Attention, diffusion maps, and magnetic Laplacians are different regimes of a single Markov geometry from pre-softmax query-scores, linked by a QK bidivergence and Schrödinger bridges into equilibrium, nonequilibrium, and driven dynamics.
Parzen, On estimation of a probability density func- tion and mode, The Annals of Mathematical Statistics 33, 1065 (1962)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
The Diffusion-Attention Connection
Attention, diffusion maps, and magnetic Laplacians are different regimes of a single Markov geometry from pre-softmax query-scores, linked by a QK bidivergence and Schrödinger bridges into equilibrium, nonequilibrium, and driven dynamics.