Loading...
Please wait, while we are loading the content...
Similar Documents
Opleiding Informatica Anomaly Detection with Deep Belief Networks
| Content Provider | Semantic Scholar |
|---|---|
| Author | Riet, Jasper Van |
| Copyright Year | 2017 |
| Abstract | When dealing with a large amount of data, using machine learning methods can often be appealing. When it concerns cybersecurity, that large volume of data is without a doubt present and so is an increasing amount of importance on safeguarding both public and private secrets. Problematic however, is the fact that machine learning ideally needs a labelled dataset that represent real-life conditions. When that is provided, in what is termed supervised learning, classifiers can achieve remarkable accuracy. Yet these labelled datasets are scarce. These datasets often contain highly sensitive data that first has to be anonymised before it can be released to the public. If it would be possible to have a similar accuracy without the need for a labelled dataset, that would be ideal. This thesis focuses on this concept of unsupervised learning. In particular, it concentrates its efforts on the applicability and feasibility of deep belief networks. These networks can ideally detect anomalies by learning a distribution of the data. If an event is not consistent with the rest of the data, a deep belief network can highlight that. In particular, this thesis trains a model for every single user from the dataset, and then proceeds to determine whether an hour of activity is anomalous behaviour or not. The results show that the usage of a deep belief network is feasible from a technical perspective, with training the model only taking a number of seconds, yet it is hard to determine the accuracy when dealing with an unlabelled dataset, thus leading the author to conclude that more research is needed. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://liacs.leidenuniv.nl/assets/Uploads/JaspervanRiet.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
