Continual Learning with Differential Privacy

Document Type

Conference Proceeding

Publication Date

1-1-2021

Abstract

In this paper, we focus on preserving differential privacy (DP) in continual learning (CL), in which we train ML models to learn a sequence of new tasks while memorizing previous tasks. We first introduce a notion of continual adjacent databases to bound the sensitivity of any data record participating in the training process of CL. Based upon that, we develop a new DP-preserving algorithm for CL with a data sampling strategy to quantify the privacy risk of training data in the well-known Averaged Gradient Episodic Memory (A-GEM) approach by applying a moments accountant. Our algorithm provides formal guarantees of privacy for data records across tasks in CL. Preliminary theoretical analysis and evaluations show that our mechanism tightens the privacy loss while maintaining a promising model utility.

Identifier

85121908638 (Scopus)

ISBN

[9783030923099]

Publication Title

Communications in Computer and Information Science

External Full Text Location

https://doi.org/10.1007/978-3-030-92310-5_39

e-ISSN

18650937

ISSN

18650929

First Page

334

Last Page

343

Volume

1517 CCIS

Grant

2041096

Fund Ref

National Science Foundation

This document is currently not available here.

Share

COinS