1 paper
George Stamatescu, Federica Gerace, Carlo Lucibello +2
The training of stochastic neural network models with binary (±1) weights and activations via continuous surrogate networks is investigated. We derive new surrogates using a no…