Rafael Pereira Pires

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

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

NVCache: A Plug-and-Play NVMM-based I/O Booster for Legacy Systems

Pascal Felber, Rafael Pereira Pires

This paper introduces NVCache, an approach that uses a non-volatile main memory (NVMM) as a write cache to improve the write performance of legacy applications. We compare NVCache against file systems tailored for NVMM (Ext4-DAX and NOVA) and with I/O-heavy applications (SQLite, RocksDB). Our evaluation shows that NVCache reaches the performance level of the existing state-of-the-art systems for NVMM, but without their limitations: NVCache does not limit the size of the stored data to the size of the NVMM, and works transparently with unmodified legacy applications, providing additional persistence guarantees even when their source code is not available.

IEEE2021

Malware in the SGX supply chain: Be careful when signing enclaves!

Pascal Felber, Rafael Pereira Pires

Malware attacks are a significant part of the new software security threats detected each year. Intel Software Guard Extensions (SGX) are a set of hardware instructions introduced by Intel in their recent lines of processors that are intended to provide a secure execution environment for user-developed applications. To our knowledge, there was no serious attempt yet to overcome the SGX protection by exploiting the weaknesses in the software supply chain infrastructure, namely at the level of the development, build or signing servers. While SGX protection does not specifically take into consideration such threats, we show in the current paper that a simple malware attack exploiting a separation between the build and signing processes can have a serious damaging impact, practically nullifying SGX integrity protection measures. We also explore two possible mitigations against the attack, one centralized leveraging SGX itself, and one distributed that relies on a smart contract deployed on a blockchain infrastructure. Our evaluation shows that both methods are feasible in practice and their added costs are acceptable for the offered protection.

2020