DNr: Berzelius-2024-228
Project Type: LiU Berzelius
Principal Investigator: Anindya Sundar Das <anindya.sundar.das@umu.se>
Affiliation: Umeå universitet
Duration: 2024-06-01 – 2024-12-01
Classification: 10201


Anomaly analysis is of great interest to diverse fields, including data mining and machine learning, and plays a critical role in a wide range of applications e.g., health, finance, and intrusion detection. Recent technological advancement such as rapid emergence of edge devices, wide application of the internet and its sources, social media have engendered lots of complex and high-dimensional data (i.e., images, text, video, audio, multimodal etc.). However high-dimensionality in the data poses additional challenges in anomaly detection, as the distance between the data objects decreases with increased dimensionality of data. Moreover, there are additional challenges to make the algorithms interpretable and robust against contamination. This project aims to research the learning models for anomaly analysis on complex data and propose trustworthy algorithms for real-world applications. The objective is three-fold 1) Robust anomaly detection, 2) Explainable anomaly detection and localization, 3) Anomaly detection in limited data regime. In objective one, we aim to study various attacking strategies and defenses against attacks and design robust models for anomaly detection. The second objective is to make model explainable by identifying the markers of anomalous instances. Here, the goal is to localize the features that explains why the examples are detected as anomalies. The third objective is to focus on the problem when data are limited and incomplete, having high missing values. The objective is also to optimize the training strategies in limited data settings where the training data is small, unlabeled or has very limited labeled unbalanced data. The project also aims to give practical and generic Anomaly analysis-based solutions for different modalities such as, rumor detection, defect identification, fake news detection etc.