Tom Schaul, Ph.D.

I am postdoc in machine learning at the Courant Institute of NYU, in the lab of Yann LeCun, focusing on learning from temporal sequences.

My research interests include (modular) reinforcement learning, stochastic (black-box) optimization and (deep and/or recurrent) neural networks. I use these techniques in application domains ranging from computer vision and neural decoding to games and robotics.

I grew up in Luxembourg and studied computer science in Switzerland (with exchanges at Waterloo and Columbia), where I obtained an MSc from the EPFL in 2005. I hold a PhD from TU Munich (2011), which I did under the supervision of Jürgen Schmidhuber at the Swiss AI Lab IDSIA.

Recent news

• 11-13/08/2013: presenting PyVGDL[34] at IEEE-CIG in Niagara Falls, Canada.
• 16-20/07/2013: presenting a new paper[32] on generalization of predictive representations at the IJCAI in Beijing, China.
• 01-05/07/2013: attending Dagstuhl seminar on Theory of Evolutionary Algorithms.
• 16-20/06/2013: presenting the no pesky learning rates paper[33] at ICML in Atlanta.
• 02-04/05/2013: presented[31], a paper on adaptive SGD for minibatches and non-smooth losses at ICLR.
• 28/04-01/05/2013: presented[W12] at the Eight Annual Barbados Workshop on Reinforcement Learning.
• 02/04/2013: plenty of new features in the latest release of a our video game description language PyVGDL.
• 25/03/2013: presentation of temporally coherent modular RL[W11] at the AAAI spring symposium on Lifelong Machine Learning (Stanford).
• 07/02/2013: invited talk at the Delft University of Technology.
• 09/01/2013: invited talk at the BrainLinks-BrainTools Symposium at the University of Freiburg.
• 18/12/2012: presenting adaptive learning rates[W10] at NIPS workshop on Optimization for Machine Learning.
• 07-09/11/2012: presenting new paper on modular RL[28] at ICDL-EpiRob in San Diego.
• 19/10/2012: talk[W9] at the 7th Annual Machine Learning Symposium at the NY Academy of Sciences.
• 08/10/2012: guest lecture at Courant: "Optimization for Machine Learning" [slides].
• 19/09/2012: second place for our bot "Clymily" at the Resistance competition of the Game/AI conference in Vienna [results].
• 09/07/2012: presented a NES convergence proof[27] at GECCO in Philadelphia.
• 07/07/2012: NES[W4][W5][W6][W7][W8] benchmarked at the Workshop for Real-Parameter Optimization.