What is it about?
Customized accelerators for Convolutional Neural Networks (CNN) can achieve better energy efficiency than general computing platforms. ACDSE provides an automatic optimization methods for CNN accelerators based on Reinforcement Learning, which may efficiently reduce the period and cost of accelerator development.
Featured Image
Photo by Laura Ockel on Unsplash
Why is it important?
ACDSE achieves its superior efficiency via adaptively compressing the design space of CNN accelerator. It can dynamic adjust the design space to concentrate its exploration on high-value subspace and filter out low-value design points. The adaptive compression also helps ACDSE keep efficiency in various circumstances with different constraint conditions.
Read the Original
This page is a summary of: ACDSE: A Design Space Exploration Method for CNN Accelerator based on Adaptive Compression Mechanism, ACM Transactions on Embedded Computing Systems, November 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3545177.
You can read the full text:
Contributors
The following have contributed to this page