Data Reconstruction from Gradient Updates in Federated Learning

Abstract

The explosive growth and diversity of machine learning applications motivate a fundamental rethinking of learning with mobile and edge devices. How can we address diverse/disparate client goals and learn with scarce heterogeneous data? While federated learning aims to address these issues, it has several bottlenecks and challenges hindering a unified solution. On the other hand, large transformer models have been shown to work across a variety of tasks often achieving remarkable few-shot adaptation. This raises the question: Can clients use a single generalpurpose model -rather than custom models for each task -while obeying device and network constraints? In this work, we investigate pretrained transformers (PTF) to achieve these on-device learning goals and thoroughly explore the roles of model size and modularity, where the latter refers to adaptation through modules such as prompts or adapters. Focusing on federated learning, we demonstrate that:(1) Larger scale shrinks the accuracy gaps between alternative approaches and improves heterogeneity robustness. Scale allows clients to run more local SGD epochs which can significantly reduce the number of communication rounds. At the extreme, clients can achieve respectable accuracy fully-locally highlighting the potential of fully-local learning. (2) Modularity, by design, enables >100× less communication in bits. Surprisingly, it also boosts the generalization capability of local adaptation methods and the robustness of smaller PTFs. Finally, it enables clients to solve multiple unrelated tasks simultaneously using a single PTF, whereas full updates are prone to catastrophic forgetting. These insights on scale and modularity motivate a new federated learning approach we call "You Only Load Once" (FedYolo): The clients load a full PTF model once and all future updates are accomplished through communication-efficient modules with limited catastrophic-forgetting, where each task is assigned to its own module.

Exact Policy Recovery in Offline RL with Both Heavy-Tailed Rewards and Data Corruption

“…Recent advances (Diakonikolas et al 2019a;Lai, Rao, and Vempala 2016) design efficient algorithms for high-dimensional robust statistics. These techniques are applied to more general machine learning tasks, including linear regression (Diakonikolas, Kong, and Stewart 2019), supervised learning (Diakonikolas et al 2019b;Prasad et al 2018) and RL (Zhang et al 2022(Zhang et al , 2021. Our work utilizes robust mean estimation to defend data corruption in offline RLs.…”

“…Adversarial RL and robust RL: RL is vulnerable to adversarial attacks (Ma et al 2019;Zhang et al 2020;Huang et al 2017;Behzadan and Munir 2017). Corruption robust RL performs policy learning under data corruption (Lykouris et al 2021;Wei, Dann, and Zimmert 2022;Zhang et al 2021Zhang et al , 2022Chen et al 2022), which usually results in a bias term in the performance guarantee due to the data corruption. (Niss and Tewari 2020;Kapoor, Patel, and Kar 2019) study multi-armed bandits under data corruption using robust statistics.…”

Fast and Accurate SNN Model Strengthening for Industrial Applications

“…Security threats and model strengthening: There are emerging security threats in deep learning [1,2], such as the well-known backdoor attacks [4][5][6], membership inference attacks [7], adversarial sample attacks [8,9], data reconstruction attacks [10][11][12], etc., leading to serious information leakage or degraded system performance [13]. Previous solutions have focused on how to detect these security threats to notify the trainer.…”