Logan Engstrom Email: engstrom@mit.edu Google Scholar: here CV/Resume: here GitHub: @lengstrom

About

I am a final year PhD student at MIT EECS advised by Aleksander Mądry and funded by a Google PhD Fellowship. I am excited about basic research across machine learning, including in data attribution/selection, adversarial examples, deep RL, the science of DL, and accelerating ML systems.

I spent four delightful years at MIT for undergrad and grew up in Massachusetts. Outside of research, I really like programming and playing pickup soccer.

Selected Work

1. Logan Engstrom*, Andrew Ilyas*, Benjamin Chen*, Axel Feldmann, William Moses, and Aleksander Mądry. Optimizing ML Training with Metagradient Descent. 2025.

2. Andrew Ilyas*, Sam Park*, Logan Engstrom*, Guillaume LeClerc, Aleksander Madry. Datamodels: Predicting Predictions from Training Data. ICML 2022.

3. Logan Engstrom*, Andrew Ilyas*, Shibani Santurkar, Dimitris Tsipras, Firdaus Janoos, Larry Rudolph, and Aleksander Mądry. Implementation Matters in Deep RL: A Case Study on PPO and TRPO. ICLR 2020 Oral Presentation.

4. Andrew Ilyas*, Shibani Santurkar*, Dimitris Tsipras*, Logan Engstrom*, Brandon Tran, and Aleksander Madry. Adversarial examples are not bugs, they are features. NeurIPS 2019 Spotlight Presentation.

5. Dimitris Tsipras*, Shibani Santurkar*, Logan Engstrom*, Alexander Turner, and Aleksander Mądry. Robustness May Be at Odds with Accuracy. ICLR 2019.

6. Anish Athalye*, Logan Engstrom*, Andrew Ilyas*, and Kevin Kwok. Synthesizing Robust Adversarial Examples. ICML 2018.

Open Source

• FFCV: (Very) fast data loading for machine learning
• Fast Style Transfer: TensorFlow CNN for fast style transfer
• Sistine: Webcam-based touch screen for any laptop
• TensorFire: accelerated neural networks in the browser using WebGL
• Tensorguard: Runtime typechecking for machine learning
• gitlinks: Go-links for GitHub pages
• Hextris: Fast paced HTML5 puzzle game played more than 15 million times
• Falcon: Chrome extension for full text history search