Full steam ahead! Unraveling Forward-Forward Neural Networks (Ep. 232)
Data Science at Home
00:00
The Forward Forward Algorithm
Geoffrey Inton has developed a new way of thinking about computer science. The forward forward algorithm works by replacing the traditional two-step learning process with just one pass. In this first pass, the network adjusts the weights in order to maximize the goodness in every hidden layer while the negative pass operates on negative data. By doing this, you know, some experiments have shown that the test error is some kind of minimized - but back propagation based neural networks don't perform as well.
Transcript
Play full episode
