Björn Bebensee Email: bebensee [at] snu.ac.kr GitHub: @qmdnls Twitter: @bebensee_ CV: pdf

About

Good day! I am currently a research engineer at Samsung Research where I work on pre-training large language models. I am broadly interested in how to train better LLMs faster. I also do research on dialogue understanding and published a paper introducing a novel span-based architecture for schema-guided dialogue state tracking

I completed my Master's degree at Seoul National University where I also worked as a Graduate Research Assistant in the Biointelligence Laboratory supervised by Byoung-Tak Zhang. There, I did research on machine learning at the intersection of language and vision. I wrote a paper proposing a co-attentional Transformer for video question answering that was accepted to ICASSP 2021.

Before getting into research I was an undergraduate student at RWTH Aachen where I studied computer science. I wrote my bachelor thesis on applying hidden Markov models to the problem space of Enterprise Architecture to automate the model creation process using network data. During my undergrad I also did a software project internship with ITERGO doing software development for ERGO Group. I also studied abroad at EWHA Womans Unversity in 2017.

Some things about me:

• Grew up near the town of Bremen, Germany
• Currently live in Seoul, Korea
• Moved to Korea for my Master's degree after initially coming as an exchange student

Research

• Span-Selective Linear Attention Transformers for Effective and Robust Schema-Guided Dialogue State Tracking,
  Björn Bebensee, Haejun Lee.
  ACL 2023
• Co-Attentional Transformers for Story-Based Video Understanding,
  Björn Bebensee, Byoung-Tak Zhang.
  ICASSP 2021
• Applying Dynamic Bayesian Networks for Automated Modeling in ArchiMate: A Realization Study,
  Björn Bebensee, Simon Hacks.
  EDOCW 2019
• Local Differential Privacy: a tutorial,
  Björn Bebensee.
  arXiv 2019

Research Interests

Large language models, i.e. how to train better models faster, and their applications to dialogue and natural language understanding.

Contact

Feel free to ping me on Twitter, or send me an email!