Network research has access to more data than ever, yet many fundamental challenges remain unsolved. This keynote takes a critical look at why progress in IP flow-based research lags behind other ML domains. We will dissect quality issues in popular benchmarks, expose the gap between research evaluations using complete historical flows and operational needs for real-time analysis, and discuss the labeling problem and potential solutions. Drawing comparisons with breakthroughs in NLP and vision, we will explore what our field might be missing and share practical insights from CUJO AI's approach to production data collection.
Adrian Pekár
Dr. Adrián Pekár is a Senior Data Scientist at CUJO AI, where he develops ML-powered security solutions that protect millions of home networks worldwide. His work focuses on evolving threat detection while maintaining industry-leading protection standards for connected devices.
With over a decade of expertise in network traffic flow measurement and categorization, Adrián bridges the gap between advanced research and real-world application. His academic journey—from his PhD in computer networks at the Technical University of Košice, Slovakia (2014) through postdoctoral research at Victoria University of Wellington, New Zealand, and his most recent role as Associate Professor at the Budapest University of Technology and Economics, Hungary—built deep insights into network behavior that now power practical security innovations.
Leveraging his unique perspective on how data flows across networks, Adrián develops intelligent methods that detect anomalies and threats in real-time across billions of connected devices. His work demonstrates how advanced traffic analytics and machine learning are essential for protecting the expanding IoT ecosystem as new technologies and encryption methods continuously reshape the threat landscape.