⏳Jürgen Schmidhuber
Original nameJürgen Schmidhuber
German Computer Scientist, Co-Inventor of the Long Short-Term Memory (LSTM) Network
Those Who Conceived That "a Machine Could Think" · Pioneers
Long Short-Term Memory (LSTM)Early Research on Recurrent Neural NetworksPublic Disputes Over Deep-Learning Priority
Who they are
Jürgen Schmidhuber (1963– ) is a German computer scientist who, with Sepp Hochreiter, in 1997 co-proposed "Long Short-Term Memory" (LSTM), an architecture that, by introducing a gating mechanism, solved the vanishing-gradient problem traditional recurrent neural networks tended to face in handling long-sequence data, thereafter widely applied for over two decades in sequence-modeling tasks such as speech recognition and machine translation, gradually yielding its mainstream place only after the rise of the Transformer architecture. Schmidhuber is one of the most active and most controversial public voices in deep learning, long publicly arguing that several core deep-learning contributions widely credited to Hinton, LeCun, Bengio, and others were in fact proposed earlier in original idea or similar method by himself and his team yet did not gain due scholarly recognition; in this series of public disputes both sides hold to their views, and scholarship has still no broadly acknowledged conclusion, and this site states only the positions the two sides have publicly expressed, without adjudicating.
Primary sourcesHochreiter & Schmidhuber, "Long Short-Term Memory" (1997)
Key stories
A Priority Case Still Without Conclusion
Schmidhuber has many times, at academic conferences and public talks, accused by name several core deep-learning ideas widely credited to Hinton and others of having "long been proposed by me and my students," including the generative adversarial network and part of the core mechanism in the Transformer architecture, holding that the mainstream scholarly and media narrative has systematically overlooked his contribution. Hinton and others and some scholarly peers reply that Schmidhuber’s statements often exaggerate the direct connection between his original work and the later breakthrough results. This years-long public dispute is still without broadly acknowledged conclusion, and this site states the two sides’ respective positions, adjudicating for neither who is the "real inventor."
Version differences
Schmidhuber has repeatedly and publicly charged that several core contributions in the deep-learning field credited to others should in fact be credited to himself and his team’s earlier research, and in this series of priority disputes each side holds its own account, academia having no widely acknowledged conclusion to this day; this site states both sides’ positions without adjudicating.
Echoes today
Below are how modern works borrow or reinterpret this name or story — not the original material. The two differ, so keep them apart.
The public image of an "underestimated pioneer"The "grievance" Schmidhuber has long publicly voiced thereafter gradually turned into a distinctive phenomenon often cited when tech media report the history of deep learning — the image of a scientist of acknowledged outstanding contribution who continually feels himself underestimated by history.