Welcome to the FlowFrontiers research group, a dedicated team within the MEDIANETS laboratory of the Department of Networked Systems and Services at the Budapest University of Technology and Economics. Our group's research focus spans various domains of networked systems, including Federated Learning, P4 programmable networks, flow-level measurement, network anomaly detection, and machine learning applications in intelligent transportation systems. We are committed to pushing the boundaries of understanding in these domains, emphasizing transparency, reproducibility, and innovative thinking.
FlowFrontiers spearheads numerous initiatives that have significant implications for the evolution of networking technology. These initiatives not only reflect our ongoing efforts to push the boundaries of network research, but also symbolize our commitment to practical, effective solutions that address real-world challenges. The following outlines our key research initiatives:
-
ChatGPT in Network Management: We are engaged in the novel application of Transformer Models, like GPT (Generative Pretrained Transformer), which forms the basis of models like ChatGPT, for network traffic management and engineering. These advanced language understanding models open up new horizons for interpreting network behavior and predicting future trends. Our interests in this area range from troubleshooting to automatic configuration and future network planning, capitalizing on the capacity of these models to automate network configurations, swiftly identify potential issues, and plan strategically for future network developments. Our ultimate aim is to create a more intelligent, adaptable, and forward-looking network infrastructure.
-
Federated Learning: Our exploration in Federated Learning (FL) centers on multi-class traffic flow classification and anomaly detection. Amid privacy concerns and the rise of decentralized model training, we strive to further uncover FL's potential in varied scenarios and improve upon the existing models for enhanced precision and robustness. Moreover, we harness the power of FL in the realm of cybersecurity, focusing on detecting and predicting network attacks. Our goal is not only to foster a more secure networking environment but also to advance the state of the art in federated anomaly detection techniques.
-
Programmable Networks and Network Function Virtualization (NFV): In the realm of P4 programmable networks and NFV, we probe into strategies to circumvent their inherent limitations. Our focus lies on optimizing data offloading to improve packet processing delay and overall system performance. Our aim is to harness the capabilities of these technologies to the fullest in advancing network operations.
-
Flow-Level Measurement and Adaptive Flow Aggregation: We specialize in flow-level measurement and adaptive flow aggregation to address the challenges posed by increasing network traffic volumes. Our goal is to understand and mitigate the impact of adaptive flow aggregation on flow-level information distortion, providing a scalable and efficient solution for future network traffic measurement.
-
Predictive Maintenance Using Telemetry Data: We are deeply invested in the development and application of predictive maintenance strategies that leverage telemetry data. Our focus is on using these hardware sensor logs to anticipate hardware failures and take preventive action. The goal is to minimize unplanned outages and service disruptions, enhancing the reliability and uptime of both network and computing infrastructure. By integrating advanced predictive analytics with real-time sensor data, we aim to shift from reactive to proactive maintenance practices, thereby reducing costs and improving the overall quality of service.
-
Real-Time Anomaly Detection: We work extensively on real-time anomaly detection in network flow. Our research is dedicated to understanding the impacts of timeliness on model performance and devising effective strategies to optimize anomaly detection, thereby improving network security and functionality.
-
Machine Learning in Intelligent Transportation Systems: We are pioneering the application of machine learning techniques to intelligent transportation systems. By employing long short-term memory for short-term public bus travel speed prediction, we aim to detect congestion effectively and provide cost-effective, scalable solutions for better transportation systems.
-
Anomaly Detection in Industrial Time Series: We compare heuristic approaches against sophisticated machine learning techniques for anomaly detection in industrial time series. Our research explores the necessity and impact of simple versus complex methods, providing a new perspective on anomaly detection strategies.
Are you driven by curiosity, committed to research, and eager to dive into the intricacies of networked systems? If so, we invite students like you to join our team. You will have the opportunity to work alongside our researchers on pioneering projects, contributing to ground-breaking research while honing your own skills and knowledge.
Whether your interest lies in federated learning, network anomaly detection, programmable networks, or intelligent transportation systems, the FlowFrontiers group offers a diverse array of opportunities. Together, we can contribute valuable insights to our field, drive technological evolution, and shape the future of networked systems and services.
Join us, and let's explore the frontiers of network research together!