- Associate Professor, Massachusetts Institute of Technology (EECS and CSAIL)1
Jacob Andreas is an associate professor at MIT in EECS and CSAIL. Andreas's research aims to understand the computational foundations of language learning and to build systems that communicate with people and learn from human guidance in natural language. One line of the work asks what features and strategies black-box models learn: Andreas co-authored a study of in-context learning in transformers using linear models, and MILAN, a procedure that labels neurons in vision models with natural-language descriptions.123
At a glance
- Current role
- Associate Professor
- Company
- MIT
- Sector
- University
- Fields
- Language models, Interpretability
- Education
- Columbia, MIT, UC Berkeley, Cambridge
- Work listed
- 3
- Honors
- 3
- Sources cited
- 7
- Profile reviewed
Career
Jacob Andreas did PhD work at UC Berkeley as a member of the Berkeley NLP Group and the Berkeley AI Research Lab. Andreas is now an associate professor at MIT in the Department of Electrical Engineering and Computer Science and the Computer Science and Artificial Intelligence Laboratory.41
Research
Andreas's group studies language as a communicative and computational tool, organized around interactive learning from natural-language supervision, scientific understanding of neural models, and human-like language understanding. Papers in the first line include work on eliciting human preferences with language models (ICLR 2025) and on learning planning representations from language guidance (ICLR 2024).15
Work on understanding neural models includes "Implicit Representations of Meaning in Neural Language Models" (ACL 2021), the MILAN procedure for describing neurons in vision models (ICLR 2022), and an analysis of in-context learning with linear models (ICLR 2023). Andreas was also first author of Neural Module Networks, which composes question-specific networks from reusable modules for visual question answering.16327
Work and contributions
What learning algorithm is in-context learning? Investigations with linear models Published research
Co-author of the ICLR 2023 paper, which tests the hypothesis that transformer in-context learners implicitly implement standard learning algorithms, using linear regression as the test problem.21
Natural Language Descriptions of Deep Visual Features (MILAN) Published research
Last-listed author of the ICLR 2022 paper introducing MILAN, a procedure that automatically labels neurons in deep vision networks with natural-language descriptions.31
Neural Module Networks Published research
First author of the paper that assembles neural networks from reusable modules according to the structure of each question, applied to visual question answering.7
Timeline
- Released the Neural Module Networks paper as first author with Marcus Rohrbach, Trevor Darrell and Dan Klein.7
- Co-authored "Implicit Representations of Meaning in Neural Language Models", later at ACL 2021.61
- Co-authored the MILAN paper on natural-language descriptions of deep visual features.3
- Co-authored the study of in-context learning as an implicit learning algorithm using linear models.2
- Co-authored "Eliciting Human Preferences with Language Models", later at ICLR 2025.51
Honors and recognition
Where they work
- MIT 26 people in the wiki
Related people
- Mathias LechnerAlso at MIT
- Karen LivescuAlso at MIT
- Alexander AminiAlso at MIT
- Alex RivesAlso at MIT
- Michael CollinsAlso at MIT
- Nick FrosstLanguage models · Interpretability
- Arthur ConmyInterpretability · Language models
- Sameer SinghInterpretability · Language models
- Hendrik StrobeltInterpretability · Language models
- David BauInterpretability · Language models
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
7 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026
- Jacob Andreas @ MIT — homepage
- What learning algorithm is in-context learning? Investigations with linear models (arXiv 2211.15661)
- Natural Language Descriptions of Deep Visual Features (arXiv 2201.11114)
- Jacob Andreas @ MIT — Bio
- Eliciting Human Preferences with Language Models (arXiv 2310.11589)
- Implicit Representations of Meaning in Neural Language Models (arXiv 2106.00737)
- Neural Module Networks (arXiv 1511.02799)