Hinton did introduce the Boltzmann machine networks that kind of resemble thermodynamics if you squint really hard. And those didn't really pan out. Of course

Hinton did introduce the Boltzmann machine networks that kind of resemble thermodynamics if you squint really hard. And those didn't really pan out. Of course the price is really for MLP backpropagation (or how it found applications) but I guess it's not physics enough. Hopfield networks never really found use either, but they are sort of related to Ising models and NNs, so I guess it's physics then.