Publications

Deep Policy Dynamic Programming applies Dynamic Programming on a search space restricted using policy derived from a Deep Neural …

Ancestral Gumbel-Top-k sampling is a generic and efficient method for sampling without replacement from discrete-valued Bayesian …

We derive a low-variance unbiased estimator for expectations over discrete random variables based on sampling without replacement.

Stochastic Beam Search finds a set of unique samples (without replacement) from a sequence model. This paper received a Best Paper …

We construct REINFORCE estimators based on multiple samples with and without replacement and obtain a baseline for free!

Recent & Upcoming Talks

Talk about learning to solve routing problems at the Eindhoven Reinforcement Learning Seminar.

Invited talk at IPAM where I talk about Deep Learning and Combinatorial Optimization, especially Deep Policy Dynamic Programming.

AMLab seminar talk about Stochastic Beam Search, Ancestral-Gumbel-Top-k Sampling and the Unordered Set Estimator.

Talk about learning to solve routing problems at the TU Eindhoven RL seminar.

Virtual Poster at ICLR 2020.

Projects

IlikeCLR

IlikeCLR allows to browse and like papers from ICLR 2020 to get instant (maximum ‘like’-lihood) recommendations.

SET® Finder

SET® Finder is an Android/iOS app that uses TensorFlow to run a Deep Neural Network on-device to find SETs in the SET® card game.

Teaching

Reinforcement Learning 2019

I have been a Teaching Assistant for Reinforcement Learning 2019 at the University of Amsterdam, for the lab / homework sessions and supervision of group projects. I also gave a guest lecture on Monte Carlo Tree Search and AlphaGo.

Reinforcement Learning 2018

I have been a Teaching Assistant for the new Reinforcement Learning 2018 course at University of Amsterdam. I developed the computer labs and supervised homework and lab sessions. Additionally I gave a guest lecture about Monte Carlo Tree Search and AlphaGo.

Machine Learning 2018

I have been a Teaching Assistant for the computer labs for Machine Learning 2 in 2018.