AI Technique Designs New Accelerators Based on Old Blueprints
Researchers at UC Berkeley and Google have announced a method for designing new accelerators using AI techniques. The new method creates architectures by drawing from existing blueprints and performance figures. According to the researchers, the approach has the ability to create accelerators that use less space and operate at lower latency. According to a post at the Google AI blog, the technique, known as PRIME, “...alleviates the need to run time-consuming simulations and enables reuse of data from past experiments, even when the set of target applications changes (e.g., an ML model for vision, language, or other objective), and even for unseen but related applications to the training set, in a zero-shot fashion.”
See the post at the Google website for more information.
Related content
- Google Announces the Google Cloud HPC Toolkit
- US DOE Sponsors Quantum Internet Blueprint Workshop
- Breakthrough in Diamond-Based Quantum Computing
-
GitLab for DevOps teams
We show you how one company uses GitLab CI as a platform for continuous integration and deployment processes. - New Technique Combines Resistive Memory with Graph Neural Networks