POSE: Phase II: An Open Source Ecosystem for Collaborative Rapid Design of Edge AI Hardware Accelerators for Integrated Data Analysis and Discovery

点击翻译按钮获取中文摘要

This project develops an ecosystem for hls4ml, which is a tool for designing machine learning inference in hardware. Machine learning models written in popular languages (e.g. PyTorch, Keras) are translated by hls4ml into specialty descriptions utilized by digital circuit designers. Hand crafting these system definitions is a process with a high barrier to entry, likely to result in poor quality results, for domain experts (e.g., scientists) without hardware design expertise. This project will innovate and deploy infrastructures and management solutions for support of the developers and users, leading to an extensive, connected, and well-supported ML hardware design automation community bridging hardware experts and domain experts. The project will develop a set of components comprising a support infrastructure for hls4ml. These components include training for users, automated testing and validation procedures for new components contributed to the tool, security validation of developers, creation of review and vetting procedures for quality control of the tool set, and a system for users and developers to report and request features. The ecosystem resulting from this project will manage access to a catalog of pre-designed and validated software packages and Intellectual Property hardware blocks, which can be used for both educational purposes and to build custom machine learning computational systems. The ecosystem will enable application and domain experts from a wide range of disciplines and affiliations (science, health, mobile, academic institutions, government laboratories, industry) to successfully utilize automated design flows to create customized machine learning hardware. This will enhance the productivity and overall ability of the underlying science discovery and technology development efforts, where these systems are deployed. The synergy catalyzed between domain experts and hardware experts will help these communities create powerful co-design methodologies and train the new generation of experts who will be proficient in applying them as part of the future workforce in data-driven disciplines. This collaborative project brings together investigators from Northwestern University, University of Illinois (Urbana-Champaign and Chicago), and Arizona State University. The project’s products and activities will be made available through https://fastmachinelearning.org/hls4ml/ . The project team plans to maintain the project repositories and website for a minimum of 3 years past the completion of this project.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.