Simple Artificial Neural Network
conv, pool, fc, input, regression, softmax, svm, lstm, mdpagent, recurrent
type, act, noise, drop, l1_decay_mul, l2_decay_mul, out, hidden, sx, sy, sz, depth, filters, stride, padx, pady, padz, gamma, eps, nw, nsd
adadelta, adagrad, adam, nesterov, sgd, windowgrad
type, lr, l1_decay, l2_decay, momentum, ro, eps, beta1, beta2, max_grad, batch, manual, bptt_step