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

Aus SDQ-Institutsseminar
Keine Bearbeitungszusammenfassung
Keine Bearbeitungszusammenfassung
Zeile 5: Zeile 5:
|betreuer=Edouard Fouché
|betreuer=Edouard Fouché
|termin=Institutsseminar/2019-03-29
|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 at the discovery of abnormal data patterns. Typical scenarios are intrusion or fraud detection and predictive maintenance.
Since in real-world settings the data is often available as an infinite and ever evolving stream,
in this thesis we propose Adaptive Variational Autoencoders (AVA) as novel approach for unsupervised outlier detection in data streams.


We present a streaming framework for training arbitrary generative models such as Variational Autoencoders (VAE) in data streams, which we name Adaptive Generative Networks (AGN).
First, we introduce a streaming framework for training arbitrary generative models on data streams.
The unique property of a generative model is its ability of generating samples from the model.
Generative models are used for sample generation to keep learned knowledge.


Adaptive Variational Autoencoders instantiate the AGN framework with a Variational Autoencoder as generative model and automatically adapt the model to an occurring concept drift.
Second, we instanciate this framework with a Variational Autoencoder as AVA. To handle data streams in a truely unsupervised manner, AVA adapts its network architecture to the dimensionality of incoming data.
AVA generates its network architecture based on 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.
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.
}}
}}

Version vom 26. März 2019, 16:16 Uhr

Vortragende(r) Florian Pieper
Vortragstyp Masterarbeit
Betreuer(in) Edouard Fouché
Termin Fr 29. März 2019
Vortragsmodus
Kurzfassung Outlier detection targets at the discovery of abnormal data patterns. Typical scenarios are intrusion or fraud detection and predictive maintenance.

Since in real-world settings the data is often available as an infinite and ever evolving stream, in this thesis we propose Adaptive Variational Autoencoders (AVA) as novel approach for unsupervised outlier detection in data streams.

First, we introduce a streaming framework for training arbitrary generative models on data streams. Generative models are used for sample generation to keep learned knowledge.

Second, we instanciate this framework with a Variational Autoencoder as AVA. To handle data streams in a truely unsupervised manner, AVA 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.