Spiral AI/ML: Co-Optimization for High-Performance, Data-Intensive Computing in Resource Constrained Environments
Abstract
Problem(s): Increasing complexity in computing architectures; Mission cost, size, weight, and power (CSWAP) constraints drive increasing use of FPGAs and ASICs (more complexity); Achieving performance from these platforms is hard; Achieving performance from data-intensive applications (graphs, ML, AI) is hard. Solution: Automatic code generation for data-intensive computations; Simultaneous, automatic co-optimization of hardware within CSWAP constraints. Approach: Identify common AI/ML/Graph computational primitives. Encode knowledge about graph, ML, and AI computational primitives into Spiral code-gen technology; Develop hardware performance models allowing Spiral to choose between components satisfying CSWAP requirements.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2019
- Accession Number
- AD1118561
Entities
People
- Franz Franchetti
- James C. Hoe
- Scott McMillan
- Tze M. Low
Organizations
- Carnegie Mellon University