Paper Radio

Two-host explainers of machine learning papers, generated locally.

Continuous Actions: The Reading List

23 Aug 2026 · 18:01

Extending Q-based Variational IRL to continuous action spaces. The two equations that break, and the literature that has already answered each one: max-entropy IRL and the partition function, soft actor-critic as amortized inference, tractable maxima over continuous actions, and the scalable GP machinery underneath.

Attention Is All You Need

23 Aug 2026 · 2:14

Why the 2017 Transformer paper replaced recurrence with attention, what that bought, and the quadratic cost the field has been fighting ever since.