A problem for Recurrent Neural Networks that occurs when the gradient used to train becomes too large/small as a result of a specific node’s activation being too high/low, causing a ripple effect in the weights during Unrolling.
A problem for Recurrent Neural Networks that occurs when the gradient used to train becomes too large/small as a result of a specific node’s activation being too high/low, causing a ripple effect in the weights during Unrolling.