Wulff, BenjaminBenjaminWulffSchücker, JannisJannisSchückerBauckhage, ChristianChristianBauckhage2022-03-142022-03-142018https://publica.fraunhofer.de/handle/publica/40287410.1007/978-3-030-01424-7_55Concerned with neural learning without backpropagation, we investigate variants of the simultaneous perturbation stochastic approximation (SPSA) algorithm. Experimental results suggest that these allow for the successful training of deep feed-forward neural networks using forward passes only. In particular, we find that SPSA-based algorithms which update network parameters in a layer-wise manner are superior to variants which update all weights simultaneously.en005006629SPSA for layer-wise training of deep networksconference paper