Sepp Hochreiter

Sepp Hochreiter

A German computer scientist whose 1991 diploma thesis identified the vanishing gradient problem: in a deep or recurrent network, the error signal shrinks exponentially as it is propagated backwards, so early layers and distant time steps barely learn at all. With Jürgen Schmidhuber he answered it in 1997 with Long Short-Term Memory, a recurrent cell with a gated memory that lets information persist across long sequences. LSTMs powered speech recognition, translation and text generation until Transformers displaced them. He now leads work on xLSTM at JKU Linz. (See also: Jürgen Schmidhuber, Backpropagation, Neural network, Transformer)