1:30 pm MCP 201
Building Quantum Field Theories Out of Neurons.
In this seminar I'll study an approach to field theory in which fields inherit their intrinsic randomness from N constituent random neurons, rather than from a function space density. A theory is defined by how the neurons are stitched together, which is called an architecture in the neural network literature. I will demonstrate a simple neuron layer that yields Euclidean-invariant neuron correlators, which is relevant for constructing Euclidean-invariant field theories, as well as Lorentz-invariant ones when reflection positivity is satisfied. A simple choice yields a tunable power spectrum. Architectures are presented that yield dual theories at infinite N, but have different symmetries at finite N. Near-Gaussianity is exhibited at large N, potentially explaining a feature of field theories in Nature.