Lecture: Are People Still Smarter than Machines?
How can a neural network learn to do something cognitively interesting? The Biologically Inspired Approach If neuron A participates in firing neuron B, strengthen the connection from A to B The Optimization Approach Adjust each connection to minimize the network’s deviation from a desired output Backpropogation algorithm became the basis of sbsequent research in the PDP framework, as well as almost all of Deep Learning Their Early Success of the Approach Models that could learn to read words and generalize to pronounceable non-words, capturing human-like response patterns: they refect statistical patterns and similarity relationships Models that captured many aspects of human semantic cognition and the disintegration of human semantic abilities resulting from neurodegenerative disease Models that showed in principle how aspects of language knowledge could be captured in a simple artificial neural network Nowadays AlexNet, ResNet and so on.