A naukri.com initiative
Arxiv
2d
31
Image Credit: Arxiv
ICE-Pruning: An Iterative Cost-Efficient Pruning Pipeline for Deep Neural Networks
- ICE-Pruning is an iterative pruning pipeline designed to compress Deep Neural Networks efficiently.
- It reduces the computational cost of fine-tuning while maintaining a similar model accuracy as existing pruning pipelines.
- ICE-Pruning comprises automatic mechanisms for determining when to perform fine-tuning, a freezing strategy for faster fine-tuning, and a custom pruning-aware learning rate scheduler.
- Evaluations on various DNN models and datasets show that ICE-Pruning can accelerate the pruning process by up to 9.61x.
Read Full Article
1 Like
For uninterrupted reading, download the app