Anne-Marie Kermarrec

Decentralized learning made easy with DecentralizePy

Akash Balasaheb Dhasade, Anne-Marie Kermarrec, Rafael Pereira Pires, Rishi Sharma, Milos Vujasinovic

Decentralized learning (DL) has gained prominence for its potential benefits in terms of scalability, privacy, and fault tolerance. It consists of many nodes that coordinate without a central server and exchange millions of parameters in the inherently iterative process of machine learning (ML) training. In addition, these nodes are connected in complex and potentially dynamic topologies. Assessing the intricate dynamics of such networks is clearly not an easy task. Often in literature, researchers resort to simulated environments that do not scale and fail to capture practical and crucial behaviors, including the ones associated to parallelism, data transfer, network delays, and wall-clock time. In this paper, we propose DecentralizePy, a distributed framework for decentralized ML, which allows for the emulation of large-scale learning networks in arbitrary topologies. We demonstrate the capabilities of DecentralizePy by deploying techniques such as sparsification and secure aggregation on top of several topologies, including dynamic networks with more than one thousand nodes.

2023

Get More for Less in Decentralized Learning Systems

Akash Balasaheb Dhasade, Anne-Marie Kermarrec, Rafael Pereira Pires, Rishi Sharma, Milos Vujasinovic, Jeffrey Wigger

Decentralized learning (DL) systems have been gaining popularity because they avoid raw data sharing by communicating only model parameters, hence preserving data confidentiality. However, the large size of deep neural networks poses a significant challenge for decentralized training, since each node needs to exchange gigabytes of data, overloading the network. In this paper, we address this challenge with JWINS, a communication-efficient and fully decentralized learning system that shares only a subset of parameters through sparsification. JWINS uses wavelet transform to limit the information loss due to sparsification and a randomized communication cut-off that reduces communication usage without damaging the performance of trained models. We demonstrate empirically with 96 DL nodes on non-IID datasets that JWINS can achieve similar accuracies to full-sharing DL while sending up to 64% fewer bytes. Additionally, on low communication budgets, JWINS outperforms the state-of-the-art communication-efficient DL algorithm CHOCO-SGD by up to 4x in terms of network savings and time.

2023

D-Cliques: Compensating for Data Heterogeneity with Topology in Decentralized Federated Learning

Anne-Marie Kermarrec, Erick Lavoie

The convergence speed of machine learning models trained with Federated Learning is significantly affected by non-independent and identically distributed (non-IID) data partitions, even more so in a fully decentralized setting without a central server. In this paper, we show that the impact of local class bias, an important type of data non-IIDness, can be significantly reduced by carefully designing the underlying communication topology. We present D-Cliques, a novel topology that reduces gradient bias by grouping nodes in interconnected cliques such that the local joint distribution in a clique is representative of the global class distribution. We also show how to adapt the updates of decentralized SGD to obtain unbiased gradients and implement an effective momentum with D-Cliques. Our empirical evaluation on MNIST and CIFAR10 demonstrates that our approach provides similar convergence speed as a fully-connected topology with a significant reduction in the number of edges and messages. In a 1000-node topology, D-Cliques requires 98% less edges and 96% less total messages, with further possible gains using a small-world topology across cliques.

2022

TEE-based decentralized recommender systems: The raw data sharing redemption

Akash Balasaheb Dhasade, Nevena Dresevic, Anne-Marie Kermarrec, Rafael Pereira Pires

Recommenders are central in many applications today. The most effective recommendation schemes, such as those based on collaborative filtering (CF), exploit similarities between user profiles to make recommendations, but potentially expose private data. Federated learning and decentralized learning systems address this by letting the data stay on user's machines to preserve privacy: each user performs the training on local data and only the model parameters are shared. However, sharing the model parameters across the network may still yield privacy breaches. In this paper, we present REX, the first enclave-based decentralized CF recommender. REX exploits Trusted execution environments (TEE), such as Intel software guard extensions (SGX), that provide shielded environments within the processor to improve convergence while preserving privacy. Firstly, REX enables raw data sharing, which ultimately speeds up convergence and reduces the network load. Secondly, REX fully preserves privacy. We analyze the impact of raw data sharing in both deep neural network (DNN) and matrix factorization (MF) recommenders and showcase the benefits of trusted environments in a full-fledged implementation of REX. Our experimental results demonstrate that through raw data sharing, REX significantly decreases the training time by 18.3x and the network load by 2 orders of magnitude over standard decentralized approaches that share only parameters, while fully protecting privacy by leveraging trustworthy hardware enclaves with very little overhead.

2022

The Universal Gossip Fighter

Rachid Guerraoui, Anne-Marie Kermarrec, Anastasiia Kucherenko, Rafaël Benjamin Pinot

The notion of adversary is a staple of distributed computing. An adversary typically models “hostile” assumptions about the underlying distributed environment, e.g., a network that can drop messages, an operating system that can delay processes or an attacker that can hack machines. So far, the goal of distributed computing researchers has mainly been to develop a distributed algorithm that can face a given adversary, the abstraction characterizing worst-case scenarios. This paper initiates the study of the somehow opposite approach. Given a distributed algorithm, the adversary is the abstraction we seek to implement. More specifically, we consider the problem of controlling the spread of messages in a large- scale system, conveying the practical motivation of limiting the dissemination of fake news or viruses. Essentially, we assume a general class of gossip protocols, called all-to-all gossip protocols, and devise a practical method to hinder the dissemination. We present the Universal Gossip Fighter (UGF). Just like classical adversaries in distributed computing, UGF can observe the status of a dissemination and decide to stop some processes or delay some messages. The originality of UGF lies in the fact that it is universal, i.e., it applies to any all-to-all gossip protocol. We show that any gossip protocol attacked by UGF ends up exhibiting a quadratic message complexity (in the total number of processes) if it achieves sublinear time of dissemination. We also show that if a gossip protocol aims to achieve a message complexity α times smaller than quadratic, then the time complexity rises exponentially in relation to α. We convey the practical relevance of our theoretical findings by implementing UGF and conducting a set of empirical experiments that confirm some of our results.

2022

Cluster-and-Conquer: When Randomness Meets Graph Locality

Anne-Marie Kermarrec, Olivier Ruas

K-Nearest-Neighbors (KNN) graphs are central to many emblematic data mining and machine-learning applications. Some of the most efficient KNN graph algorithms are incremental and local: they start from a random graph, which they incrementally improve by traversing neighbors-of-neighbors links. Unfortunately, the initial random graph exhibits a poor graph locality, leading to many unnecessary similarity computations. In this paper, we remove this drawback with Cluster-and-Conquer (C-2 for short). Cluster-and-Conquer boosts the starting configuration of greedy algorithms thanks to a novel lightweight clustering mechanism, dubbed FastRandomHash. FastRandomHash leverages randomness and recursion to precluster similar nodes at a very low cost. Our extensive evaluation on real datasets shows that Cluster-and-Conquer significantly outperforms existing approaches, including LSH, yielding speedups of up to x4.42 and even improving the KNN quality.

IEEE COMPUTER SOC2021

FeGAN: Scaling Distributed GANs

Rachid Guerraoui, Arsany Hany Abdelmessih Guirguis, Anne-Marie Kermarrec

Existing approaches to distribute Generative Adversarial Networks (GANs) either (i) fail to scale for they typically put the two components of a GAN (the generator and the discriminator) on different machines, inducing significant communication overhead, or (ii) they face GAN training specific issues, exacerbated by distribution. We propose FeGAN, the first middleware for distributing GANs over hundreds of devices addressing the issues of mode collapse and vanishing gradients. Essentially, we revisit the idea of Federated Learning, co-locating a generator with a discriminator on each device (addressing the scaling problem) and having a server aggregate the devices' models using balanced sampling and Kullback-Leibler (KL) weighting, mitigating training issues and boosting convergence. Through extensive experiments, we show that FeGAN generates high-quality dataset samples in a scalable and devices' heterogeneity tolerant manner. In particular, FeGAN achieves up to 5× throughput gain with 1.5× less bandwidth compared to the state-of-the-art GAN distributed approach (named MD-GAN), while scaling to at least one order of magnitude more devices. We demonstrate that FeGAN boosts training by 2.6× w.r.t. a baseline application of Federated Learning to GANs while preventing training issues.

2020

FLeet: Online Federated Learning via Staleness Awareness and Performance Prediction

Georgios Damaskinos, Rachid Guerraoui, Anne-Marie Kermarrec, Vlad-Tiberiu Nitu, Rhicheek Patra

Federated Learning (FL) is very appealing for its privacy benefits: essentially, a global model is trained with updates computed on mobile devices while keeping the data of users local. Standard FL infrastructures are however designed to have no energy or performance impact on mobile devices, and are therefore not suitable for applications that require frequent (online) model updates, such as news recommenders. This paper presents FLeet, the first Online FL system, acting as a middleware between the Android OS and the machine learning application. FLeet combines the privacy of Standard FL with the precision of online learning thanks to two core components: (i) I-Prof, a new lightweight profiler that predicts and controls the impact of learning tasks on mobile devices, and (ii) AdaSGD, a new adaptive learning algorithm that is resilient to delayed updates. Our extensive evaluation shows that Online FL, as implemented by FLeet, can deliver a 2.3× quality boost compared to Standard FL, while only consuming 0.036% of the battery per day. I-Prof can accurately control the impact of learning tasks by improving the prediction accuracy up to 3.6× (computation time) and up to 19× (energy). AdaSGD outperforms alternative FL approaches by 18.4% in terms of convergence speed on heterogeneous data.

Association for Computing Machinery2020

Smaller, Faster & Lighter KNN Graph Constructions

Rachid Guerraoui, Anne-Marie Kermarrec, Olivier Ruas

We propose GoldFinger, a new compact and fast-to-compute binary representation of datasets to approximate Jaccard’s index. We illustrate the effectiveness of GoldFinger on the emblematic big data problem of K-Nearest-Neighbor (KNN) graph construction and show that GoldFinger can drastically accelerate a large range of existing KNN algorithms with little to no overhead. As a side effect, we also show that the compact representation of the data protects users’ privacy for free by providing k-anonymity and l-diversity. Our extensive evaluation of the resulting approach on several realistic datasets shows that our approach delivers speedups of up to 78.9% compared to the use of raw data while only incurring a negligible to moderate loss in terms of KNN quality. To convey the practical value of such a scheme, we apply it to item recommendation and show that the loss in recommendation quality is negligible.

2020