Lisa Alazraki

Hi! I am a PhD student in the NLP Group at Imperial College London, advised by Marek Rei. I am broadly interested in generalisable learning, OOD robustness, and the relationship between reasoning and language.

Currently, I am a Research Scientist Intern in the Llama Reasoning and Planning team at Meta GenAI, where I am working with Akhil Mathur. Previously, I was a Research Intern at Cohere (2024) and Google (2022, 2023).

News

Aug 2025 We have released the AgentCoMa benchmark!

Aug 2025 No Need for Explanations: LLMs can implicitly learn from mistakes in-context is accepted to EMNLP Main 🎉

May 2025 New preprint! Reverse Engineering Human Preferences with Reinforcement Learning

Apr 2025 Going to NAACL to present Meta-Reasoning Improves Tool Use in Large Language Models ✈️

Research

My favourite recent papers are highlighted.

	AgentCoMa: A Compositional Benchmark Mixing Commonsense and Mathematical Reasoning in Real-World Scenarios Lisa Alazraki, Lihu Chen, Ana Brassard, Joe Stacey, Hossein A. Rahmani, Marek Rei arXiv 2025 paper / dataset / code / website
	How to Improve the Robustness of Closed-Source Models on NLI Joe Stacey, Lisa Alazraki, Aran Ubhi, Beyza Ermis, Aaron Mueller, Marek Rei arXiv 2025 paper / code
	Reverse Engineering Human Preferences with Reinforcement Learning Lisa Alazraki, Tan Yi-Chern, Jon Ander Campos, Maximilian Mozes, Marek Rei, Max Bartolo arXiv 2025 paper
	No Need for Explanations: LLMs can implicitly learn from mistakes in-context Lisa Alazraki, Maximilian Mozes, Jon Ander Campos, Tan Yi-Chern, Marek Rei, Max Bartolo EMNLP 2025 paper / code
	Enhancing LLM Robustness to Perturbed Instructions: An Empirical Study Aryan Agrawal, Lisa Alazraki*, Shahin Honarvar, Marek Rei (Equal contribution) ICLR 2025 BuildingTrust paper / code
	How Can Representation Dimension Dominate Structurally Pruned LLMs? Mingxue Xu, Lisa Alazraki, Danilo P. Mandic ICLR 2025 SLLM paper
	Meta-Reasoning Improves Tool Use in Large Language Models Lisa Alazraki, Marek Rei NAACL 2025 Findings paper / code
	How (not) to ensemble LVLMs for VQA Lisa Alazraki, Lluis Castrejon, Mostafa Dehghani, Fantine Huot, Jasper Uijlings, Thomas Mensink NeurIPS 2023 ICBINB paper

Teaching

I am a Teaching Assistant for 70050 Intro to Machine Learning, 70016 Natural Language Processing, 70010 Deep Learning and 40008 Graphs and Algorithms.

Supervision

2025 Kevin Zhou 'Optimal Data Mixing Strategies at Different Token Budgets for Language Model Training', MSc dissertation, co-supervised with Marek Rei (Imperial College London) and Kris Cao (Cohere).

2024 Aryan Agrawal 'Improving the Robustness of LLMs to Prompt Perturbations', MSc dissertation, co-supervised with Marek Rei (Imperial College London), Thomas Mensink (Google Research) and Shahin Honarvar (Imperial College London).

2023 Ziheng Zhang 'Knowledge-Enhanced Supportive Memory for Chatbots', MEng dissertation, co-supervised with the Algorithmic Human Development group (Imperal College London) and Zissis Poulos (University of Toronto).

2022 Alicia Law 'A Multilingual Chatbot for Self-Attachment Therapy', MSc dissertation (Distinguished Dissertation award), co-supervised with the Algorithmic Human Development group (Imperial College London).

Service

Conference reviewing: ACL 2025, AACL 2025, COLING 2025, COLM 2025, EMNLP 2025, NAACL 2025
Workshop reviewing: ACL-SRW 2025, BuildingTrust @ICLR 2025, ICBINB @ICLR 2025, MARW @AAAI 2025

Forked from Jon Barron's website.