Pacifo speeds encrypted ResNet20 inference while keeping most accuracy
Original title: Pacifo: An Efficient Fully Homomorphic Encryption-based Lightweight ResNet for Trustworthy Machine Learning as a Service
Runtime is 3.7 to 21.7 times faster than prior FHE ResNet20 work, retaining 95% to 99% plaintext accuracy.
Pacifo adapts fine-tuning so ResNet20 can be evaluated under fully homomorphic encryption for private machine-learning inference. A public-private layer split and a cheaper ReLU approximation reduce private layers and costly encrypted operations. Clients can select different security settings, and the code is published on GitHub.
- Only the final inference result is decrypted on the client.
- The evaluated model is ResNet20.
- The project is open-sourced as Pacifo.
- Flexible security configurations match different privacy needs.
From the original article
1 min read at arxiv.orgMachine Learning as a Service (MLaaS) has become a popular paradigm for deploying AI applications. However, it raises significant privacy concerns that the raw client data needs to be uploaded to the server for inference services. Private inference using Fully Homomorphic Encryption (FHE) offers a solution by allowing computations directly on encrypted data, with only the final results decrypted on a client side.
By Yang Li · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.