Daniel Situnayake

I think 2025 will be the year of foundation models at the edge. Get ready by joining us next week for a panel discussion! It's at 18:00 CET, 09:00 PST on Wednesday 15th 🙌🏼

www.wevolver.com/article/disc...

I'm looking forward to being a keynote speaker at the Generative AI for Edge Computing Symposium, at the AAAI 2025 Spring Symposium from March 31 to April 2nd!

The Call for Papers is open and you can find the details here, please consider submitting: sites.google.com/view/gen-ai-...

Not so long ago I was driving around LA with MapQuest printouts. Today you can ride in a self-driving Waymo. Wild.

stuck a jax laptop sticker up in a nightclub bathroom friday night. hoping some spaced out dev has a 3am WTF moment...

Basement JAX

And peeping from behind the door is another black box

We'll be following soon with more serious industrial datasets, all hosted on the same portal and licensed permissively. Thank you to Louis Moreau and @doncarlos999.bsky.social who are leading our data gathering efforts!

Explore Edge AI With Our Public Datasets Whether you're working on object detection, audio classification, or visual anomaly detection, these datasets will help you grasp the data requirements and challenges you'll face at the edge.

We're attempting to address this problem at @edgeimpulsecom.bsky.social by releasing our own free public datasets, permissively licensed. We just launched with four simple datasets designed for learning and experimentation.

www.edgeimpulse.com/blog/explore...

It's super difficult to find datasets representative of industrial edge ML problems. This holds back the entire field: if we're benchmarking on eclectic web-scraped nonsense, we optimize for that—not for real world tasks.

Integration Edge AI has come a long way, but sensory perception is not enough. What will it take to build systems that can act?

Edge AI is maturing, and we’re getting pretty good at sensory perception. What comes next? In this post I explore what we’ll need to bring real intelligence to the field.

I’d love to hear what you think 👇🏽

open.substack.com/pub/dansitu/...

Histogram peaked at 3 minutes and 2 weeks since sent

When I will respond to your email

I love these sorts of design and engineering challenges. Deciding that all insights must be actionable changes the way you think about presenting data! And the best path is not always obvious.

I'd love to hear from anyone who has solved these types of design challenges in their work.

Instead of printing raw metrics, we need to help the user interpret the data and take action if necessary: for example, perhaps the results indicate that the Bayesian model is underfit, and the user needs to continue the search a while longer to be confident of getting the best results.

For example: We're working upgrading our hyperparameter tuning system to use Bayesian search, and there are tons of cool metrics you can log about the quality of the Bayesian model.

But to the average user, these stats are meaningless at best—and potentially misleading or confusing.

I've learned that a major challenge in making ML tools usable by all developers is ensuring everything is actionable. There's no point surfacing raw metrics if the user doesn't know how to interpret or improve them.

Some people want to explore, others prefer to stay home. We need both types, but I’m glad to be the former!

It's pretty amazing how much this feels like the old Twitter—or at least my memories of it. But now it's closer to actually being a piece of open web infrastructure.

🕳️

The term stopped making sense when it was picked up by people who have never lived in the US and don’t know what a “bro” is

Thank you Patrick!!

Check out dans substack if you’re looking for insightful reads on the state of ml seasoned with a pinch of edge ai.

🤞🏽I’ll be satisfied with a finish!

Good luck with the races!! I also had big plans for running this year and was stymied by major life events. But I’m running a 50 miler on Saturday with probably insufficient training, hopefully I survive 😅

Another great person to add is @matpalm.com

starter pack cover

Another starter pack with Machine Learning and Artificial Intelligence Researchers. go.bsky.app/gPKeHz We can learn a lot from people like: @abeba.bsky.social @eckles.bsky.social @maosbot.bsky.social @andrew.heiss.phd @jessicahullman.bsky.social @melaniemitchell.bsky.social @wattenberg.bsky.social

Thank you, I will try my best! I’m excited to be back on what feels like a much healthier platform.

Great list, thank you! Would you be open to adding me?

Integration Edge AI has come a long way, but sensory perception is not enough. What will it take to build systems that can act?

Edge AI is maturing, and we’re getting pretty good at sensory perception. What comes next? In this post I explore what we’ll need to bring real intelligence to the field.

I’d love to hear what you think 👇🏽

open.substack.com/pub/dansitu/...