Andrew Ilyas

This other kind of fairness, we also want to impose this other constraint on whatever outcome we're going to get and things like that.

In some sense, like our saving grace in theory land is that what we care about is recovery of parameters.

There is a ground truth that we're after, and that's what allows us to actually make progress there.

But I think when you go to machine learning, there's no longer this like ground truth parameter that you're searching after, and it makes everything much harder and much murkier.

So I will be starting a professorship, not this year, but the next at CMU.

So hopefully have a lab website up by the time, I don't know, it won't happen soon, but hopefully by the time I start, there'll be a lab website.

You can reach me by email if you're ever curious.

It's just my first letter A and my last name, I-L-Y-A-S at MIT.edu.

And my website is just myfullname.com.

I think I'd like to continue working along this ML pipeline direction that I was talking about.

I'm really flexible in terms of where exactly in the pipeline people want to work, but I think things that advance our understanding of what are the key drivers and things that make machine learning models behave the way they do in production is really interesting.

I think this is a tricky one, I think, with starting out as a professor.

I've been advised that maybe you want to do that during a postdoc and then not do it so much during your first couple of years.

I still haven't really decided what I'd like to do.

I definitely want to work on problems with practical impact, whether that's in the form of industry collaborations or I think something else I'd be really excited about is collaborating with domain experts in, for example, sciences or in robotics or in

sort of these fields that I have less exposure to.

It's been great.

Thank you so much.