Šimon Mandlík, a PhD candidate specializing in machine learning for cybersecurity at the Czech Technical University, dives into the intriguing world of fraud detection using graph-based techniques. He explains how graphs can unveil malicious activities by analyzing relationships within vast datasets. The discussion highlights the advantages of his hierarchical multi-instance learning method over traditional approaches, tackling challenges like scalability and heterogeneous graphs. Mandlík emphasizes the 'locality assumption' in fraud detection, resulting in faster and more accurate outcomes.