Benjamin Winter is a recent addition to the DATEXIS group at Beuth University of Applied Sciences Berlin working as a research assistant. He recently graduated with distinction in the master of media informatics at Beuth University. His masters thesis "Applying Reinforcement Learning to small scale combat scenarios in the strategy game StarCraft" [git|pdf] was one of the first examples of Reinforcement Learning algorithms being able to beat expert human players at real competitive scenarios. An article based on this thesis was published on the Beuth Research Day 2018.

Currently he is working on the FashionBrain project backed by Zalando among others, which aims to gather a clearer understanding of Europe's Fashion Universe through the use of Big Data techniques.

Research Interests

  • Active Learning
  • Reinforcement Learning
  • Imitation Learning
  • Autonomous Driving

Publications

  • Benjamin Winter, Alexei Figueroa, Amy Siu, Felix Gers, Alexander Löser. KIMERA: Injecting Domain Knowledge into Vacant Transformer Heads. LREC 2022
  • Betty van Aken, Benjamin Winter, Felix A. Gers and Alexander Löser. VisBERT: Hidden-State Visualizations for Transformers. Accepted to The Web Conference 2020 (WWW'20) Demo Track [Code]
  • Betty van Aken, Benjamin Winter, Felix A. Gers and Alexander Löser. How Does BERT Answer Questions? A Layer-Wise Analysis of Transformer Representations. The 28th ACM International Conference on Information and Knowledge Management (CIKM'19) [PDF] [Code]

Contact

E-Mail: benjamin.winter (at) bht-berlin.de