V. Attention Seeking
three independent derivations of softmax attention — from thermodynamics, associative memory, and kernel regression — and the variational principle that unites them.
read more →connecting dots between mathematics, machines, and the mind.
three independent derivations of softmax attention — from thermodynamics, associative memory, and kernel regression — and the variational principle that unites them.
read more →uncovering how spectral outliers from low-rank structure rise above chaotic bulk dynamics in recurrent neural networks, and how these isolated eigenvalues can serve as early-warning signals for bifurcations in driven systems.
read more →developing a philosophical framework for emergence that takes higher-level patterns seriously without surrendering reductive grounding — effective field theory as a model for the mind.
read more →exploring theoretical justifications for rapid mixing in high-dimensional Markov chains.
read more →tracing the shared formalism between classical random walks and quantum evolution on graphs, revealing how diffusion and interference arise from a common linear backbone.
read more →