We present algorithms for learning depth two neural networks where the
hidden nodes are threshold gates with constant fan-in. The transfer
function of the output node might be more general: we have results for
the cases when the threshold function, the logistic function or the
identity function is ... more >>>