Naomi Saphra (original) (raw)

Naomi Saphra

Naomi Saphra

Research Fellow

About Me

I am a research fellow at the Kempner Institute at Harvard University and incoming Assistant Professor in Boston University’s faculty of Computing & Data Science. I am interested in NLP training dynamics: how models learn to encode linguistic patterns or other structure and how we can encode useful inductive biases into the training process. Recently, I have begun collaborating with natural and social scientists to use interpretability to understand the world around us. I have become particularly interested in fish. Previously, I earned a PhD from the University of Edinburgh on Training Dynamics of Neural Language Models; worked at NYU, Google and Facebook; and attended Johns Hopkins and Carnegie Mellon University. Outside of research, I play roller derby under the name Gaussian Retribution, perform standup comedy, and shepherd disabled programmers into the world of code dictation.

I am recruiting PhD students to begin in 2026 at Boston University.

Download CV

Interests

Education

My Research

My core agenda focuses on a single goal: to completely and comprehensively understand language model training. This objective combines linguistics, optimization, learning dynamics, science of deep learning, interpretability, and behavioral analysis. Recently, I have begun using similar approaches to study scientific discovery models and enhance broader scientific understanding.

My current publication list is available on my Google Scholar.

Apr 24, 2025

Sep 17, 2023

Jun 7, 2022

Against Monodomainism

Apr 28, 2021

A petty rant on the exceptional treatment of computer vision applications, directed at the machine learning community.

Apr 28, 2021

USVSN Sai Prashanth, Alvin Deng, Kyle O'Brien, Jyothir S V, Mohammad Aflah Khan, Jaydeep Borkar, Christopher A. Choquette-Choo, Jacob Ray Fuehne, Stella Biderman, Tracy Ke, Katherine Lee, Naomi Saphra (2025).Recite, Reconstruct, Recollect: Memorization in LMs as a Multifaceted Phenomenon.International Conference on Learning Representations (ICLR).

CiteURL