Mining multimodal records: explainable patterns and anomalies discovery

Abstract

The discovery of anomalies and patterns is essential to many applications and areas, such as COVID-19 detection in chest X-Rays and vertebral fracture detection in MRI for health applications, as well as credit analysis in finance, and bot detection in social networks. The mining approach is application-dependent, and the discovered patterns should be shown to specialists using appropriate tools for boosting explainability and understanding, usually by taking advantage of visualization metaphors. In this internship proposal, we aim at exploring multimodal data from multiple application scenarios and types (e.g., images, graphs, electronic health records, and financial transactions) by proposing modular, scalable, and explainable methods. The internship will be carried out at Carnegie Mellon University (CMU), USA. The objectives meet the interests of both the FAPESP thematic project "Mining, Indexing and Visualizing Big Data in Clinical Decision Support Systems - (MIVisBD)" and the current research carried out at CMU by the supervisor and co-supervisor.