Logo image
Home Academic units
Sign in
Deep Convolutional NMF Net
Thesis   Open access

Deep Convolutional NMF Net

Hangyan Jiang
Brandeis University
Bachelor of Science (BS), Brandeis University, School of Arts and Sciences
2018
Handle:
https://hdl.handle.net/10192/36632

Abstract

In this paper, we explored the possible buildup of a neural network based on Non-negative Matrix Factorization (NMF). We constructed a neural network with NMF layers that process input data in a convolutional manner. This ensures those novel layers first focus features locally and gradually expand their scope as they stack on each other. After testing we found that NMF preserves the original signal very well and has the potential to accelerate network.
pdf
JiangThesis2018473.93 kBDownloadView
PDF Open Access

Metrics

40 File views/ downloads
131 Record Views

Details

Logo image