Panel on machine learning in networking #infocom18 
First, Mohammad talks about the tradeoff between universality and performance. ML is about trying to achieve both
Second, Xin talks about increasing network complexity (multiple players, high dimensionality) presenting opportunities for data-driven networking. Among supervised/unsupervised/reinforcement learning, her preferred method is reinforcement learning.
Aaron sees opportunities to use machine learning for providing management/debugging insights to network operators
Hui sees opportunities to replace heuristics in video, TCP, fraud with ML. She also raises interesting challenges in applying ML: accuracy, interpretability, *privacy* and *bias*
Zhi-Li also talks about both opportunities (leveraging big network data) and challenges (interpretability)
Thyaga warns against using ML for the sake of using ML, especially if you can’t explain why you got better performance results
