search

TaehyeonKim-pyomu / CNN_compression_rank_selection_BayesOpt

Bayesian Optimization-Based Global Optimal Rank Selection for Compression of Convolutional Neural Networks, IEEE Access
MIT License
14 stars 4 forks source link

How to implement SVD single rank decomposition by partial_ tucker() #1

Closed filoliufei closed 2 years ago

filoliufei commented 3 years ago

In your paper, we see the performance of the proposed method in SVD. Is it implemented on the basis of Tucker decomposition? Or do you need to write separate SVD decomposition code?

TaehyeonKim-pyomu commented 3 years ago

The SVD experiment result was reported by a modified version of the uploaded code. I think changing my code to SVD-based CNN compression is not hard work. If you need further help, please let me know.

filoliufei commented 3 years ago

Hello, we are learning CNN compression based on SVD. Can you upload the modified version of the upload code(BayesOpt-based Global Optimal Rank Selection)?

filoliufei commented 3 years ago

Hello, we are very interested in the work of you and your team. What is the accuracy rate of SVD after decomposition and before fine tuning? Our test results show that, taking vgg19 as an example, the accuracy of SVD decomposition before fine tuning is basically 0. thank you!