Optimiranje konvolucijskih modela iterativnim podrezivanjem

master's thesis

• Title (english): Optimizing convolutional models by iterative pruning
• Author: Stjepan Močilac
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Committee member: Vladimir Čeperić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Ovaj rad se bavi pronalažanjem podmreža malog kapaciteta i dobre generalizacije unutar većeg dubokog modela. Pretraživanje se obavlja postupkom strukturnog podrezivanja i podrezivanja težina. Eksperimenti su rađeni nad VGG-19 i ResNet-18 arhitekturama. Pokazuje se da postoji podmreža mreže ResNet-18 koja nad CIFAR-10 skupom za učenje postiže 1.37\% veću točnost i ima 20\% parametara originalnog modela. Uz to, takva podmreža postiže 2.5 puta bržu evaluaciju na GPU uređaju od originalne mreže.

Abstract (english)

The goal of this thesis is to find small capacity and good generalization subnet within a larger deep network. The search is performed by the process of structural and weight pruning. Experiments were done using VGG-19 and ResNet-18 architectures. It is shown that there is a sub-network of the ResNet-18 network that achieves 1.37 \% higher accuracy on the CIFAR-10 dataset and has 20 \% parameters of the original model. Also, it achieves 2.5 times faster evaluations on the GPU device than the original network.

• Keywords: Duboko učenje; podrezivanje; neuronska mreža; hipoteza dobitnog listića lutrije; CNN; VGG; ResNet
• Keywords (english): Deep learning; pruning; neural network; the lottery ticket hypothesis; CNN; VGG; ResNet
• Language: croatian
• URN:NBN: urn:nbn:hr:168:003963
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 10:29:27