Around 2006 - 2008 I participated in a Research Experiences for Undergraduates program at an AI lab. I used to get into arguments with grad students when I ass

Around 2006 - 2008 I participated in a Research Experiences for Undergraduates program at an AI lab. I used to get into arguments with grad students when I asserted it was intuitively obvious that given a neural network which recognizes two features, it should be possible to extract a trimmed network which recognizes one feature but not the other.