Schmidhuber Shares 1991 Munich Lab Scene With Early Neural Net Diagrams

Professor @maier_ak has written a thoughtful substack article on the roots of today’s trillion-dollar AI boom in 1991 when Munich was the epicentre of AI: https://akmaier.substack.com/p/five-months-in-munich-revisiting

Professor @maier_ak has written a thoughtful substack article on the origins of today’s AI boom in the early 1990s in Munich: https://akmaier.substack.com/p/five-months-in-munich-revisiting

@SchmidhuberAI @hardmaru The honest reading honours both halves: the 1991 originals AND the engineering that scaled them. Vaswani et al., He et al., Goodfellow et al., DeepSeek — turning ideas into working systems is itself original research, not a footnote.

@SchmidhuberAI and @hardmaru just published a calendar of contributions from the March-August 1991 TUM group that maps cleanly onto today's AI stack.

@SchmidhuberAI @hardmaru FKI-147-91 → Fast Weight Programmers (linear Transformer, 26 years before Vaswani). FKI-148-91 → unsupervised pre-training + distillation. Hochreiter's 15 June thesis → residual learning. 31 August 1991 → first GAN-style adversarial paper.

@SchmidhuberAI @hardmaru Read the full piece: https://akmaier.substack.com/p/five-months-in-munich-revisiting

@maier_ak This great work! That you for explaining the problem!