Hi there!
I’m a third-year PhD candidate at the Department of CSAI at Tilburg University. I’m part of the consortium project InDeep: Interpreting Deep Learning Models for Text and Sound, and happy to be supervised by Afra Alishahi, Willem Zuidema, and Grzegorz Chrupała.
Currently, I’m doing a research visit at ILCC, School of Informatics, University of Edinburgh, working with Ivan Titov.
Research
My research focuses on analyzing and interpreting deep neural models of language (written or spoken) by treating them as mathematical functions. I try to develop analysis methods that can faithfully elucidate the interplay and flow of information within neural networks.
Background
I completed my Master’s degree in Artificial Intelligence at Iran University of Science and Technology, where my research revolved around the interpretation of pre-trained language models and the utilization of interpretability techniques to accelerate their inference time, under the supervision of Mohammad Taher Pilehvar.
Before that, I got my Bachelor’s in Computer Engineering from Ferdowsi University of Mashhad. During that, I was working under the supervision of Ahad Harati as a team member of the Nexus RoboCup Simulation Team.
Services
- I’m co-organizing: InDeep Journal Club (since 2022), BlackboxNLP Workshop (2023, 2024)
- I reviewed for the following conferences: EMNLP (2022, 2023), ACL 2023, EACL 2023, ACL Rolling Review (since 2022)
News
- Mar 2024: 🚀 Materials (slides, notebooks, etc.) for EACL 2024 tutorial on “Transformer-specific Interpretability” are available here.
- Mar 2024: DecoderLens has been accepted to findings of NAACL 2024.
- Jan 2024: Started a research visit at ILCC, University of Edinburgh.
- Dec 2023: 🏅 Got an Outstanding Paper Award for Homophone Disambiguation Reveals Patterns of Context Mixing in Speech Transformers at EMNLP 2023!
- Mar 2023: New blog post: A few thoughts on why Value Zeroing.
- Jan 2023: Value Zeroing is out, a new interpretability method customized for Transformers (accepted to EACL’23 main conference).
- Dec 2022: BlackboxNLP will be back in 2023 at EMNLP! Happy to be serving as a co-organizer.
- Feb 2022: AdapLeR is out, up to 22x infrence speedup while retaining performance (ACL’22 main).
- Nov 2021: Moved to the Netherlands to join the consortium project: InDeep.
- Sep 2021: 🎓 Successfully defended my Master’s thesis titled “Interpretability and Transferability of Linguistic Knowledge in Pre-trained Language Models”.
- Sep 2021: Two papers accepted to EMNLP’21 (main conference and BlackboxNLP).
- Apr 2021: Our pre-print intepretability work is ready! Exploring the Role of BERT Token Representations to Explain Sentence Probing Results.