This hits home for edge ML: SOTA architectures assume abundant resources. On the edge, we need new ML systems designed for constraints, not scaled-down versions of unconstrained ones.

Connections
Puzzle #670
🟨🟨🟨🟨
🟦🟦🟦🟦
🟩🟩🟩🟩
🟪🟪🟪🟪

I refreshed it couple of times, thinking there was an error 😆

Connections
Puzzle #656
🟦🟦🟨🟦
🟩🟦🟦🟦
🟦🟦🟦🟦
🟨🟩🟨🟨
🟨🟨🟨🟨
🟪🟪🟪🟪
🟩🟩🟩🟩

😂