Very nice!

@jay.bsky.team @support.bsky.team Can something please be done about these?

Congratulations! 🙂

These typically work when the graph structure is assumed, and parameters are learned. Eg for question — robot is being trained to cook. Grasping, placing, cutting, sprinkling, mixing, measuring are all tasks that apply across multiple ingredients, mixtures, or dishes. Data is 3D voxel-labeled video.

Thank you!

Trying my luck here: compositionality in learning. Automated decomposition of sequential tasks into modules that can be composed.

Structured analysis of hidden convexity. What are problems where it is possible, how to sniff them out, good reviews, etc.

Yep, I’ll go on a limb and say impossible to imagine.

Yep! Same reason applies for composition.

One might argue that it is closed form if one can do analysis and algebra on it. Composition, derivatives, etc.

Black hole man of India. His book was delightful.

Whoa! 🙂

What’s the backstory?

Congratulations, Yoshua!