Reinforcement learning for chip design
Practical AI
00:00
Exploring Alternatives in Chip Design
This chapter examines alternative techniques to reinforcement learning for chip design, focusing on evolutionary strategies and supervised learning. The discussion addresses the challenges of data labeling and generalization in architecture searches, while also analyzing the complexities of sequential decision-making akin to natural language processing. Additionally, it highlights the role of graph neural networks, emphasizing the importance of edge features and the training of architectures using pseudo labels to improve predictive capabilities.
Play episode from 17:34
Transcript
