RSS 2.0 feedQuick Summary
- This is a PhD-level course focusing on the latest advancements in Network Management.
- Focus on research: read, review and present papers, final mini research project
- Graduate students or advanced undergrads (with consent of instructor)
- Not a traditional networking course (not CSCI1680)
Overview
The central motivation of data-driven networking is that existing heuristics and protocols are special purpose and are based on human insights. With the constantly growing expectations and growing diversity in networks guess to design of new devices (e.g., 5G networks), due to growing internet usage (e.g., IoT or self driving cars), and due to expanding internet ecosystem (e.g, internet drones in Africa or 2G networks in S.E. Asia), there is a need for new algorithms that learn to automatically adapt to the growing diversity in requirements and expectations. Moreover, to support these more dynamic learning-based algorithms, there is a need for more flexible and adaptable system designs and hardware that allows for quick and efficient reconfiguration to enforces the dynamic policies generated by the learning-based algorithms.
The course will begin with a series of lectures on domain problems (e.g., IoT challenges, cellular challenges, or on-demand video challenges), a survey of ML technologies (i.e., bayesian optimization, multi-armed bandits, reinforcement learning), a discussion on emerging technologies that is enabling fine-grained reconfiguration in data centers, cellular networks, and web servers (i.e., eBPF, NFV, SDN, P4). After these background classes, the course will transition into a paper reading seminar class.
- Lecture time: Tu 4-6:20
- Location: CIT 477 (for now)
