In primary visual cortex, Hubel and Wiesel discovered by chance that the neurons responded well to luminance ``edges''. In higher visual areas and other sensory areas, it has been more difficult to discover the properties that makes the neurons fire. One recently popular technique is, instead of trying to span the space of possible stimuli, to take simple random stimuli and extract what is common from all the stimuli that cause a spike. The basic version of this technique is limited by being linear; it depends on the responses to stimuli throughout the receptive field being additive which is not true for higher order sensory neurons. Hsin-Hao Yu (my graduate student) and I are investigating the use of non-linear pattern finding algorithms for automatic discovery of non-linear receptive fields. This work has a definite synergistic flavor. We are using machine learning algorithms to elucidate cortical physiology, which will in turn lead to better machine learning pattern recognition algorithms.