We should coordinate our research expeditions to make layered scientific maps.

Nyquist-Shannon-style bandlimited signal recovery is really just linear regression.

Word embedding models learn interpretable topic-level concepts one at a time.

There's only one degree of freedom in choosing how hyperparameters scale with network width.

Using einsums for pen-and-paper matrix differentiation

How to derive order-of-magnitude expressions for radiation formulas

Using statistical physics to understand what makes a good activation function

My senior thesis won an award

The "cavity with a hole" is soo confusing.