On dissipativity of cross-entropy loss in training ResNets — A turnpike towards architecture search

Key Points

• Dissipativity of cross-entropy loss positively influences model optimization, enhancing ResNet training.
• Effective architecture search may leverage the dissipative nature found in cross-entropy loss.
• Assessment involves mathematical modeling of training dynamics in deep learning frameworks like ResNets.
• Findings highlight the potential for improved neural network designs, emphasizing optimization strategies.