Must-read papers by Jürgen Schmidhuber

Author: Ziming Liu (刘子鸣)

LSTM: A search space odyssey (Link)

Reason: walk you through the development of LSTMs

The binding problem

1. On the Binding Problem in Artificial Neural Networks (Link)
2. Neural Expectation Maximization (Link)

Reason: Today’s foundation models suffer a lot from the binding problem.

Continual learning

1. Compete to Compute (Link)
2. The Two-Dimensional Organization of Behavior (Link)
3. Adaptive decomposition of time (Link)

Reason: a hot topic right now – many ideas (e.g., multiple time scales) can be attributed to Schmidhuber’s work.

Algorithmic complexity

1. Low-Complexity Art (Link)
2. Facial Beauty and Fractal Geometry (Link)

Reason: I believe that compression (algorithmic simplicity) is key to intelligence

Citation

If you find this article useful, please cite it as:

BibTeX:

@article{liu2026Jurgen-Schmidhuber,
  title={Must-read papers of Jürgen Schmidhuber},
  author={Liu, Ziming},
  year={2026},
  month={April},
  url={https://KindXiaoming.github.io/blog/2026/Jurgen-Schmidhuber/}
}