Name
Abstract | ||
|---|---|---|
ABSTRACTRecently one deep learning technique, Convolutional Neural Networks (CNN), has gained immense popularity. Their success is particularly noticeable on image data, but falls short on non-image data. New methods have been developed to transform non-image data to exhibit image like local structures. That would enable the transformed data to take advantage of CNN architectures. Question then arises, how to measure the presence of local structures, the quality of those local structures, and how to know if there is any optimal shape of the local structures that might result in superior performance for CNN. In this paper, we answer these three questions. We present three methods to identify presence of local structures by measuring entropy. We show experimental results that provide intuitions about the quality of the local structures. Finally, we provide results showing that the performances of CNN models corresponding to the lowest entropy producing datasets were superior. |
Year | DOI | Venue |
|---|---|---|
2021 | 10.1145/3409334.3452061 | ACM-SE |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 2 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Torumoy Ghoshal | 1 | 0 | 0.34 |
| Yixin Chen | 2 | 4326 | 299.19 |