Efficient Pruning for Machine Learning under Homomorphic Encryption

Abstract

Homomorphic encryption (HE) based privacy-preserving machine learning (PPML) solutions are gaining widespread popularity. However, model sizes in PPML are limited by computational speed and memory requirements for inference. Pruning the model parameters improves latency and memory in plaintext ML, but has little impact if directly applied to HE-based PPML. This presentation introduces a framework called HE-PEx that carefully applies parameter pruning and permutation methods, on top of a packing technique called tile tensors, for reducing the latency and memory of PPML inference.

About the speakers

Dr. Subhankar Pal is a research scientist at IBM T.J. Watson Research Center in Yorktown Heights, USA. He currently works on hardware and software co-designed approaches for accelerating homomorphic encryption, and SoC-level design methodologies. In the past, he has worked on reconfigurable computer architectures, machine learning hardware and compiler techniques, resource scheduling, among others.

Unable to make it?

After the event has finished, resources (including slides, video, papers, etc) will be available at the resource page at https://fhe.org/meetups/041-Efficient_Pruning_for_Machine_Learning_under_Homomorphic_Encryption

Never miss an update

1. The newsletter where we post community announcements: https://fheorg.substack.com/
2. The discord server where you can discuss FHE related topics with the community: https://discord.fhe.org

Make sure to join either (or both) of these to stay informed about future events!