Why graphs break message passing
Over-smoothing, over-squashing, and the limits of how far information can travel across a graph.
Notes & essays
Writing.
Thinking out loud about the mathematics of learning — and trying to explain the hard parts simply.
Coming soon
The first post is brewing.
I'm planning to write about the ideas I find most beautiful in AI and mathematics — clearly, and from first principles. Here's a taste of what's coming.
L1 vs L2, intuitively
What the choice of norm really does to a regression — robustness, sparsity, and sensitivity to outliers.
Brains as inductive bias
What neuroscience-inspired architectures might teach us about building more human-like models.
Surviving distribution shift
Notes from benchmarking models when the test data refuses to look like the training data.