Abstract
In this paper, we propose a Deep Neural Networks (DNN) Design Index which would aid a DNN designer during the designing phase of DNNs. We study the designing aspect of DNNs from model-specific and data-specific perspectives with focus on three performance metrics: training time, training error and, validation error. We use a simple example to illustrate the significance of the DNN design index. To validate it, we calculate the design indices for four benchmark problems. This is an elementary work aimed at setting a direction for creating design indices pertaining to deep learning.
Original language | English |
---|---|
Pages (from-to) | 131-138 |
Number of pages | 8 |
Journal | Procedia Computer Science |
Volume | 88 |
DOIs | |
State | Published - 2016 |
Externally published | Yes |
Event | 7th Annual International Conference on Biologically Inspired Cognitive Architectures, BICA 2016 - Newyok City, United States Duration: Jul 16 2016 → Jul 19 2016 |
Funding
The research was supported by funding from the Air Force Research Lab (AFRL). We are grateful to Rensselaer Polytechnic Institute for providing the state of the art infrastructure, without which this study would not have been possible.
Keywords
- Deep Learning
- Deep Neural Networks
- Design Index
- Neural Network Design
- Performance Evaluation