Skip to content

Investigation on speeding up the forward pass computation for DNN "lottery tickets". Class project for ECE 6524 Sp'20

Notifications You must be signed in to change notification settings

kfallah/DNNLotteryTicket

Repository files navigation

DNNLoteryTicket

Investigation on speeding up the forward pass computation for DNN "lottery tickets". Class project for Georgia Tech ECE 6524 Sp'20.

Contributors: Daniel Bolya, Kion Fallah, Gregory Hessler, Shreyas Patil, Cameron Taylor

Experiment

This experiment is concerned with speeding the forward pass of sparse networks found by the lottery ticket hypothesis [1]. If a neural network with sparse matrices is found through the lottery ticket hypothesis, then the forward pass can be sped up through the use of structured pruning and sparse matrix multiplication.

This script proceeds by training on a DNN with fully-connected layers trained on MNIST digit classification. A pruned network is found using the original library ticket hypothesis and a new structured block pruning method. These methods are tested for forward pass speed.

Results

TBD

Running

Dependencies

Python 3.0+, PyTorch

Scripts

Testing forward pass speed:

python speed_test.py

Training baseline network:

python network.py

Pruning network:

python pruning.py

References

[1] Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neuralnetworks. In ICLR, 2018.

About

Investigation on speeding up the forward pass computation for DNN "lottery tickets". Class project for ECE 6524 Sp'20

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages