Adaptive Variational Autoencoders for Outlier Detection in Data Streams: Unterschied zwischen den Versionen

Aktuelle Version vom 26. März 2019, 23:19 Uhr

Vortragende(r)	Florian Pieper
Vortragstyp	Masterarbeit
Betreuer(in)	Edouard Fouché
Termin	Fr 29. März 2019
Vortragsmodus
Kurzfassung	Outlier detection targets the discovery of abnormal data patterns. Typical scenarios, such as are fraud detection and predictive maintenance are particularly challenging, since the data is available as an infinite and ever evolving stream. In this thesis, we propose Adaptive Variational Autoencoders (AVA), a novel approach for unsupervised outlier detection in data streams. Our contribution is two-fold: (1) we introduce a general streaming framework for training arbitrary generative models on data streams. Here, generative models are useful to capture the history of the stream. (2) We instantiate this framework with a Variational Autoencoder, which adapts its network architecture to the dimensionality of incoming data. Our experiments against several benchmark outlier data sets show that AVA outperforms the state of the art and successfully adapts to streams with concept drift.

@@ Zeile 5: / Zeile 5: @@
 |betreuer=Edouard Fouché
 |termin=Institutsseminar/2019-03-29
-|kurzfassung=Outlier detection targets at the discovery of abnormal data patterns. Adaptive Variational Autoencoders (AVA) are a novel approach for unsupervised outlier detection in data streams.
+|kurzfassung=Outlier detection targets the discovery of abnormal data patterns. Typical scenarios, such as are fraud detection and predictive maintenance are particularly challenging, since the data is available as an infinite and ever evolving stream. In this thesis, we propose Adaptive Variational Autoencoders (AVA), a novel approach for unsupervised outlier detection in data streams.
-We show that they outperform recent state-of-the-art approaches on several datasets in a static as well as in a streaming setting.
-AVA exceeds the competition regarding streams with concept drift and an evolving feature space.
-Furthermore, we propose a new approach to automatically construct an appropriate autoencoder network architecture in a completely unsupervised manner only parametrized by the dimensionality of the incoming stream.
+Our contribution is two-fold: (1) we introduce a general streaming framework for training arbitrary generative models on data streams. Here, generative models are useful to capture the history of the stream. (2) We instantiate this framework with a Variational Autoencoder, which adapts its network architecture to the dimensionality of incoming data.
-We show via benchmarking that AVA is able to handle infinite data streams and fulfils all requirements for efficient stream processing such as constant memory consumption and constant processing times per point.
+Our experiments against several benchmark outlier data sets show that AVA outperforms the state of the art and successfully adapts to streams with concept drift.
 }}