Salma Abdel Magid

My last name is Abdel Magid (two words with a space).

Presidential Postdoctoral Scholar
Computer Science, Princeton University

Office:  Friend Center 207
         Princeton, NJ 08540
Email:   sabdelmagid [at] princeton [dot] edu

Education & Training

2025–27
Postdoctoral Fellow, Princeton University
Advisor: Dr. Olga Russakovsky
2019–25
Ph.D. Computer Science, Harvard University
Advisor: Dr. Hanspeter Pfister
2023–24
PhD Research Intern, Adobe
2016–18
M.Eng. Computer Engineering, Santa Clara University
Advisor: Dr. Behnam Dezfouli
2015–17
B.S. Computer Science, Santa Clara University
Salma Abdel Magid

I’m a Presidential Postdoctoral Fellow in Computer Science at Princeton, working with Dr. Olga Russakovsky in the Princeton Visual AI Lab, and a Future Values Fellow at the University Center for Human Values. My research aims to make generative AI safer by understanding how image and vision-language models cause harm and developing ways to measure and prevent it.

News

Nov 2026
I’m on the academic job market this year! I’m applying for faculty positions starting in 2027. Feel free to reach out.
Nov 2026
See you at NeurIPS 2026 in Sydney (Dec 6–12)! We’re presenting BodyBench and our hallucination-detection paper in the Evaluations and Datasets Track.
Nov 2026
BodyBench will also appear at the WiML Workshop and as a poster at TAE (Trust-AI-Eval), a NeurIPS 2026 workshop on whether we can trust AI evaluation.

Research

My vision is safer generative AI that mitigates real-world harm.

But, we traditionally study one model on one task. In practice, a generator’s outputs are collected into datasets that train other models (scorers, classifiers, the next generator), and those models act back on the generator by filtering, finetuning, and evaluating it.

Harm from generative AI is produced by an ecosystem of models, data, and people, so safety has to be understood, evaluated, and controlled at the level of that ecosystem. My current work is on image generation and editing, but the same loop holds for language models and other modalities.

UNDERSTAND

How harm spreads through models and data

Concept2Concept, TMLR 2025
Bias at the End of the Score, CVPR 2026

EVALUATE

Against the harm people actually face

BodyBench, NeurIPS 2026
Hallucination signals, NeurIPS 2026
Concept unlearning, T&S 2026

CONTROL

By intervening inside the model

Copying as Defense, preprint 2026

I’m always happy to collaborate, especially across disciplines. If any of this overlaps with your work, please reach out.

Papers

* equal contribution

  1. 2026
    Copying as Defense: Protecting Images from Editing by Eliciting Model Copying
    Salma Abdel Magid, Sanghyuk Chun, Eric Slyman, Kushal Kafle, Olga Russakovsky. Under review.
  2. 2026
    BodyBench: Evaluating Adversarial Image Defenses Against AI Nudification Inpainting
    Qiwei Li*, Salma Abdel Magid*, Olga Russakovsky, Eric Gilbert, Sarita Schoenebeck. NeurIPS 2026, Evaluations and Datasets Track. Also at the WiML Workshop and, as a poster, at TAE (Trust-AI-Eval): Can We Trust AI Evaluation?, a NeurIPS 2026 workshop. [pdf] [bib]
  3. 2026
    Robustness of Internal Signals for Hallucination Detection in Vision-Language Models
    Allison Chen, William Yang, Salma Abdel Magid, Jonathan Williams, Olga Russakovsky, Esin Tureci. NeurIPS 2026, Evaluations and Datasets Track. [pdf] [bib]
  4. 2026
    Bias at the End of the Score
    Salma Abdel Magid, Grace Guo, Esin Tureci, Vikram Ramaswamy, Amaya Dharmasiri, Hanspeter Pfister, Olga Russakovsky. CVPR 2026. [pdf] [bib]
  5. 2025
    Is What You Ask For What You Get? Understanding Concept Associations in Text-to-Image Models
    Salma Abdel Magid, Weiwei Pan, Simon Warchol, Grace Guo, Junsik Kim, Wanhua Li, Mahia Rahman, Hanspeter Pfister. TMLR 2025. [pdf] [bib]
  6. 2023
    Masked Image Training for Generalizable Image Denoising
    Haoyu Chen, Jinjin Gu, Yihao Liu, Salma Abdel Magid, Chao Dong, Lei Zhu. CVPR 2023. [pdf] [bib]
  7. 2022
    Texture Based Error Analysis of Image Super-Resolution
    Salma Abdel Magid, Zudi Lin, Donglai Wei, Yulun Zhang, Jinjin Gu, Hanspeter Pfister. CVPR 2022. [pdf] [bib]
  8. 2021
    Dynamic High-Pass Filtering and Multi-Spectral Attention for Image Super-Resolution
    Salma Abdel Magid, Yulun Zhang, Donglai Wei, Won-Dong Jang, Zudi Lin, Yun Fu, Hanspeter Pfister. ICCV 2021. [pdf] [bib]
  9. 2020
    Channel Embedding for Informative Protein Identification from Highly Multiplexed Images
    Salma Abdel Magid, Won-Dong Jang, Denis Schapiro, Donglai Wei, James Tompkin, Peter Sorger, Hanspeter Pfister. MICCAI 2020. [pdf] [bib]
  10. 2019
    Image Classification on IoT Edge Devices: Profiling and Modeling
    Salma Abdel Magid, Francesco Petrini, Behnam Dezfouli. Cluster Computing, 2019. [pdf] [bib]

Preprints and under review

  1. Conjunction Junction, What's Your Erasure Function? A Set-Theoretic Analysis of Representation Scrubbing
    Salma Abdel Magid, Sanghyuk Chun, William Yang, Esin Tureci, Olga Russakovsky. Under review.
  2. Mitigating Associative Bias in CLIP using Synthetic Counterfactuals
    Salma Abdel Magid, Jui-Hsien Wang, Kushal Kafle, Hanspeter Pfister. arXiv 2024. [arXiv]
  3. Revisiting RCAN: Improved Training for Image Super-Resolution
    Zudi Lin, Prateek Garg, Atmadeep Banerjee, Salma Abdel Magid, Deqing Sun, Yulun Zhang, Donglai Wei, Hanspeter Pfister. arXiv 2023. [arXiv]

Recent talks

  • Nov 2026  In the Lab ≠ In the Wild: Making Generative AI Safer. Department of Computer Science and Engineering Seminar. Santa Clara University
  • Sep 2026  Auditing the Text-to-Image Feedback System. Department of Computer Science Graduate Seminar. University of Maine
  • Aug 2026  Text-to-Image Feedback System. Beery Lab. Massachusetts Institute of Technology
  • May 2026  Auditing the Text-to-Image Feedback System. Computer Vision Seminar. Cornell University
  • Mar 2026  Bias at the End of the Score. Bias-in-AI Seminar. Princeton University
  • Oct 2024  Identifying and Mitigating Associative Bias in Vision-Language Models. Department of Computer Science Colloquium. Harvard University

Teaching

  • Nov 2026  Guest lecture: Generative AI Safety. COS 324: Introduction to Machine Learning, Princeton University
  • Apr 2026  Guest lecture: Machine Learning Safety. COS 324: Introduction to Machine Learning, Princeton University
  • Spr 2021  Teaching Fellow. DS 10: Elements of Data Science, Harvard University
  • Spr 2020  Teaching Fellow. DS 10: Elements of Data Science, Harvard University
  • Spr 2018  Grader. COEN 148/290: Computer Graphics, Santa Clara University
  • Spr 2018  Grader. AMTH 351: Quantum Computing, Santa Clara University
  • Fall 2017  Grader. CSCI 163: Theory of Algorithms, Santa Clara University
  • Fall 2017  Grader. COEN 210: Computer Architecture, Santa Clara University
  • Fall 2016  Teaching Assistant. COEN 21: Logic Design, Santa Clara University

Mentoring

  • Ruoming Shen (Princeton B.S. thesis, 2026–27, in progress)
  • Jessica Yan (Princeton B.S. thesis, 2025–26)
  • Mahia Rahman (Harvard B.S. thesis, 2024–25)
  • Beamlak Birhane (GenAI summer research intern, 2024)
  • Sein Yun (GenAI summer research intern, 2024)
  • Esha Pahwa (research intern, 2022)
  • Prachi Agrawal (research intern, 2021)

Honors

  • Princeton Presidential Postdoctoral Fellowship (2025–27)
  • Future Values Fellowship, University Center for Human Values (2025–26)
  • Siebel Scholarship (2024)
  • Harvard Generative AI Summer Research Grant (2024)
  • Harvard Graduate Prize Fellowship (2019–24)
  • MICCAI NIH Award (2019)

Service

Reviewer for ICLR (2025, 2026), CVPR (2023, 2025, 2026), ECCV (2020, 2024, 2026), ICCV (2023, 2025), and the CVPR Demographic Diversity Workshop (2025).