FedPEAT: Convergence of Federated Learning, Parameter-Efficient Fine Tuning, and Emulator Assisted Tuning for Artificial Intelligence Foundation Models with Mobile Edge Computing

TL;DR

FedPEAT combines federated learning, parameter-efficient fine-tuning, and emulator-assisted tuning to enable privacy-preserving, memory-efficient deployment and adaptation of large foundation models in mobile edge computing environments.

Contribution

This paper introduces FedPEAT, a novel framework integrating emulator-assisted tuning with federated learning for efficient, privacy-preserving foundation model adaptation.

Findings

FedPEAT enhances privacy and memory efficiency in federated foundation model tuning.

The approach is adaptable to various neural network architectures.

Experimental results demonstrate effective collaborative tuning with a server in federated settings.

Abstract

The emergence of foundation models, including language and vision models, has reshaped AI's landscape, offering capabilities across various applications. Deploying and fine-tuning these large models, like GPT-3 and BERT, presents challenges, especially in the current foundation model era. We introduce Emulator-Assisted Tuning (EAT) combined with Parameter-Efficient Fine-Tuning (PEFT) to form Parameter-Efficient Emulator-Assisted Tuning (PEAT). Further, we expand this into federated learning as Federated PEAT (FedPEAT). FedPEAT uses adapters, emulators, and PEFT for federated model tuning, enhancing model privacy and memory efficiency. Adapters adjust pre-trained models, while emulators give a compact representation of original models, addressing both privacy and efficiency. Adaptable to various neural networks, our approach also uses deep reinforcement learning for hyper-parameter optimization. We tested FedPEAT in a unique scenario with a server participating in collaborative federated tuning, showcasing its potential in tackling foundation model challenges.