Zach Lipton

I wasn't like resistant.

I think a lot of people who had spent a lot of time becoming like learning a lot of mathematics in order to like reason about like Hilbert spaces because they're working on kernel machines were like, oh, this neural network stuff is pretty, pretty, pretty ad hoc.

You know, it's pretty scruffy.

I think like I saw that wave coming clear eyed because I wasn't deep in any rabbit hole.

and it hit image recognition then it hit speech recognition then it started hitting natural language processing and i think a big moment for me personally was around 2014 2015 recognizing hey the the source of problems that this was these were neural sequence models basically the the the forerunners or the the the ancestor of what became like the transformer is now the dominant sequence architecture at the time was the recurrent neural network that was the dominant workhorse

and i saw that there was there was the kinds of problems we had in language were very similar to the kind of problems we had with medical record data like you can't just it's like an image where you could just crop every image to be the same size and just feed it into the model you sort of have like a long sequence and some documents are one sentence and some are one paragraph and some are two pages and the order in which the words occur kind of matters you've got this like big sequence of events

And I saw the like sequence of events, whether it was like lab readings or medications administered or diagnoses applied to a medical record kind of has a similar structure of the order matters.

So some could be long, some could be short.

There were some additional problems, like you might go long periods of time without an observation.

But for me, I got to basically be like the first paper that applied like modern neural sequence models to electronic health record data.

And I got that to be like

My first paper in PhD.

That was like a huge moment.

And I think I was just very lucky and to sort of point in the right direction, sitting at the intersection of the right two fields at the right time.

And just to be in the field at a moment of great change, you know, I think when the field is classified and.

plateaued at that point, it's really run by the people with the deepest kind of like tribal knowledge.

And the moment when like a new lane opens up, it's actually the scrappiest people who don't know anything and don't have any commitment that get in.

And so I got to get in very early in my PhD and that kind of opened up

a path to sort of have a voice in some kind of like leadership in this community and other opportunities to collaborate.

Because if you were looking around in 2015 saying, hey, I want to do deep learning in healthcare, there were like three people you could talk to.