Loading...
Please wait, while we are loading the content...
Similar Documents
Extracting Rules from Deep Neural Networks
| Content Provider | Semantic Scholar |
|---|---|
| Author | Gloeckner, Peter Mencía, Eneldo Loza |
| Copyright Year | 2015 |
| Abstract | Neural network classifiers are known to be able to learn very accurate models in a wide variety of application domains. In the recent past, researchers have even been able to train neural networks with multiple hidden layers (deep neural networks) more efficiently. However, the major downside of neural networks is that it is not trivial to understand the way how they derive their classification decisions. To solve this problem, there has been research on extracting better understandable rules from neural networks. Although there are quite some algorithms available, to the best of our knowledge, none of them has ever been explicitly tested on deep neural networks. Furthermore, most authors focus on nets with only one single hidden layer, and hardly any paper even mentions deep neural networks. The present thesis fills in this gap and analyses the specific challenges of extracting rules from deep neural networks. A new algorithm – DeepRED – is presented that is able to perform this task. Its main strategy is to utilize C4.5 to extract layer-wise rules that are getting merged afterwards. The evaluation of the proposed algorithm shows its ability to outperform a baseline on several tasks. This work also provides general instructions on how to use an arbitrary deep neural network rule extraction algorithm to analyse single neurons in any neural network. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.ke.tu-darmstadt.de/lehre/arbeiten/master/2015/Zilke_Jan.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
