Herke van Hoof
Herke van Hoof is a reinforcement-learning researcher at the University of Amsterdam. He earned bachelor’s and master’s degrees in artificial intelligence at Groningen, completed a doctorate under Jan Peters at TU Darmstadt in November 2016, and held a postdoctoral position at McGill. His research focuses on modular and hierarchical reinforcement learning: sharing information across tasks can reduce the amount of data required to learn. His coauthored publications also cover sequence sampling, planning with a learned policy basis and program synthesis with neural components. He is an associate professor at Amsterdam as of October 2026.
- Now at
- University of Amsterdam
- Works
- 82
- Citations
- 4,850
- h-index
- 23
Publication numbers from OpenAlex, as of 14 October 2026.
Timeline
-
Before November 2016
Artificial-intelligence training
Van Hoof earned bachelor’s and master’s degrees in artificial intelligence at Groningen. He completed his doctorate at TU Darmstadt in November 2016 under Jan Peters.
-
After doctorate
McGill and Amsterdam
He held a McGill postdoctoral appointment before moving to Amsterdam. He collaborated with Joelle Pineau, Dave Meger and Gregory Dudek at McGill.
-
2019–2024
Sampling and planning
His publications include work on the Gumbel-Top-k method for sequence sampling in 2019 and planning with a learned policy basis in 2024. These methods address how algorithms sample possibilities and combine learned behavior.
-
2026
Reusable learning structure
A 2026 paper examines program synthesis with neurally interpreted languages. His research focus is modular reinforcement learning, including hierarchical structure that shares information between tasks and uses prior knowledge.
Sources and credits
Photo: amlab-amsterdam.github.io; AMLab | Amsterdam Machine Learning Lab | Herkevanhoof; creator not stated. License: Not stated by source. Source: amlab-amsterdam.github.io. Cropped to a square and resized. Ask us to remove or replace this photo.