Sepp Hochreiter is an Austrian computer scientist and professor who heads the Institute for Machine Learning at Johannes Kepler University Linz. His research covers deep learning, reinforcement learning, bioinformatics, and the foundations of neural networks. He is one of the central figures in the European deep-learning community.
Hochreiter is best known as the co-inventor of Long Short-Term Memory (LSTM), the recurrent neural-network architecture he developed with his doctoral advisor Juergen Schmidhuber, published in 1997. As a student he had identified the “vanishing gradient” problem, the reason ordinary recurrent networks fail to learn long-range dependencies in sequences. LSTM solved it with a gated memory cell that can hold information over many time steps. For roughly two decades LSTM was the workhorse behind speech recognition, machine translation, and other sequence tasks, before the transformer architecture took over much of that ground.
For the library’s reader, Hochreiter is the name behind the LSTM entries and a counterweight to the usual framing that deep learning began around 2012. His 1997 work is one of the key technical reasons neural networks could eventually handle language and time-series data at all. He remains a vocal advocate for the continued relevance of recurrent and memory-based approaches alongside the dominant transformer paradigm.