An algorithm for in-the-loop training based on activation function derivative approximation | IEEE Conference Publication | IEEE Xplore