ANDREW ILYAS

I am a PhD student at MIT, fortunate to be advised by Costis Daskalakis and Aleksander Madry. I am supported by the Open Philanthropy AI Fellowship.

I am on the academic job market this year.

I went to MIT for undergrad, double majoring in CS and Math. Outside of research, I enjoy playing soccer and table tennis.

Recent updates:

Research

I want to build machine learning systems that we can rely on. To this end, my research pursues a precise understanding of the entire ML pipeline, from training data (and the way we collect it), to learning algorithms, to the data we test on. A recurring theme of my work is that the data distribution we train on is usually not the one we care about—the former may be biased or low-quality, and the latter may be dynamic or adversarial. My main interests include:




  1. Tracing predictions back to training data: developing tools that decompose ML predictions as a function of individual training points, scaling these tools to modern algorithms and datasets, and leveraging them to improve our understanding of algorithms.
  2. Data bias: identifying cases where seemingly innocuous data pipelines introduce bias due to missing data or strategic behavior; building algorithms that can learn from data that is incomplete or strategically reported.
  3. Robustness: studying the implications and causes of adversarial vulnerability from a data perspective, and the ensuing properties of robust models.

I also like thinking more broadly about trust in AI systems, and have had the privilege of contributing to writings on social media regulation and AI deployment.

Selected Papers

* denotes equal contribution

Show all

TRAK: Attributing Model Behavior at Scale
Sung Min Park*, Kristian Georgiev*, Andrew Ilyas*, Guillaume Leclerc, Aleksander Madry (2023)
Project Page, Blog Post, GitHub
Oral presentation, ICML 2023

ModelDiff: A Framework for Comparing Learning Algorithms
Harshay Shah*, Sung Min Park*, Andrew Ilyas*, Aleksander Madry (2023)
Blog Post, GitHub
ICML 2023

Raising the Cost of Malicious AI-Powered Image Editing
Hadi Salman*, Alaa Khaddaj*, Guillaume Leclerc*, Andrew Ilyas, Aleksander Madry (2023)
Blog Post, GitHub
Oral presentation, ICML 2023

Rethinking Backdoor Attacks
Alaa Khaddaj*, Guillaume Leclerc*, Aleksandar Makelov*, Kristian Georgiev*, Hadi Salman, Andrew Ilyas, Aleksander Madry (2023)
Blog Post, GitHub
ICML 2023

When does Bias Transfer in Transfer Learning?
Hadi Salman*, Saachi Jain*, Andrew Ilyas, Logan Engstrom, Eric Wong, Aleksander Madry (2022)
Blog Post, GitHub

What Makes A Good Fisherman? Linear Regression under Self-Selection Bias
Yeshwanth Cherapanamjeri, Constantinos Daskalakis, Andrew Ilyas, Manolis Zampetakis (2022)
STOC 2023

Estimation of Standard Auction Models
Yeshwanth Cherapanamjeri, Constantinos Daskalakis, Andrew Ilyas, Manolis Zampetakis (2022)
Slides
EC 2022

Datamodels: Predicting Predictions from Training Data
Andrew Ilyas*, Sung Min Park*, Logan Engstrom*, Guillaume Leclerc, Aleksander Madry (2022)
Blog Post Part 1, Part 2, Data
ICML 2022

Constructing and adjusting estimates for household transmission of SARS-CoV-2 from prior studies, widespread-testing and contact-tracing data
Mihaela Curmei*, Andrew Ilyas*, Jacob Steinhardt, Owain Evans (2021)
medRxiv (previous draft), GitHub (Code and Data)
International Journal of Epidemiology

3DB: A Framework for Debugging Computer Vision Models
Guillaume Leclerc*, Hadi Salman*, Andrew Ilyas*, Sai Vemprala, Logan Engstrom, Vibhav Vineet, Kai Xiao, Pengchuan Zhang, Shibani Santurkar, Greg Yang, Ashish Kapoor, Aleksander Madry (2021)
Blog Post and Walkthrough, GitHub (Code and Demos), Quickstart and API Documentation

Unadversarial Examples: Designing Objects for Robust Vision
Hadi Salman*, Andrew Ilyas*, Logan Engstrom, Sai Vemprala, Aleksander Madry, Ashish Kapoor (2020)
Blog Post, GitHub
NeurIPS 2021

Do Adversarially Robust ImageNet Models Transfer Better?
Hadi Salman*, Andrew Ilyas*, Logan Engstrom, Ashish Kapoor, Aleksander Madry (2020)
Blog Post, GitHub (Code and Models)
Oral presentation, NeurIPS 2020

Noise or Signal: The Role of Image Backgrounds in Object Recognition
Kai Xiao, Logan Engstrom, Andrew Ilyas, Aleksander Madry (2020)
Blog Post
ICLR 2021

From ImageNet to Image Classification: Contextualizing Progress on Benchmarks
Dimitris Tsipras*, Shibani Santurkar*, Logan Engstrom, Andrew Ilyas, Aleksander Madry (2020)
Blog Post
ICML 2020

Identifying Statistical Bias in Dataset Replication
Logan Engstrom*, Andrew Ilyas*, Shibani Santurkar, Dimitris Tsipras, Jacob Steinhardt, Aleksander Madry (2020)
Blog Post
ICML 2020

Implementation Matters in Deep Policy Gradient Algorithms
Logan Engstrom*, Andrew Ilyas*, Shibani Santurkar, Dimitris Tsipras, Firdaus Janoos, Larry Rudolph, Aleksander Madry (2020)
Slides and video
Oral presentation, ICLR 2020

A Closer Look at Deep Policy Gradient Algorithms
Andrew Ilyas*, Logan Engstrom*, Shibani Santurkar, Dimitris Tsipras, Firdaus Janoos, Larry Rudolph, Aleksander Madry (2020)
Slides and video
Oral presentation, ICLR 2020

Image Synthesis with a Single (Robust) Classifier
Shibani Santurkar*, Dimitris Tsipras*, Brandon Tran*, Andrew Ilyas*, Logan Engstrom*, Aleksander Madry (2019)
NeurIPS 2019. Blog Post, Github

Adversarial Robustness as a Prior for Learned Representations
Logan Engstrom*, Andrew Ilyas*, Shibani Santurkar*, Dimitris Tsipras*, Brandon Tran*, Aleksander Madry (2019)
Blog Post, Github

Adversarial Examples are not Bugs, They are Features
Andrew Ilyas*, Shibani Santurkar*, Dimitris Tsipras*, Logan Engstrom*, Brandon Tran, Aleksander Madry (2019)
Blog Post, Datasets
Spotlight presentation, NeurIPS 2019

Prior Convictions: Black-box Adversarial Attacks with Bandits and Priors
Andrew Ilyas*, Logan Engstrom*, Aleksander Madry
ICLR 2019. Github

How Does Batch Normalization Help Optimization?
Shibani Santurkar*, Dimitris Tsipras*, Andrew Ilyas*, Aleksander Madry
Blog Post, Video (3 minutes)
Oral presentation, NeurIPS 2018.

Black-box Adversarial Attacks with Limited Queries and Information
Andrew Ilyas*, Logan Engstrom*, Anish Athalye*, Jessy Lin*
ICML 2018. Partial-Information/GCV Blog Post, Label-Only Attack Blog Post, Github

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

Training GANs with Optimism
Constantinos Daskalakis*, Andrew Ilyas*, Vasilis Syrgkanis*, Haoyang Zeng*
ICLR 2018. Github

Extracting Syntactic Patterns From Databases
Andrew Ilyas, Joana M.F. da Trindade, Raul C. Fernandez, Samuel Madden
ICDE 2018. Github

MicroFilters: Harnessing Twitter for Disaster Managment
Andrew Ilyas
Chairman's award winner, IEEE GHTC 2015.

Short Papers/Miscellanea

"On AI Deployment" Blog Post Series
Sarah Cen, Aspen Hopkins, Andrew Ilyas, Aleksander Madry, Isabella Struckman, Luis Videgaray
Part 1 Part 2 Part 3 Part 4

Social Media Blog Post Series
Sarah Cen, Andrew Ilyas, Aleksander Madry
Part 1 Part 2 Part 3 Part 4

FFCV: Fast Forward Computer Vision
Python Library. Homepage

The robustness python library
GitHub repository/PyPI package. Documentation on ReadTheDocs

A Game-Theoretic Perspective on Trust in Recommender Systems
Sarah Cen, Andrew Ilyas, Aleksander Madry (2022)
Talk Recording, Poster
Oral presentation, ICML Workshop on Responsible Decision-Making 2022

Evaluating and Understanding the Robustness of Adversarial Logit Pairing
Logan Engstrom*, Andrew Ilyas*, Anish Athalye* (2018)
NeurIPS Security in Machine Learning Workshop 2018