF-PS at a fast time scale of per coherence time. To address the critical issue that training overhead increases linearly with the number of RIS elements, we consider various pilot structures and sampling patterns in time and space to evaluate the efficiency and sum-rate performance of the proposed two time-scale design. Our results show that the proposed two time-scale design can achieve good spectral efficiency when taking into account the pilot overhead required for training. The proposed design also outperforms a direct BF-PS design that does not employ a channel extrapolator. These demonstrate the feasibility of applying RIS in time-varying channels with reasonable pilot overhead. 

In this work, we propose a two-stage multi-agent deep deterministic policy gradient (TS-MADDPG) algorithm for communication-free, multi-agent reinforcement learning (MARL) under partial states and observations. In the first stage, we train prototype actor-critic networks using only partial states at actors. In the second stage, we incorporate partial observations resulting from prototype actions as side information at actors to enhance actor-critic training. This side information is useful to infer the unobserved states and hence, can help reduce the performance gap between a network with fully observable states and a partially observable one. Using a case study of building energy control in the power distribution network, we successfully demonstrate that the proposed TS-MADDPG can greatly improve the performance of single-stage MADDPG algorithms that use partial states only. This is the first work that utilizes partial local voltage measurements as observations to improve the MARL performance for a distributed power network. 

Our extensive real measurements over Amazon EC2 show that the virtual instances often have different computing speeds even if they share the same configurations. This motivates us to study heterogeneous Coded Storage Elastic Computing (CSEC) systems where machines, with different computing speeds, join and leave the network arbitrarily over different computing steps. In CSEC systems, a Maximum Distance Separable (MDS) code is used for coded storage such that the file placement does not have to be re-defined with each elastic event. Computation assignment algorithms are used to minimize the computation time given computation speeds of different machines. While previous studies of heterogeneous CSEC do not include stragglers - the slow machines during the computation, we develop a new framework in heterogeneous CSEC that introduces straggler tolerance. Based on this framework, we design a novel algorithm using our previously proposed approach for heterogeneous CSEC such that the system can handle any subset of stragglers of a specified size while minimizing the computation time. Furthermore, we establish a trade-off in computation time and straggler tolerance. Another major limitation of existing CSEC designs is the lack of practical evaluations using real applications. In this paper, we evaluate the performance of our designs on Amazon EC2 for applications of the power iteration and linear regression. Evaluation results show that the proposed heterogeneous CSEC algorithms outperform the state-of-the-art designs by more than 30%.