Brian Dolhansky

Skills

Research: Recommender systems, sequence modeling, representation learning, adversarial machine learning, computer vision
Engineering: Python, Scala, PyTorch, Spark, Bazel, C++, large-scale dataset construction and release, agentic workflows
Other: Behavior modeling, responsible AI, adtech, submodular functions, digital signal processing

Experience

Yobi

Founding Engineer, Core Signals Lead

Jul 2023 – present
Remote

Led the team responsible for developing the behavioral foundation model that powered Yobi’s Audiences product.¹
Developed very-large-scale unified graph and sequence models trained on heterogeneous datasets (trillions of rows).
Built the infrastructure supporting weekly versioning, production, and delivery of embeddings within a changing data landscape.
Worked with partner teams to develop an offline/online testing framework for rapid model and feature iteration.

Facebook

Senior Research Engineer, Facebook AI

Feb 2017 – Feb 2021
Seattle & New York

Founding engineer, tech and research lead of the AI Red Team,² which discovers AI vulnerabilities within the Facebook ecosystem.
Tech lead for the $1M DeepFake Detection Challenge^{3, 4, 5} and the release of the largest current DeepFake dataset.
Developed and published a novel algorithm to automatically open closed or blinking eyes in photographs.^{6, 7}
Designed and deployed a logo detection system for integrity purposes at web scale (millions of images per second).
Prototyped the first version of the background “in-painting” system used for the 3D Photos in Feed product.
Built and deployed a computer vision system to intelligently remove (“in-paint”) text from advertisement images.
Contributed to selfie-segmentation computer vision technology used in a variety of Facebook products.

Internships

2016
Machine Learning Intern, Growth Team, Instagram. Built new ranking models for users with low follow counts that raised the app-wide follow rate.
2015
Machine Learning Intern, Dato/Turi (since acquired by Apple). Developed and shipped the “Diversity” GraphLab toolkit for diverse sampling over billions of items.
2014
Machine Learning Scientist Intern, Demand Forecasting Team, Amazon. Developed a forecasting model that outperformed Amazon’s production model for the tools and equipment vertical.

Academia

2012–16
Ph.D. Candidate, University of Washington. Developed the first mathematical foundations of deep submodular functions and studied submodular approaches to dropout, determinantal point processes, and graph sampling.
2012
Teaching Assistant, CIS 520, University of Pennsylvania. Taught lectures for the graduate machine learning course and designed and ran the final project.
2011
Research Assistant, DAAD RISE Fellow, Technische Universität Dortmund. Built a power consumption model and monitoring system for Android mobile devices.

Education

University of Washington

Ph.D. Candidate, M.S. Computer Science · transferred from the University of Pennsylvania

Aug 2012 – Dec 2016
Seattle, WA

Drexel University

M.S. Electrical Engineering, B.S. Electrical Engineering, Computer Science minor

Aug 2007 – Jun 2012
Philadelphia, PA

Publications, Presentations, and Patents

2023
Systems and methods for making high value account recommendations. US Patent No. 11,669,915. Si, Dolhansky, et al.
2022
Exposing vulnerabilities of deepfake detection systems with robust attacks. Hussain, Dolhansky, et al.
2021
Towards measuring fairness in AI: the Casual Conversations dataset. Hazirbas, Dolhansky, et al.
2021
Systems and methods for protecting data. US Patent No. 10,915,663. Ferrer, Dolhansky, et al.
2020
Adversarial Collision Attacks on Image Hashing Functions. Dolhansky, Ferrer.
2020
The DeepFake Detection Challenge Dataset. Dolhansky, Bitton, Pflaum, Lu, Howes, Wang, Ferrer.
2020
Adversarial Evaluation of Multimodal Models under Gray Box Assumptions. Evtimov, Howes, Dolhansky, Firooz, Ferrer.
2020
Adversarial Threats to DeepFake Detection: A Practical Perspective. Neekhara, Dolhansky, Bitton, Ferrer.
2020
Automated detection of tampered images. US Patent No. 10,810,725. Dolhansky, Ferrer, Hsin.
2020
Image in-painting for modified viewpoint rendering. US Patent No. 10,789,723. Kopf, Dolhansky, Alsisan.
2020
DeepFakes and the DeepFake Detection Challenge. Presentation to the Partnership on AI. Dolhansky.
2020
Building the DeepFake Detection Challenge. AI NEXTcon. Dolhansky.
2019
Deep Poisoning Functions: Towards Robust Privacy-safe Image Data Sharing. Guo, Dolhansky, Hsin, Dinh, Wang, Ferrer.
2019
Automatic image correction using machine learning. US Patent No. 10,388,002. Ferrer, Dolhansky, Meyer, Morton.
2018
Eye in-painting with exemplar generative adversarial networks. CVPR 2018. Dolhansky, Ferrer.
2016
Deep submodular functions: Definitions and learning. NeurIPS 2016. Dolhansky, Bilmes.
2012
Analyzing the perceptual salience of audio features for musical emotion recognition. Schmidt, Dolhansky, et al.
2012
Relating perceptual and feature space invariances in music emotion recognition. Schmidt, Dolhansky, et al.
2011
Designing an expressive virtual percussion instrument. Dolhansky, McPherson, Kim.
2011
Teaching STEM concepts through music technology and DSP. Kim, Dolhansky, et al.

Full list on Google Scholar.

Awards and Honors

2012
GRFP Recipient, National Science Foundation
2012
ASEE SMART Scholarship Recipient (declined), Department of Defense
2011
RISE Fellow, Deutscher Akademischer Austauschdienst
2007
Drexel Presidential Scholarship, Drexel University
2007
Robert C. Byrd Academic Scholarship Recipient, State of New Jersey
2007
US Presidential Scholars Nominee, Department of Education
2007
Valedictorian, Triton Regional High School