Education
Saarland University
Oct 2025 - present
Ph.D. in Computer Science
Advisor - Prof. Dr. Michael Hahn
Saarland University
Oct 2023 - Aug 2025
M.Sc. in Computational Linguistics
GPA - 1.1, with distinction
Manipal Institute of Technology
Aug 2016 - Jul 2020
B.Tech. in Computer Science and Engineering, Minor in Intelligent Systems
CGPA - 9.09 / 10Publications
- Barriers to Universal Reasoning With Transformers (And How to Overcome Them). Kraus O.†, Sarrof, Y.†, Yao, Y., Koller, A., & Hahn, M. (2026). COLM 2026. (†Equal contribution)
- On the Ability of Transformers to Verify Plans. Sarrof, Y., Du, Y., Stein, K., Koller, A., Thi’ebaux, S., & Hahn, M. (2026). ICML 2026.
- Born a Transformer–Always a Transformer? On the Effect of Pretraining on Architectural Abilities. Jobanputra, M., Veitsman, Y., Sarrof, Y., Bakalova, A., Demberg, V., Pavlick, E., & Hahn, M. NeurIPS 2025.
- A Formal Framework for Understanding Length Generalization in Transformers. Huang, X., Yang, A., Bhattamishra, S., Sarrof, Y., Krebs, A., Zhou, H., Nakkiran, P., & Hahn, M. ICLR 2025.
- Homophonic Pun Generation in Code Mixed Hindi English. Sarrof, Y.. 1st Workshop on Computational Humour, COLING 2025.
- The Expressive Capacity of State Space Models: A Formal Language Perspective. Sarrof, Y., Veitsman, Y., & Hahn, M. NeurIPS 2024.
Research Experience
Research Assistant, Language, Computation and Cognition Lab, Saarland University
Dec 2023 - Sep 2025
Advisor - Prof. Dr. Michael Hahn
- Theoretically analyzed the capabilities and limitations of neural sequence-modeling architectures, especially Transformers and State Space Models.
- Studied length generalization in Transformers, including formal-language and algorithmic settings.
Teaching
- Theory of Machine Learning for Language Models, Co-Instructor, Summer 2026: Saarland University; co-taught with Michael Hahn.
- Milestones in Machine Learning, Lead Instructor, Winter 2025: Saarland University.
- Computational Linguistics, Tutor, Winter 2024: Saarland University.
Invited Talks and Presentations
- Formal Languages and Neural Networks Seminar (FLaNN), June 29, 2026: Length Generalization of Transformers with a Growing Test-Time Alphabet.
- AG1 Mittagsseminar, Max Planck Institute for Informatics, June 16, 2026: Length Generalization of Transformers with a Growing Test-Time Alphabet.
- Ploutus AI, 2026: Born a Transformer–Always a Transformer? On the Effect of Pretraining on Architectural Abilities.
- LCT Summer School 2025, 2025: Prompting Workshop.
- Computational Humour Workshop, COLING 2025, 2025: Homophonic Pun Generation in Code Mixed Hindi English.
- Formal Languages and Neural Networks Seminar (FLaNN), 2024: The Expressive Capacity of State Space Models: A Formal Language Perspective.
- TaCoS, Student Conference on Computational Linguistics, 2024: Transformers or RNNs or SSMs: Who’s More Sensitive?
Workshops and Summer Schools
- Eastern European Machine Learning Summer School (EEML), 2026: Accepted participant.
- FLaNN Workshop 2026, 2026: Poster, On the Ability of Transformers to Verify Plans.
- FLaNN Workshop 2026, 2026: Poster, On the Learnability of Chain of Thought.
- 1st Workshop on Computational Humour, COLING 2025, 2025: Workshop paper, Homophonic Pun Generation in Code Mixed Hindi English.
- Athens NLP Summer School, 2024: Participant.
Honors and Awards
- Outstanding Reviewer, CoNLL 2026: Reviewer recognition.
- Gold Reviewer, ICML 2026: Recognized among the top 25% of reviewers of the conference.
- Top Reviewer, NeurIPS 2025: Recognized among the top 10% of reviewers of the conference.
- Master’s with Distinction: Awarded for achieving an overall grade of 1.1 at Saarland University.
- Winner, Smart India Hackathon Software Edition 2019: Led a 6-member team to 1st place in a national 36-hour hackathon involving over 300,000 participants.
- Finalist, GE HackElt Pan India 2019: Selected as a top-10 national finalist for developing a predictive hospital management system to reduce patient wait times.
Academic Service and Organization
- Reviewer: COLM 2026, ICML 2026, CoNLL 2026, NeurIPS 2025.
- Organizer, LangLunches, 2025 - present: Biweekly research talk series at Saarland University.
- Organizer, ELLIS Poster Sessions, Saarbrücken, 2024 and 2025: Local ELLIS poster sessions for NeurIPS-associated work.
- Student Council Member, Department of Language Science and Technology, Saarland University, Apr 2024 - Mar 2025: Participated in departmental meetings and helped organize department-wide student events.
Industry Experience
Machine Learning Engineer, Glib.ai
Dec 2020 - Aug 2023
R&D team; financial document intelligence
- Lead developer on Finray, a financial statement analyzer that reduced manual financial spreading from approximately 9 to 2 person-hours per report.
- Implemented models for automated extraction of relevant tables from long documents.
- Configured and trained DETR-style models to detect table bounding boxes in document images.
- Developed internal libraries for information retrieval from unstructured text and models for document image restoration.
Student Trainee, Samsung R&D Bangalore
Jan 2020 - Jun 2020
On-Device AI division
- Designed and deployed compact deep learning models for text extraction and script identification.
- Curated datasets and automated parts of data cleaning for Samsung’s Alt Z Features.
- Developed data augmentation techniques for natural-scene images containing Indic-language text.
Project Intern, Novartis India Ltd.
Nov 2018 - Jul 2019
Hyderabad, India
- Implemented an end-to-end solution for comparing medical and pharmaceutical documents.
- Developed models to compare dosage information and detect medical terms in text corpora.
- Built a domain-specific algorithm for semantic similarity between medical sentences.
Selected Projects
Reproducibility Study of SAdam: An Adam Variant for Strongly Convex Functions
2020
OpenReview reproducibility project
- Implemented SAdam in PyTorch to complement the TensorFlow implementation by the authors.
- Reproduced and verified reported results across datasets and models from the original paper.
- Trained a language model on the Penn Treebank dataset using SAdam.
Parallelizing Mean-Filter-Based Edge Detection
2019
CUDA and MPI implementation
- Implemented adaptive edge detection using Wiener filters to counter Gaussian noise.
- Parallelized the implementation using CUDA and MPI in C.
Detect Earthquakes on a Live Camera Feed
2018
Computer-vision application
- Detected and tracked static background objects in video using OpenCV.
- Built a Django application to view multiple camera feeds and alert emergency contacts upon detecting oscillations.
Technical Skills
- Languages: Python, C, C++.
- Machine Learning: PyTorch, TensorFlow, Hugging Face ecosystem.
- Web/Tools: Django, Flask, Git, LaTeX, Docker, Linux.
- Certifications: Deep Learning Specialization, Coursera; Data Science Specialization, Coursera.
Last updated: 19th July 2026