Redirigiendo al acceso original de articulo en 17 segundos...
ARTÍCULO
TITULO

Efficient and Controllable Model Compression through Sequential Knowledge Distillation and Pruning

Leila Malihi and Gunther Heidemann    

Resumen

Efficient model deployment is a key focus in deep learning. This has led to the exploration of methods such as knowledge distillation and network pruning to compress models and increase their performance. In this study, we investigate the potential synergy between knowledge distillation and network pruning to achieve optimal model efficiency and improved generalization. We introduce an innovative framework for model compression that combines knowledge distillation, pruning, and fine-tuning to achieve enhanced compression while providing control over the degree of compactness. Our research is conducted on popular datasets, CIFAR-10 and CIFAR-100, employing diverse model architectures, including ResNet, DenseNet, and EfficientNet. We could calibrate the amount of compression achieved. This allows us to produce models with different degrees of compression while still being just as accurate, or even better. Notably, we demonstrate its efficacy by producing two compressed variants of ResNet 101: ResNet 50 and ResNet 18. Our results reveal intriguing findings. In most cases, the pruned and distilled student models exhibit comparable or superior accuracy to the distilled student models while utilizing significantly fewer parameters.

 Artículos similares

       
 
Andrea Momblanch, Ian P. Holman and Sanjay K. Jain    
Global change is expected to have a strong impact in the Himalayan region. The climatic and orographic conditions result in unique modelling challenges and requirements. This paper critically appraises recent hydrological modelling applications in Himala... ver más
Revista: Water

 
Margarita Garcia-Vila, Rodrigo Morillo-Velarde and Elias Fereres    
Process-based crop models such as AquaCrop are useful for a variety of applications but must be accurately calibrated and validated. Sugar beet is an important crop that is grown in regions under water scarcity. The discrepancies and uncertainty in past ... ver más
Revista: Water

 
Piotr Parasiewicz, Pawel Prus, Christos Theodoropoulos, Knut Alfredsen, Mikolaj Adamczyk, Claudio Comoglio and Paolo Vezza    
Hydraulic habitat simulation models were designed for the quantitative determination of environmental flows that consider the needs of aquatic fauna in rivers and streams. In the past 50 years, the modeling techniques were significantly developed, but ex... ver más
Revista: Water

 
Yaru Guo, Tian-Chyi Jim Yeh and Yonghong Hao    
Karst aquifers are prominent sources of water worldwide; they store large amounts of water and are known for their beautiful springs. However, extensive groundwater development and climate variation has resulted in a decline in the flow of most karst spr... ver más
Revista: Water

 
Haekyung Park and Dong Kun Lee    
Korea experienced an unexpected drought in the southern Seoul metropolitan area from 2015 to 2017. After that, the Korean government has been drafting various policies to mitigate the effects of drought. However, these are primarily long-term drought pol... ver más
Revista: Water