Schickard Leibniz Gauss Kurt Goedel Ray Solomonoff at IDSIA Konrad Zuse Turing Archimedes, greatest scientist ever?
Very Deep Learning since 1991
Brainstorm Open Source Software for Neural Networks
Recipient of the 2016 IEEE CIS Neural Networks Pioneer Award (announced in 2015) for pioneering contributions to deep learning and neural networks.
Handwriting Recognition with Fast Deep Neural Nets & LSTM Recurrent Nets (Juergen Schmidhuber)
Switzerland - best country in the world? Leading the world in science, Nobel Prizes, patents, publications, citations, quality of life, competitiveness, happiness, many sports
2011: First Superhuman Visual Pattern Recognition
The formal theory of creativity by Juergen Schmidhuber explains the desire to learn motor skills, to do science, to produce art
Learning Robots
Resilient machine with Continuous Self-Modeling
Statistical Robotics
Best robot car so far (Dickmanns, 1995)
Unsupervised Learning
Attentive vision
STIFF - EU research project on enhancing biomorphic agility of robot arms and hands through variable stiffness & elasticity
The Swiss AI Lab IDSIA
What's new in Juergen Schmidhuber's page?
Master's Degree in Informatics with a Major in Intelligent Systems -  a master's in computer science, with a specialization in Artificial Intelligence
Home Page of Juergen Schmidhuber
Juergen Schmidhuber at Singularity Summit 2009 - Compression Progress: The Algorithmic Principle Behind Curiosity and Creativity and Art and Science
The EU - a new kind of empire?
Juergen Schmidhuber and his iCub baby robot
Juergen Schmidhuber at TU Munich Computer Science
robot population explosion
Deep Learning
AMA (ask me anything)
Lugano, Switzerland
Click at squares above for related links

Update of Oct 7: Most positions are filled, but one postdoc position is still open - closes Oct 31!

Are you interested in continual reinforcement learning (RL) in realistic environments where traditional RL (for board games etc) does not work well, where we need some sort of search for programs running on general purpose computers such as recurrent neural networks (RNNs), and where a good RL machine should learn to plan and reason in hierarchical and other abstract ways?

Join the Deep Learning team (since 1991) that won more competitions than any other. We are seeking PhD students and postdocs for Jürgen Schmidhuber's research group at the Swiss AI Lab IDSIA, for the RNNAIssance Project based on arXiv:1511.09249 on "learning to think." It is about general purpose artificial intelligence for agents living in partially observable environments, controlled by reinforcement learning RNNs, supported by unsupervised predictive RNN world models. Our first papers on such RL control systems based on two interacting RNNs date back to 1990 and 1991, but now, a quarter century later, finally we know how to do it right. (You will also collaborate with other lab members on other projects - we are one big family!)

Duration: initially for 2 years, with possibility of prolongation.

We are looking for outstanding candidates with experience / interest in topics such as deep learning neural networks, RNNs and Long Short-Term Memory (LSTM), evolutionary computation, RNN evolution, compressed network search, reinforcement learning & policy gradients for partially observable environments, unsupervised learning, curiosity-driven learning and the Formal Theory of Fun & Creativity, adaptive robotics, hierarchical reinforcement learning, general artificial intelligence, universal learning machines. The general goal is to advance the state of the art in machine learning and AI. Preferred start: As soon as possible.

INSTRUCTIONS: Submit your CV, a brief statement of why you'd like to work with us on such topics, and a list of 3 references and their email addresses to In the subject header, mention your full name followed by the keyword rnnai2016. (Add "PhD" in case of PhD students. For example, if your name is Jo Mo, and you are applying as a PhD student, use subject: Jo Mo rnnai2016 PhD)

As a PhD student, you must also submit your material to SUPSI through THIS ONLINE FORM (now closed). As a postdoc, you must also submit your material to SUPSI through THIS ONLINE FORM (open again until Oct 31).

PhD student salary: Initially CHF 46'000 per year (>US$ 46k as of 21 Oct 2016). Standard postdoc salary: CHF 80'000 per year (>US$ 80k as of 21 Oct). Low Swiss taxes. There is travel funding in case of papers accepted at important conferences. There are also exciting connections to the world of AI startups.

There is an opportunity to participate in teaching courses (in English) on machine learning, e.g., in the master's program at the University of Lugano.

Requirements are also published within the "Direttive interne SUPSI" (direttiva 7A, Art. 2) and within "Regolamento del personale SUPSI" on the website (follow SUPSI, Documenti ufficiali). Official language at IDSIA is English.

(Filled: one additional 4-month PhD student position in the Prototouch project with this online form.)

The Swiss AI Lab IDSIA was the smallest of the world's top ten AI labs listed in the 1997 "X-Lab Survey" by Business Week magazine, and ranked in fourth place in the category "Computer Science - Biologically Inspired". IDSIA's most important work was done after 1997 though. It is small but visible, competitive, and influential. For example, it won many international pattern recognition competitions. Its highly cited Ant Colony Optimization Algorithms broke numerous benchmark records and are now widely used in industry for routing, logistics etc (today entire conferences specialize on Artificial Ants). IDSIA is also the origin of the first mathematical theory of optimal Universal Artificial Intelligence and self-referential Universal Problem Solvers (previous work on general AI was dominated by heuristics). IDSIA's artificial Recurrent Neural Networks learn to solve numerous previous unlearnable sequence processing tasks through gradient descent, Artificial Evolution and other methods. In particular, IDSIA's LSTM RNNs are now available to billions of smartphone users, e.g., for speech recognition. Research topics also include complexity and generalization issues, unsupervised learning and information theory, forecasting, learning robots. IDSIA's results were reviewed not only in science journals such as Nature, Science, Scientific American, but also in numerous popular press articles in TIME magazine, the New York Times, der SPIEGEL, and many others. Numerous TV shows on Tech & Science helped to popularize IDSIA's achievements. IDSIA is affiliated with the University of Lugano (USI) and USI's Faculty of Informatics and SUPSI. Many IDSIA alumni went on to become professors. Google DeepMind is heavily influenced by our former students: DeepMind's first PhDs in AI and Machine Learning came from our lab, one of them co-founder, one of them first employee; others joined later. IDSIA is located just outside the beautiful city of Lugano in Ticino (pics), the scenic southern Swiss province. Milano, Italy's center of fashion and finance, is 1 hour away, Venice 3 hours.

Switzerland is the world's leading science nation. It is the origin of special relativity (1905) and the World Wide Web (1990), is associated with 105 Nobel laureates, and boasts far more Nobel prizes per capita than any other nation with over 1m population. It also has the world's highest number of publications per capita, the highest number of patents per capita, the highest citation impact factor, the most cited single-author paper, etc, etc. As of 2016, Switzerland is the world's most competitive country for the 7th year in a row, according to the World Economic Forum. It's also the happiest country (1990s average), according to the Happiness Foundation. More.

Videos of talks by Juergen Schmidhuber
Robot Learning
Artificial Curiosity
Feedback Network
Deep Learning in Neural Networks: an Overview
Reinforcement Learning
The RNN Book
My first Deep Learner of 1991 + Deep Learning timeline 1962-2013
Subgoal learning
Deep Learning neural nets won the MICCAI 2013 Grand Challenge on Mitosis Detection
Pybrain Machine Learning Library
Evolino for time series prediction
Genetic Programming
SSA learns a complex task involving two agents and two keys
Optimal Ordered Problem Solver
Universal AI
Goedel machine
Theory of Beauty and Low- complexity Art
Randomness and Kolmogorov complexity
Speed Prior
Computing the Universe
videos of talks on deep learning in the US
Juergen Schmidhuber's talks at DLD 2016 - Europe's hottest conference ticket
Juergen Schmidhuber at the International Health Forum 2015. Credits: Wort & Bild Verlag / Eleana Hegerich
Click at squares above for related links