doi.bio/sebastian_gehrmann

Sebastian Gehrmann

Overview

Sebastian Gehrmann is a researcher in the fields of natural language processing (NLP) and machine learning. He currently serves as Head of NLP in the CTO office at Bloomberg, where he works on long-term strategy and research for the company's products. Prior to this role, Gehrmann was a researcher at Google and holds a Ph.D. from Harvard University.

Education

Gehrmann obtained his Ph.D. from Harvard University.

Career

Gehrmann currently serves as the Head of NLP in the CTO office at Bloomberg, where he supports the development of language technology across the company. He previously worked as a researcher at Google.

Research Interests and Publications

Gehrmann's research interests include natural language generation, model evaluation, and interpretability. He has published extensively in these areas, with a particular focus on large multi-disciplinary collaborations such as the GEM benchmark. Some of his notable publications include:

• "The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics" (co-authored with Tosin Adewumi, Karmanya Aggarwal, and others)
• "Investigating gender bias in language models using causal mediation analysis" (co-authored with Jesse Vig, Yonatan Belinkov, Sharon Qian, and others)
• "LSTM Networks Can Perform Dynamic Counting" (co-authored with Mirac Suzgun, Yonatan Belinkov, and Stuart Shieber)
• "GLTR: Statistical Detection and Visualization of Generated Text" (co-authored with Hendrik Strobelt and Alexander M Rush)
• "Visual Interaction with Deep Learning Models through Collaborative Semantic Inference" (co-authored with Hendrik Strobelt, Robert Krüger, Hanspeter Pfister, and Alexander M Rush)

Co-authors

Throughout his career, Gehrmann has collaborated with several prominent researchers and academics in the field, including:

• Hendrik Strobelt (IBM Research / MIT-IBM Watson AI Lab)
• Alexander M. Rush (Associate Professor, Cornell University)
• Franck Dernoncourt (NLP/ML Researcher, MIT PhD)
• Yonatan Belinkov (Technion)
• Hanspeter Pfister (An Wang Professor of Computer Science, Harvard University)
• Stuart Shieber (Harvard University)
• Ankur P Parikh (Staff Research Scientist, Google DeepMind)
• Tom Sercu (Facebook AI Research)
• Cicero Nogueira dos Santos (Research Scientist, Google Research)
• Joy Tzung-yu Wu (Stanford Nuclear Medicine/Radiology and IBM Research Almaden)
• Leo Anthony Celi (Massachusetts Institute of Technology)
• Eric T Carlson (Director, Global Data Science, Merck)
• Thibault Sellam (Google Research)
• Dipanjan Das (Senior Director of Research, Google Deepmind)
• Yuntian Deng (Postdoc at AI2; Incoming Assistant Professor, University of Waterloo)
• Benjamin Hoover (IBM Research; GATech)
• Jesse Vig (Lead Research Scientist, Salesforce)
• Michael Behrisch (Assistant Professor of Computer Science, Utrecht University, Netherlands)
• Adam Perer (Carnegie Mellon University)

- Mirac Suzgun (Stanford University)

Sebastian Gehrmann

Overview

Sebastian Gehrmann is a researcher and Head of NLP in the CTO office at Bloomberg, where he supports the development of language technology with applications in finance and news. He previously worked as a researcher at Google and holds a Ph.D. from Harvard. His research interests include natural language generation, model evaluation, and interpretability. Gehrmann has published extensively in the fields of natural language processing, machine learning, and artificial intelligence.

Education

Gehrmann holds a Ph.D. from Harvard University.

Career

Bloomberg

• Head of NLP in the CTO office (current)

Google

• Researcher (former)

Publications

Gehrmann has numerous publications in prestigious conferences and journals, including:

• Proceedings of the 12th International Conference on Natural Language Generation (2018): "Generating Abstractive Summaries with Finetuned Language Models" and "Margin Call: an Accessible Web-based Text Viewer with Generated Paragraph Summaries in the Margin".
• EMNLP (2018): "Bottom-Up Abstractive Summarization".
• IEEE Transactions on Visualization and Computer Graphics (2018): "Seq2Seq-Vis: A Visual Debugging Tool for Sequence-to-Sequence Models".
• PLoS One (n.d.): "Comparing deep learning and concept extraction based methods for patient phenotyping from clinical narratives".
• International Journal of Medical Informatics (n.d.): "Behind the Scenes: A Medical Natural Language Processing Project".
• NeurIPS (2020): "Investigating gender bias in language models using causal mediation analysis".
• Nature Biomedical Engineering (n.d.): "Accelerating Antimicrobial Discovery with Controllable Deep Generative Models and Molecular Dynamics".
• ICLR Workshop on Deep Generative Models for Highly Structured Data (2019): "Interactive Visual Exploration of Latent Space (IVELS) for Peptide Auto-Encoder Model Selection".
• Proceedings of the 11th International Conference on Natural Language Generation (n.d.): "End-to-End Content and Plan Selection for Natural Language Generation" and "Towards Controllable Generation of Diverse Natural Language".
• IEEE Transactions on Visualization and Computer Graphics (2017): "LSTMVis: A Tool for Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks".

Co-authors

• Hendrik Strobelt
• Alexander M. Rush
• Franck Dernoncourt
• Yonatan Belinkov
• Hanspeter Pfister
• Stuart Shieber
• Ankur P. Parikh
• Tom Sercu
• Cicero Nogueira dos Santos
• Joy Tzung-yu Wu
• Leo Anthony Celi
• Eric T. Carlson
• Thibault Sellam
• Dipanjan Das
• Yuntian Deng
• Benjamin Hoover
• Jesse Vig
• Michael Behrisch
• Adam Perer
• Mirac Suzgun