Wireline EE here. ML in the context of wireline communications has been widely studied (grant money works that way). However, at least in wireline communicatio

Wireline EE here. ML in the context of wireline communications has been widely studied (grant money works that way). However, at least in wireline communications there isn’t a lot of gain to be had. This is because we have good physical models of underlying impairments and our ASICS are optimized using those physical models, so ML isn’t able to improve upon that much. Where it can help is where we don’t have good data on underlying physical parameters of the channel. So in this case, you can get a bit more capacity than you would otherwise. So for wireless where you have signal fading and huge variation in multi path interference, maybe there would be benefit as these phenomena are hard to model. Also, you may not realize but we’ve been using ML techniques for decades in communications. Gradient descent is used to optimize equalizers; maximum likelihood estimation(and equalizer optimization) used for phase estimation in high order QAM. Plenty of other examples. So you probably are already familiar with much of the basic tool kit. I had a wannabe startup founder in ML tell me that there’s no way I could possibly understand the stuff if I didn’t have PhD in that area in CS (I am physics). I just smiled and nodded.