Characterizing the Training Dynamics of Private Fine-tuning with Langevin diffusion

2024/02/29 by Ke, Shuqi, Hou, Charlie, Oh, Sewoong +1 · 2 citations
#Artificial Intelligence (cs.AI) #Cryptography and Security (cs.CR) #FOS: Computer and information sciences #FOS: Mathematics #Machine Learning (cs.LG) #Optimization and Control (math.OC)

paper · doi:10.48550/arxiv.2402.18905

Abstract

We show that differentially private full fine-tuning (DP-FFT) can distort pre-trained backbone features based on both theoretical and empirical results. We identify the cause of the distortion as the misalignment between the pre-trained backbone and the randomly initialized linear head. We prove that a sequential fine-tuning strategy can mitigate the feature distortion: first-linear-probing-then-fine-tuning (DP-LP-FFT). A new approximation scheme allows us to derive approximate upper and lower bounds on the training loss of DP-LP and DP-FFT, in a simple but canonical setting of 2-layer neural networks with ReLU activation. Experiments on real-world datasets and architectures are consistent with our theoretical insights. We also derive new upper bounds for 2-layer linear networks without the approximation. Moreover, our theory suggests a trade-off of privacy budget allocation in multi-phase fine-tuning methods like DP-LP-FFT.

Cited by

• On Optimal Hyperparameters for Differentially Private Deep Transfer Learning
• Lower Bounds for Public-Private Learning under Distribution Shift

Related

• Fine-Tuning with Differential Privacy Necessitates an Additional Hyperparameter Search
• Differentially Private Neural Network Training Under the Hidden State Assumption
• Optimizer Dynamics at the Edge of Stability with Differential Privacy
• An Optimization Framework for Differentially Private Sparse Fine-Tuning
• Understanding Private Learning From Feature Perspective