In the recent years, we deep dived into the research of AI Journalism. Together with the Faculty of Social Sciences at the Charles University, we obtained funding from the Technology Agency of the Czech Republic (TAČR) that allows us to explore the exciting field of automated journalism and fact-checking. Our long-term goal is to develop tools for journalists that will allow them to get the most out of digital technologies and allow newsrooms to embrace automated journalism. Research into the methods of verification and authorization of AI-generated content is our top priority right now.
Security Applications of ML
The group led by Václav Šmídl and Tomáš Pevný is interested in development of AI methods for non-standard tasks that are relevant in industrial applications. In collaboration with the Avast company, we look into non-standard data and non-standard objectives. Data in this domain are commonly available in the form of tree structures. Substantial effort is thus dedicated to learning algorithms that process tree (or other types of graph) data. Since security application is very sensitive to false-positive errors, we develop methods minimizing those. On theoretical level, we are interested in interpretable models, which we approach by learning sparse representations. For example, we design neural units that have interpretation as algebraic mathematical equations.
A research cluster around Jan Drchal develops methods similar to those used in Natural Language Processing, but instead of human spoken languages, our focus is on log messages. Our main interest lies in transforming them into forms appropriate for downstream ML tasks, i.e., log embeddings.
The results are highly applicable in many areas. We can help administrators understand the logs better by smart visualizations of the messages based on their clustering. We work on methods of log anomaly detection with Red Hat. We also develop log-based systems for the prediction of manufacturing process durations for Škoda Auto.