1st March 2025, Las Vegas, Nevada, USA
ANT-ACE is currently being developed at the Ant Research Institute to establish a cutting-edge compiler ecosystem, specifically tailored to meet the diverse compilation needs of domain-specific languages and advanced computer architectures. ANT-ACE is dedicated to providing a stable and robust compiler infrastructure that features a flexible and extensible type system, complemented by robust support for domain-specific operations.
Leveraging the ANT-ACE compiler infrastructure, the Ant Research Institute has developed an FHE compiler designed to facilitate neural network inference on encrypted data. This compiler accepts ONNX models as input and produces optimized C/C++ programs as output, utilizing the institute’s own open-source FHE library. Initial evaluations conducted on CPUs indicate that ANT-ACE delivers a 2.24x performance enhancement in processing ResNet models, surpassing the results of expert-coded implementations.
This tutorial is based on the CGO’25 Tool Paper titled “An FHE Compiler Framework for Automating Neural Network Inference,” recently accepted for presentation at CGO’25. The primary goal of this tutorial is to introduce participants to the ANT-ACE compiler ecosystem, emphasizing the design objectives and foundational principles of the infrastructure. We will discuss how to enhance this infrastructure to create a domain-specific compiler. The session will provide a detailed examination of both the compiler infrastructure and the FHE compiler. Participants will learn how a pre-trained neural network model, provided in an ONNX file, is compiled into a Fully Homomorphic Encryption (FHE) program, enabling the performance of neural network inference on encrypted data. This thorough overview is designed to equip attendees with the knowledge necessary to understand and potentially extend the ANT-ACE ecosystem for their own domain-specific applications.
The target audience for this tutorial encompasses researchers and graduate students specializing in compiler technology and Fully Homomorphic Encryption (FHE), with particular interest in cryptographic application development, hardware accelerator research, and cryptographic algorithm research.