Sepp Hochreiter

Sepp Hochreiter is an Austrian computer scientist and one of the principal architects of modern deep learning. His 1991 diploma thesis, supervised by Jürgen Schmidhuber, identified and formally analyzed the vanishing gradient problem in recurrent neural networks — a discovery that rendered plain RNNs theoretically incapable of learning long-range temporal dependencies. This work laid the foundation for the Long Short-Term Memory (LSTM) architecture, which Hochreiter and Schmidhuber introduced in 1997 and which became the dominant approach to sequence modeling for two decades.

Hochreiter's research has consistently pursued the question of how neural networks can learn to represent and manipulate structured information over time. His later work on deep learning theory, bioinformatics, and drug discovery extends the same principle: that the key to intelligent systems is not merely scale but the right representational geometry. The LSTM was the first demonstration that learned gating could replace hand-designed memory structures — a principle now ubiquitous in machine learning.