NeRCC: Nested-Regression Coded Computing for Resilient Distributed Prediction Serving Systems

Resilience against stragglers is a critical element of prediction serving systems, tasked with executing inferences on input data for a pre-trained machine-learning model. In this paper, we propose NeRCC, as a general straggler-resistant framework for approximate coded computing. NeRCC includes three layers: (1) encoding regression and sampling, which generates coded data points, as a combination of original data points, (2) computing, in which a cluster of workers run inference on the coded data points, (3) decoding regression and sampling, which approximately recovers the predictions of the original data points from the available predictions on the coded data points. We argue that the overall objective of the framework reveals an underlying interconnection between two regression models in the encoding and decoding layers. We propose a solution to the nested regressions problem by summarizing their dependence on two regularization terms that are jointly optimized. Our extensive experiments on different datasets and various machine learning models, including LeNet5, RepVGG, and Vision Transformer (ViT), demonstrate that NeRCC accurately approximates the original predictions in a wide range of stragglers, outperforming the state-of-the-art by up to 23%.

$\texttt{NeRCC}$: Nested-Regression Coded Computing for Resilient Distributed Prediction Serving Systems

Resilience against stragglers is a critical element of prediction serving systems, tasked with executing inferences on input data for a pre-trained machine-learning model. In this paper, we propose NeRCC, as a general straggler-resistant framework for approximate coded computing. NeRCC includes three layers: (1) encoding regression and sampling, which generates coded data points, as a combination of original data points, (2) computing, in which a cluster of workers run inference on the coded data points, (3) decoding regression and sampling, which approximately recovers the predictions of the original data points from the available predictions on the coded data points. We argue that the overall objective of the framework reveals an underlying interconnection between two regression models in the encoding and decoding layers. We propose a solution to the nested regressions problem by summarizing their dependence on two regularization terms that are jointly optimized. Our extensive experiments on different datasets and various machine learning models, including LeNet5, RepVGG, and Vision Transformer (ViT), demonstrate that NeRCC accurately approximates the original predictions in a wide range of stragglers, outperforming the state-of-the-art by up to 23%.

Byzantine-Resistant Secure Aggregation for Federated Learning Based on Coded Computing and Vector Commitment

In this paper, we propose an efficient secure aggregation scheme for federated learning that is protected against Byzantine attacks and privacy leakages. Processing individual updates to manage adversarial behavior, while preserving privacy of data against colluding nodes, requires some sort of secure secret sharing. However, communication load for secret sharing of long vectors of updates can be very high. To resolve this issue, in the proposed scheme, local updates are partitioned into smaller sub-vectors and shared using ramp secret sharing. However, this sharing method does not admit bi-linear computations, such as pairwise distance calculations, needed by outlier-detection algorithms. To overcome this issue, each user runs another round of ramp sharing, with different embedding of data in the sharing polynomial. This technique, motivated by ideas from coded computing, enables secure computation of pairwise distance. In addition, to maintain the integrity and privacy of the local update, the proposed scheme also uses a vector commitment method, in which the commitment size remains constant (i.e. does not increase with the length of the local update), while simultaneously allowing verification of the secret sharing process.

SwiftAgg+: Achieving Asymptotically Optimal Communication Load in Secure Aggregation for Federated Learning

We propose SwiftAgg+, a novel secure aggregation protocol for federated learning systems, where a central server aggregates local models of $N\in\mathbb{N}$ distributed users, each of size $L \in \mathbb{N}$, trained on their local data, in a privacy-preserving manner. SwiftAgg+ can significantly reduce the communication overheads without any compromise on security, and achieve the optimum communication load within a diminishing gap. Specifically, in presence of at most $D$ dropout users, SwiftAgg+ achieves average per-user communication load of $(1+\mathcal{O}(\frac{1}{N}))L$ and the server communication load of $(1+\mathcal{O}(\frac{1}{N}))L$, with a worst-case information-theoretic security guarantee, against any subset of up to $T$ semi-honest users who may also collude with the curious server. The proposed SwiftAgg+ has also a flexibility to reduce the number of active communication links at the cost of increasing the the communication load between the users and the server. In particular, for any $K\in\mathbb{N}$, SwiftAgg+ can achieve the uplink communication load of $(1+\frac{T}{K})L$, and per-user communication load of up to $(1-\frac{1}{N})(1+\frac{T+D}{K})L$, where the number of pair-wise active connections in the network is $\frac{N}{2}(K+T+D+1)$.

SwiftAgg: Communication-Efficient and Dropout-Resistant Secure Aggregation for Federated Learning with Worst-Case Security Guarantees

We propose SwiftAgg, a novel secure aggregation protocol for federated learning systems, where a central server aggregates local models of $N$ distributed users, each of size $L$, trained on their local data, in a privacy-preserving manner. Compared with state-of-the-art secure aggregation protocols, SwiftAgg significantly reduces the communication overheads without any compromise on security. Specifically, in presence of at most $D$ dropout users, SwiftAgg achieves a users-to-server communication load of $(T+1)L$ and a users-to-users communication load of up to $(N-1)(T+D+1)L$, with a worst-case information-theoretic security guarantee, against any subset of up to $T$ semi-honest users who may also collude with the curious server. The key idea of SwiftAgg is to partition the users into groups of size $D+T+1$, then in the first phase, secret sharing and aggregation of the individual models are performed within each group, and then in the second phase, model aggregation is performed on $D+T+1$ sequences of users across the groups. If a user in a sequence drops out in the second phase, the rest of the sequence remain silent. This design allows only a subset of users to communicate with each other, and only the users in a single group to directly communicate with the server, eliminating the requirements of 1) all-to-all communication network across users; and 2) all users communicating with the server, for other secure aggregation protocols. This helps to substantially slash the communication costs of the system.

Optimal Communication-Computation Trade-Off in Heterogeneous Gradient Coding

Gradient coding allows a master node to derive the aggregate of the partial gradients, calculated by some worker nodes over the local data sets, with minimum communication cost, and in the presence of stragglers. In this paper, for gradient coding with linear encoding, we characterize the optimum communication cost for heterogeneous distributed systems with \emph{arbitrary} data placement, with $s \in \mathbb{N}$ stragglers and $a \in \mathbb{N}$ adversarial nodes. In particular, we show that the optimum communication cost, normalized by the size of the gradient vectors, is equal to $(r-s-2a)^{-1}$, where $r \in \mathbb{N}$ is the minimum number that a data partition is replicated. In other words, the communication cost is determined by the data partition with the minimum replication, irrespective of the structure of the placement. The proposed achievable scheme also allows us to target the computation of a polynomial function of the aggregated gradient matrix. It also allows us to borrow some ideas from approximation computing and propose an approximate gradient coding scheme for the cases when the repetition in data placement is smaller than what is needed to meet the restriction imposed on communication cost or when the number of stragglers appears to be more than the presumed value in the system design.

Berrut Approximated Coded Computing: Straggler Resistance Beyond Polynomial Computing

One of the major challenges in using distributed learning to train complicated models with large data sets is to deal with stragglers effect. As a solution, coded computation has been recently proposed to efficiently add redundancy to the computation tasks. In this technique, coding is used across data sets, and computation is done over coded data, such that the results of an arbitrary subset of worker nodes with a certain size are enough to recover the final results. The major challenges with those approaches are (1) they are limited to polynomial function computations, (2) the size of the subset of servers that we need to wait for grows with the multiplication of the size of the data set and the model complexity (the degree of the polynomial), which can be prohibitively large, (3) they are not numerically stable for computation over real numbers. In this paper, we propose Berrut Approximated Coded Computing (BACC), as an alternative approach, which is not limited to polynomial function computation. In addition, the master node can approximately calculate the final results, using the outcomes of any arbitrary subset of available worker nodes. The approximation approach is proven to be numerically stable with low computational complexity. In addition, the accuracy of the approximation is established theoretically and verified by simulation results in different settings such as distributed learning problems. In particular, BACC is used to train a deep neural network on a cluster of servers, which outperforms repetitive computation (repetition coding) in terms of the rate of convergence.

A Hybrid-Order Distributed SGD Method for Non-Convex Optimization to Balance Communication Overhead, Computational Complexity, and Convergence Rate

In this paper, we propose a method of distributed stochastic gradient descent (SGD), with low communication load and computational complexity, and still fast convergence. To reduce the communication load, at each iteration of the algorithm, the worker nodes calculate and communicate some scalers, that are the directional derivatives of the sample functions in some \emph{pre-shared directions}. However, to maintain accuracy, after every specific number of iterations, they communicate the vectors of stochastic gradients. To reduce the computational complexity in each iteration, the worker nodes approximate the directional derivatives with zeroth-order stochastic gradient estimation, by performing just two function evaluations rather than computing a first-order gradient vector. The proposed method highly improves the convergence rate of the zeroth-order methods, guaranteeing order-wise faster convergence. Moreover, compared to the famous communication-efficient methods of model averaging (that perform local model updates and periodic communication of the gradients to synchronize the local models), we prove that for the general class of non-convex stochastic problems and with reasonable choice of parameters, the proposed method guarantees the same orders of communication load and convergence rate, while having order-wise less computational complexity. Experimental results on various learning problems in neural networks applications demonstrate the effectiveness of the proposed approach compared to various state-of-the-art distributed SGD methods.

Corella: A Private Multi Server Learning Approach based on Correlated Queries

The emerging applications of machine learning algorithms on mobile devices motivate us to offload the computation tasks of training a model or deploying a trained one to the cloud. One of the major challenges in this setup is to guarantee the privacy of the client's data. Various methods have been proposed to protect privacy in the literature. Those include (i) adding noise to the client data, which reduces the accuracy of the result, (ii) using secure multiparty computation, which requires significant communication among the computing nodes or with the client, (iii) relying on homomorphic encryption methods, which significantly increases computation load. In this paper, we propose an alternative approach to protect the privacy of user data. The proposed scheme relies on a cluster of servers where at most $T$ of them for some integer $T$, may collude, that each running a deep neural network. Each server is fed with the client data, added with a $\textit{strong}$ noise. This makes the information leakage to each server information-theoretically negligible. On the other hand, the added noises for different servers are $\textit{correlated}$. This correlation among queries allows the system to be $\textit{trained}$ such that the client can recover the final result with high accuracy, by combining the outputs of the servers, with minor computation efforts. Simulation results for various datasets demonstrate the accuracy of the proposed approach.

Coded Fourier Transform

We consider the problem of computing the Fourier transform of high-dimensional vectors, distributedly over a cluster of machines consisting of a master node and multiple worker nodes, where the worker nodes can only store and process a fraction of the inputs. We show that by exploiting the algebraic structure of the Fourier transform operation and leveraging concepts from coding theory, one can efficiently deal with the straggler effects. In particular, we propose a computation strategy, named as coded FFT, which achieves the optimal recovery threshold, defined as the minimum number of workers that the master node needs to wait for in order to compute the output. This is the first code that achieves the optimum robustness in terms of tolerating stragglers or failures for computing Fourier transforms. Furthermore, the reconstruction process for coded FFT can be mapped to MDS decoding, which can be solved efficiently. Moreover, we extend coded FFT to settings including computing general $n$-dimensional Fourier transforms, and provide the optimal computing strategy for those settings.