arXiv subjects

Salman Avestimehr

Publications and source records attributed to Salman Avestimehr.

At least 127 records · Page 7Linked to original sources

arXiv2020

Group Knowledge Transfer: Federated Learning of Large CNNs at the Edge

Scaling up the convolutional neural network (CNN) size (e.g., width, depth, etc.) is known to effectively improve model accuracy. However, the large model size impedes training on resource-constrained edge devices. For instance, federated learning (FL) may place undue burden on the compute capability of edge nodes, even though there is a strong practical need for FL due to its privacy and confidentiality properties. To address the resource-constrained reality of edge devices, we reformulate FL as a group knowledge transfer training algorithm, called FedGKT. FedGKT designs a variant of the alternating minimization approach to train small CNNs on edge nodes and periodically transfer their knowledge by knowledge distillation to a large server-side CNN. FedGKT consolidates several advantages into a single framework: reduced demand for edge computation, lower communication bandwidth for large CNNs, and asynchronous training, all while maintaining model accuracy comparable to FedAvg. We train CNNs designed based on ResNet-56 and ResNet-110 using three distinct datasets (CIFAR-10, CIFAR-100, and CINIC-10) and their non-I.I.D. variants. Our results show that FedGKT can obtain comparable or even slightly higher accuracy than FedAvg. More importantly, FedGKT makes edge training affordable. Compared to the edge training using FedAvg, FedGKT demands 9 to 17 times less computational power (FLOPs) on edge devices and requires 54 to 105 times fewer parameters in the edge CNN. Our source code is released at FedML (https://fedml.ai).

cs.LG↗

arXiv2019

Jupiter: A Networked Computing Architecture

In the era of Internet of Things, there is an increasing demand for networked computing to support the requirements of the time-constrained, compute-intensive distributed applications such as multi-camera video processing and data fusion for security. We present Jupiter, an open source networked computing system that inputs a Directed Acyclic Graph (DAG)-based computational task graph to efficiently distribute the tasks among a set of networked compute nodes regardless of their geographical separations and orchestrates the execution of the DAG thereafter. This Kubernetes container-orchestration-based system supports both centralized and decentralized scheduling algorithms for optimally mapping the tasks based on information from a range of profilers: network profilers, resource profilers, and execution time profilers. While centralized scheduling algorithms with global knowledge have been popular among the grid/cloud computing community, we argue that a distributed scheduling approach is better suited for networked computing due to lower communication and computation overhead in the face of network dynamics. To this end, we propose and implement a new class of distributed scheduling algorithms called WAVE on the Jupiter system. We present a set of real world experiments on two separate testbeds - one a world-wide network of 90 cloud computers across 8 cities and the other a cluster of 30 Raspberry pi nodes, over a simple networked computing application called Distributed Network Anomaly Detector (DNAD). We show that despite using more localized knowledge, a distributed WAVE greedy algorithm can achieve similar performance as a classical centralized scheduling algorithm called Heterogeneous Earliest Finish Time (HEFT), suitably enhanced for the Jupiter system.

cs.DC↗

arXiv2019

Coded Merkle Tree: Solving Data Availability Attacks in Blockchains

In this paper, we propose coded Merkle tree (CMT), a novel hash accumulator that offers a constant-cost protection against data availability attacks in blockchains, even if the majority of the network nodes are malicious. A CMT is constructed using a family of sparse erasure codes on each layer, and is recovered by iteratively applying a peeling-decoding technique that enables a compact proof for data availability attack on any layer. Our algorithm enables any node to verify the full availability of any data block generated by the system by just downloading a $Θ(1)$ byte block hash commitment and randomly sampling $Θ(\log b)$ bytes, where $b$ is the size of the data block. With the help of only one connected honest node in the system, our method also allows any node to verify any tampering of the coded Merkle tree by just downloading $Θ(\log b)$ bytes. We provide a modular library for CMT in Rust and Python and demonstrate its efficacy inside the Parity Bitcoin client.

cs.CR↗

arXiv2019

Train Where the Data is: A Case for Bandwidth Efficient Coded Training

Training a machine learning model is both compute and data-intensive. Most of the model training is performed on high performance compute nodes and the training data is stored near these nodes for faster training. But there is a growing interest in enabling training near the data. For instance, mobile devices are rich sources of training data. It may not be feasible to consolidate the data from mobile devices into a cloud service, due to bandwidth and data privacy reasons. Training at mobile devices is however fraught with challenges. First mobile devices may join or leave the distributed setting, either voluntarily or due to environmental uncertainties, such as lack of power. Tolerating uncertainties is critical to the success of distributed mobile training. One proactive approach to tolerate computational uncertainty is to store data in a coded format and perform training on coded data. Encoding data is a challenging task since erasure codes require multiple devices to exchange their data to create a coded data partition, which places a significant bandwidth constraint. Furthermore, coded computing traditionally relied on a central node to encode and distribute data to all the worker nodes, which is not practical in a distributed mobile setting. In this paper, we tackle the uncertainty in distributed mobile training using a bandwidth-efficient encoding strategy. We use a Random Linear Network coding (RLNC) which reduces the need to exchange data partitions across all participating mobile devices, while at the same time preserving the property of coded computing to tolerate uncertainties. We implement gradient descent for logistic regression and SVM to evaluate the effectiveness of our mobile training framework. We demonstrate a 50% reduction in total required communication bandwidth compared to MDS coded computation, one of the popular erasure codes.

cs.DC↗

arXiv2019

Collage Inference: Using Coded Redundancy for Low Variance Distributed Image Classification

MLaaS (ML-as-a-Service) offerings by cloud computing platforms are becoming increasingly popular. Hosting pre-trained machine learning models in the cloud enables elastic scalability as the demand grows. But providing low latency and reducing the latency variance is a key requirement. Variance is harder to control in a cloud deployment due to uncertainties in resource allocations across many virtual instances. We propose the collage inference technique which uses a novel convolutional neural network model, collage-cnn, to provide low-cost redundancy. A collage-cnn model takes a collage image formed by combining multiple images and performs multi-image classification in one shot, albeit at slightly lower accuracy. We augment a collection of traditional single image classifier models with a single collage-cnn classifier which acts as their low-cost redundant backup. Collage-cnn provides backup classification results if any single image classification requests experience slowdown. Deploying the collage-cnn models in the cloud, we demonstrate that the 99th percentile tail latency of inference can be reduced by 1.2x to 2x compared to replication based approaches while providing high accuracy. Variation in inference latency can be reduced by 1.8x to 15x.

cs.CV↗

arXiv2019

Slack Squeeze Coded Computing for Adaptive Straggler Mitigation

While performing distributed computations in today's cloud-based platforms, execution speed variations among compute nodes can significantly reduce the performance and create bottlenecks like stragglers. Coded computation techniques leverage coding theory to inject computational redundancy and mitigate stragglers in distributed computations. In this paper, we propose a dynamic workload distribution strategy for coded computation called Slack Squeeze Coded Computation ($S^2C^2$). $S^2C^2$ squeezes the compute slack (i.e., overhead) that is built into the coded computing frameworks by efficiently assigning work for all fast and slow nodes according to their speeds and without needing to re-distribute data. We implement an LSTM-based speed prediction algorithm to predict speeds of compute nodes. We evaluate $S^2C^2$ on linear algebraic algorithms, gradient descent, graph ranking, and graph filtering algorithms. We demonstrate 19% to 39% reduction in total computation latency using $S^2C^2$ compared to job replication and coded computation. We further show how $S^2C^2$ can be applied beyond matrix-vector multiplication.

cs.DC↗

arXiv2019

Interactive Verifiable Polynomial Evaluation

Cloud computing platforms have created the possibility for computationally limited users to delegate demanding tasks to strong but untrusted servers. Verifiable computing algorithms help build trust in such interactions by enabling the server to provide a proof of correctness of his results which the user can check very efficiently. In this paper, we present a doubly-efficient interactive algorithm for verifiable polynomial evaluation. Unlike the mainstream literature on verifiable computing, the soundness of our algorithm is information-theoretic and cannot be broken by a computationally unbounded server. By relying on basic properties of error correcting codes, our algorithm enforces a dishonest server to provide false results to problems which become progressively easier to verify. After roughly $\log d$ rounds, the user can verify the response of the server against a look-up table that has been pre-computed during an initialization phase. For a polynomial of degree $d$, we achieve a user complexity of $O(d^ε)$, a server complexity of $O(d^{1+ε})$, a round complexity of $O(\log d)$ and an initialization complexity of $O(d^{1+ε})$.

cs.CC↗

arXiv2019

Coded State Machine -- Scaling State Machine Execution under Byzantine Faults

We introduce an information-theoretic framework, named Coded State Machine (CSM), to securely and efficiently execute multiple state machines on untrusted network nodes, some of which are Byzantine. The standard method of solving this problem is using State Machine Replication, which achieves high security at the cost of low efficiency. We propose CSM, which achieves the optimal linear scaling in storage efficiency, throughput, and security simultaneously with the size of the network. The storage efficiency is scaled via the design of Lagrange coded states and coded input commands that require the same storage size as their origins. The computational efficiency is scaled using a novel delegation algorithm, called INTERMIX, which is an information-theoretically verifiable matrix-vector multiplication algorithm of independent interest. Using INTERMIX, the network nodes securely delegate their coding operations to a single worker node, and a small group of randomly selected auditor nodes verify its correctness, so that computational efficiency can scale almost linearly with the network size, without compromising on security.

cs.IT↗

arXiv2019

Collage Inference: Achieving low tail latency during distributed image classification using coded redundancy models

Reducing the latency variance in machine learning inference is a key requirement in many applications. Variance is harder to control in a cloud deployment in the presence of stragglers. In spite of this challenge, inference is increasingly being done in the cloud, due to the advent of affordable machine learning as a service (MLaaS) platforms. Existing approaches to reduce variance rely on replication which is expensive and partially negates the affordability of MLaaS. In this work, we argue that MLaaS platforms also provide unique opportunities to cut the cost of redundancy. In MLaaS platforms, multiple inference requests are concurrently received by a load balancer which can then create a more cost-efficient redundancy coding across a larger collection of images. We propose a novel convolutional neural network model, Collage-CNN, to provide a low-cost redundancy framework. A Collage-CNN model takes a collage formed by combining multiple images and performs multi-image classification in one shot, albeit at slightly lower accuracy. We then augment a collection of traditional single image classifiers with a single Collage-CNN classifier which acts as a low-cost redundant backup. Collage-CNN then provides backup classification results if a single image classification straggles. Deploying the Collage-CNN models in the cloud, we demonstrate that the 99th percentile tail latency of inference can be reduced by 1.47X compared to replication based approaches while providing high accuracy. Also, variation in inference latency can be reduced by 9X with a slight increase in average inference latency.

cs.DC↗

arXiv2019

A Sampling Theory Perspective of Graph-based Semi-supervised Learning

Graph-based methods have been quite successful in solving unsupervised and semi-supervised learning problems, as they provide a means to capture the underlying geometry of the dataset. It is often desirable for the constructed graph to satisfy two properties: first, data points that are similar in the feature space should be strongly connected on the graph, and second, the class label information should vary smoothly with respect to the graph, where smoothness is measured using the spectral properties of the graph Laplacian matrix. Recent works have justified some of these smoothness conditions by showing that they are strongly linked to the semi-supervised smoothness assumption and its variants. In this work, we reinforce this connection by viewing the problem from a graph sampling theoretic perspective, where class indicator functions are treated as bandlimited graph signals (in the eigenvector basis of the graph Laplacian) and label prediction as a bandlimited reconstruction problem. Our approach involves analyzing the bandwidth of class indicator signals generated from statistical data models with separable and nonseparable classes. These models are quite general and mimic the nature of most real-world datasets. Our results show that in the asymptotic limit, the bandwidth of any class indicator is also closely related to the geometry of the dataset. This allows one to theoretically justify the assumption of bandlimitedness of class indicator signals, thereby providing a sampling theoretic interpretation of graph-based semi-supervised classification.

cs.LG↗

arXiv2019

Lagrange Coded Computing: Optimal Design for Resiliency, Security and Privacy

We consider a scenario involving computations over a massive dataset stored distributedly across multiple workers, which is at the core of distributed learning algorithms. We propose Lagrange Coded Computing (LCC), a new framework to simultaneously provide (1) resiliency against stragglers that may prolong computations; (2) security against Byzantine (or malicious) workers that deliberately modify the computation for their benefit; and (3) (information-theoretic) privacy of the dataset amidst possible collusion of workers. LCC, which leverages the well-known Lagrange polynomial to create computation redundancy in a novel coded form across workers, can be applied to any computation scenario in which the function of interest is an arbitrary multivariate polynomial of the input dataset, hence covering many computations of interest in machine learning. LCC significantly generalizes prior works to go beyond linear computations. It also enables secure and private computing in distributed settings, improving the computation and communication efficiency of the state-of-the-art. Furthermore, we prove the optimality of LCC by showing that it achieves the optimal tradeoff between resiliency, security, and privacy, i.e., in terms of tolerating the maximum number of stragglers and adversaries, and providing data privacy against the maximum number of colluding workers. Finally, we show via experiments on Amazon EC2 that LCC speeds up the conventional uncoded implementation of distributed least-squares linear regression by up to $13.43\times$, and also achieves a $2.36\times$-$12.65\times$ speedup over the state-of-the-art straggler mitigation strategies.

cs.IT↗

arXiv2019

Pipe-SGD: A Decentralized Pipelined SGD Framework for Distributed Deep Net Training

Distributed training of deep nets is an important technique to address some of the present day computing challenges like memory consumption and computational demands. Classical distributed approaches, synchronous or asynchronous, are based on the parameter server architecture, i.e., worker nodes compute gradients which are communicated to the parameter server while updated parameters are returned. Recently, distributed training with AllReduce operations gained popularity as well. While many of those operations seem appealing, little is reported about wall-clock training time improvements. In this paper, we carefully analyze the AllReduce based setup, propose timing models which include network latency, bandwidth, cluster size and compute time, and demonstrate that a pipelined training with a width of two combines the best of both synchronous and asynchronous training. Specifically, for a setup consisting of a four-node GPU cluster we show wall-clock time training improvements of up to 5.4x compared to conventional approaches.

cs.LG↗

arXiv2018

GradiVeQ: Vector Quantization for Bandwidth-Efficient Gradient Aggregation in Distributed CNN Training

Data parallelism can boost the training speed of convolutional neural networks (CNN), but could suffer from significant communication costs caused by gradient aggregation. To alleviate this problem, several scalar quantization techniques have been developed to compress the gradients. But these techniques could perform poorly when used together with decentralized aggregation protocols like ring all-reduce (RAR), mainly due to their inability to directly aggregate compressed gradients. In this paper, we empirically demonstrate the strong linear correlations between CNN gradients, and propose a gradient vector quantization technique, named GradiVeQ, to exploit these correlations through principal component analysis (PCA) for substantial gradient dimension reduction. GradiVeQ enables direct aggregation of compressed gradients, hence allows us to build a distributed learning system that parallelizes GradiVeQ gradient compression and RAR communications. Extensive experiments on popular CNNs demonstrate that applying GradiVeQ slashes the wall-clock gradient aggregation time of the original RAR by more than 5X without noticeable accuracy loss, and reduces the end-to-end training time by almost 50%. The results also show that GradiVeQ is compatible with scalar quantization techniques such as QSGD (Quantized SGD), and achieves a much higher speed-up gain under the same compression ratio.

cs.LG↗

arXiv2017

Distributed Solution of Large-Scale Linear Systems via Accelerated Projection-Based Consensus

Solving a large-scale system of linear equations is a key step at the heart of many algorithms in machine learning, scientific computing, and beyond. When the problem dimension is large, computational and/or memory constraints make it desirable, or even necessary, to perform the task in a distributed fashion. In this paper, we consider a common scenario in which a taskmaster intends to solve a large-scale system of linear equations by distributing subsets of the equations among a number of computing machines/cores. We propose an accelerated distributed consensus algorithm, in which at each iteration every machine updates its solution by adding a scaled version of the projection of an error signal onto the nullspace of its system of equations, and where the taskmaster conducts an averaging over the solutions with momentum. The convergence behavior of the proposed algorithm is analyzed in detail and analytically shown to compare favorably with the convergence rate of alternative distributed methods, namely distributed gradient descent, distributed versions of Nesterov's accelerated gradient descent and heavy-ball method, the block Cimmino method, and ADMM. On randomly chosen linear systems, as well as on real-world data sets, the proposed method offers significant speed-up relative to all the aforementioned methods. Finally, our analysis suggests a novel variation of the distributed heavy-ball method, which employs a particular distributed preconditioning, and which achieves the same theoretical convergence rate as the proposed consensus-based method.

cs.LG↗

arXiv2016

Active Learning for Community Detection in Stochastic Block Models

The stochastic block model (SBM) is an important generative model for random graphs in network science and machine learning, useful for benchmarking community detection (or clustering) algorithms. The symmetric SBM generates a graph with $2n$ nodes which cluster into two equally sized communities. Nodes connect with probability $p$ within a community and $q$ across different communities. We consider the case of $p=a\ln (n)/n$ and $q=b\ln (n)/n$. In this case, it was recently shown that recovering the community membership (or label) of every node with high probability (w.h.p.) using only the graph is possible if and only if the Chernoff-Hellinger (CH) divergence $D(a,b)=(\sqrt{a}-\sqrt{b})^2 \geq 1$. In this work, we study if, and by how much, community detection below the clustering threshold (i.e. $D(a,b)<1$) is possible by querying the labels of a limited number of chosen nodes (i.e., active learning). Our main result is to show that, under certain conditions, sampling the labels of a vanishingly small fraction of nodes (a number sub-linear in $n$) is sufficient for exact community detection even when $D(a,b)<1$. Furthermore, we provide an efficient learning algorithm which recovers the community memberships of all nodes w.h.p. as long as the number of sampled points meets the sufficient condition. We also show that recovery is not possible if the number of observed labels is less than $n^{1-D(a,b)}$. The validity of our results is demonstrated through numerical experiments.

cs.LG↗

arXiv2015

MISO Broadcast Channel with Hybrid CSIT: Beyond Two Users

We study the impact of heterogeneity of channel-state-information available at the transmitters (CSIT) on the capacity of broadcast channels with a multiple-antenna transmitter and $k$ single-antenna receivers (MISO BC). In particular, we consider the $k$-user MISO BC, where the CSIT with respect to each receiver can be either instantaneous/perfect, delayed, or not available; and we study the impact of this heterogeneity of CSIT on the degrees-of-freedom (DoF) of such network. We first focus on the $3$-user MISO BC; and we completely characterize the DoF region for all possible heterogeneous CSIT configurations, assuming linear encoding strategies at the transmitters. The result shows that the state-of-the-art achievable schemes in the literature are indeed sum-DoF optimal, when restricted to linear encoding schemes. To prove the result, we develop a novel bound, called Interference Decomposition Bound, which provides a lower bound on the interference dimension at a receiver which supplies delayed CSIT based on the average dimension of constituents of that interference, thereby decomposing the interference into its individual components. Furthermore, we extend our outer bound on the DoF region to the general $k$-user MISO BC, and demonstrate that it leads to an approximate characterization of linear sum-DoF to within an additive gap of $0.5$ for a broad range of CSIT configurations. Moreover, for the special case where only one receiver supplies delayed CSIT, we completely characterize the linear sum-DoF.

cs.IT↗

arXiv2013

A Latent Social Approach to YouTube Popularity Prediction

Current works on Information Centric Networking assume the spectrum of caching strategies under the Least Recently/ Frequently Used (LRFU) scheme as the de-facto standard, due to the ease of implementation and easier analysis of such strategies. In this paper we predict the popularity distribution of YouTube videos within a campus network. We explore two broad approaches in predicting the popularity of videos in the network: consensus approaches based on aggregate behavior in the network, and social approaches based on the information diffusion over an implicit network. We measure the performance of our approaches under a simple caching framework by picking the k most popular videos according to our predicted distribution and calculating the hit rate on the cache. We develop our approach by first incorporating video inter-arrival time (based on the power-law distribution governing the transmission time between two receivers of the same message in scale-free networks) to the baseline (LRFU), then combining with an information diffusion model over the inferred latent social graph that governs diffusion of videos in the network. We apply techniques from latent social network inference to learn the sharing probabilities between users in the network and apply a virus propagation model borrowed from mathematical epidemiology to estimate the number of times a video will be accessed in the future. Our approach gives rise to a 14% hit rate improvement over the baseline.

cs.SI↗

arXiv2011

Low-Complexity Near-Optimal Codes for Gaussian Relay Networks

We consider the problem of information flow over Gaussian relay networks. Similar to the recent work by Avestimehr \emph{et al.} [1], we propose network codes that achieve up to a constant gap from the capacity of such networks. However, our proposed codes are also computationally tractable. Our main technique is to use the codes of Avestimehr \emph{et al.} as inner codes in a concatenated coding scheme.

cs.IT↗