AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Using Multiplier Interactions in RNN Modeling
The way that we, the physical mechanism that we use, it's called a multiplicative gate. All that means is that we take the outputs of these two hippos and we multiply them. So now we can bring out Obama as the, as the next word. And then the second hippo that is remembering words through the whole sequence,. Then you get to Michelle, and then you can say, oh, there's a president that appeared some time ago. I think this gating idea actually goes back a long time to even in RNN things like long term, long short term memory LSTM.