Multimodal Anomaly Detection

Program Description

Anomaly detection is highly important in many machine learning-driven fields, yet developing mostly independently for different data modalities such as audio, EEG signals, images, text, videos or time-series in general. A lack of knowledge transfer between different modality-specific communities is a wasted opportunity, especially because multiple complementary data modalities can be combined into a single system to get a more complete view of monitored objects or scenes.

The purpose of this program is to bring together experts in anomaly detection from different communities and to promote joint research activities of its members. There is already strong evidence for research excellence of program members related to anomaly detection in industrial or medical applications. Examples are acoustic machine condition monitoring, video surveillance and detection of industrial defects, medical imaging, agricultural monitoring, visual insect monitoring, monitoring of wind turbines, skin lesion classification with Raman spectra or theoretical work.