.png){kind=icon}
.png){kind=icon}
Return
SEArch: A self-evolving framework for network architecture optimization
.png){kind=icon}
OA
AI
DOI：10.1016/j.neucom.2025.130980.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose a self-evolving network architecture optimization framework incorporating the advantages of network architecture pruning, knowledge distillation, and neural architecture search approaches. It can optimize an existing network into a more efficient network with better accuracy and smaller parameter size. • We propose two new designs for network architecture optimization, the first one is the attention layer to identify the bottleneck of the network performance; the second one is edge-splitting scheme for efficient network modification and construction. • We demonstrate our framework through comprehensive experiments achieving state-of-the-art accuracy and parameter sizes compared to existing network pruning and knowledge distillation algorithms.
Keywords:
Deep neural network
Knowledge distillation
Neural network pruning
Automated machine learning
Neural architecture search
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
Organization
No organization information available