An interesting proof of concept, but separately and precisely controlling the temperature of each 'neuron' to adjust it's activation seems like a nonstarter. Th

An interesting proof of concept, but separately and precisely controlling the temperature of each 'neuron' to adjust it's activation seems like a nonstarter. The behavior of the NN will change unpredictability if it heats up even a little. [thinks a bit] It might be practical to get it working identically at a wide range of temperatures for a binary neural network architecture if you incorporate 'normal' RLUs to round the 'neuron' output to 1s and 0s.