Ameet Talwalkar, Carnegie Mellon ML professor and Chief Scientist at Datadog, joins Ben Lorica to trace time series foundation models from skepticism to Datadog’s Toto V1 and V2.<br>Subscribe to the Gradient Flow Newsletter 📩 https://gradientflow.substack.com/<br>Subscribe: Apple · Spotify · Overcast · Pocket Casts · AntennaPod · Podcast Addict · Amazon · RSS.<br><br>Detailed show notes and transcript, can be found on The Data Exchange web site.

Why Observability May Be AI’s Next Frontier

The Data Exchange with Ben Lorica

All right, today we have an old friend, Amit Tauwaker, professor in the machine learning department at Carnegie Mellon and also chief scientist at Datadog, which you can find at datadoghq.com. Taglines are full stack observability and security built for enterprise scale. See inside any stack, any app at any scale, anywhere. And with that, Amit, welcome to the podcast. Thanks, Ben. Always great to chat with you.

All right. So the first topic is time series foundation models, which will segue to what you folks have released and announced at Datadog. But just basically, Amit, just to set the baseline for our audience, so time series and foundation models. So foundation models for most people means I have a base model that I can then customize, post-train and use for a bunch of things.

So then the idea here is that people have built such a thing for time series. And I know that I may be misremembering this, but I think you helped run a workshop recently on time series foundation model. Yeah. I mean, so my exposure to time series foundation models goes back a few years. It starts with actually with my academic hat on working with folks at CMU.

I would say, you know, there was a trend a couple of years ago of a bunch of folks in academia or elsewhere, you know, developing times or developing foundation models for X. Time series for, you know, geospatial modeling for, you know, genomics for anything, right? But it kind of made sense that we have foundation models for X. Can we take that kind of same idea of free training on massive amounts of unlabeled data in some sort of self-supervised way?

Why Observability May Be AI’s Next Frontier

Read the full transcript

The candid version is on a podcast,
not the earnings call.

Why Observability May Be AI’s Next Frontier

Read the full transcript

The candid version is on a podcast, not the earnings call.

The candid version is on a podcast,
not the earnings call.