BingoCGN graph neural network accelerator (IMAGE)

Institute of Science Tokyo

BingoCGN graph neural network accelerator

Caption

BingoCGN employs cross-partition message quantization to summarize inter-partition message flow, which eliminates the need for irregular off-chip memory access and utilizes a fine-grained structured strong lottery theory-based training algorithm to improve computational efficiency. 

Credit

Institute of Science Tokyo, Japan

Usage Restrictions

For use in scientific news only. Attribution is required. The image may not be modified in ways that alter its meaning or context.

License

Original content

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.