Stanford University B.S. `18, M.S. `20
gangus [at] stanford [dot] edu
gangus [at] stanford [dot] edu
Geoffrey Angus
Resident
Computer Science
Google Research

I am a Resident in Google Research working on multimodal deep learning models for Image Search. I received a BS and MS in Computer Science from Stanford University, specializing in machine learning. I worked in both the Artificial Intelligence in Medical Imaging (AIMI) Center and HazyResearch lab, advised by Matthew Lungren and Chris Ré.

My most recent publication involves abnormality localization in full body PET-CT scans using weak supervision. We propose a method that allows for the automatic processing of thousands of natural language radiologist reports in the span of a few hours, for the purpose of training highly specialized abnormality classifiers. The full article can be found here.

I received my B.S. in Computer Science from Stanford University in 2018. I was class president, violinist, and teaching assistant in my spare time.
Projects & Papers

Cross-modal weak supervision enables abnormality localization in full-body PET-CT

Geoffrey Angus*, Sabri Eyuboglu*, Bhavik Patel, Anuj Pareek, Guido Davidzon, Jared Dunnmon, Matthew Lungren

We are among the first to build a high-performance machine learning model for abnormality localization in full body PET-CT scans. Our model detects lymph node, lung and liver abnormalities with median AUROCs of 87%, 85% and 92%, respectively. We propose a new method to generate labels from natural language radiologist reports to train weakly supervised image classifiers.

Nature Communications, 2021
[Code] [Paper] [Slides]

No Subclass Left Behind: Fine-Grained Robustness in Coarse-Grained Classification Problems

Nimit S. Sohoni, Jared A. Dunnmon, Geoffrey Angus, Albert Gu, Christopher Ré

Developed a novel weak supervision approach for recent distributionally robust optimization algorithms, reducing worst-case generalization error by up to 74% compared to standard procedures, without fine-grained label annotations.

NeurIPS, 2020
[Code] [Paper]

Via: Illuminating Undergraduate Academic Pathways at Scale

Geoffrey Angus*, Richard Diehl Martinez*, Mitchell Stevens, Andreas Paepcke

We synthesized 19 years of university enrollment data as a probabilistic graphical model to derive insights from the behavior of >52,000 students.

Learning @ Scale ACM conference, 2019
[Code] [Paper]

Regulatory Activity Prediction with Attention-based Models

Geoffrey Angus*, Sabri Eyuboglu*

We propose a new neural network model that leverages attention mechanisms in order to predict transcriptional profiles from a given DNA sequence.

CS273B: Deep Learning in Genomics and Bioinformatics.
[Code] [Paper] [Poster]

Abnormality Detection in Carotid Ultrasounds with Convolutional Networks

Geoffrey Angus*, Sabri Eyuboglu*, Pierce Freeman*, Bhavik Patel, Mu Zhou, Katie Shpanskaya, Kristen Yeom, Matthew Lungren

We attempt to use Computer Vision to find stenosis in the carotid artery. Spoiler: it's harder than it sounds.

CS231N: Convolutional Neural Networks for Visual Recognition.
[Code] [Paper] [Poster]

Exercise Transcription using a Long-Term Recurrent Convolutional Network

Geoffrey Angus*, Sabri Eyuboglu*, Pierce Freeman*, Rooz Mahdavian*

We train a model on a dataset consisting of micro-electromagnetic sensor data streams recorded at the gym in order to build an exercise transcription application for wearables.

CS230: Deep Learning.
[Code] [Paper] [Poster]

Wine Recommender System Using Multidimensional Clustering Methods

Geoffrey Angus*, Richard Diehl Martinez*, Rooz Mahdavian*

We train a recommender system on the sommelier reviews of over 150 thousand wines. Tell us a price point and we'll tell you what you want! 

CS229: Machine Learning.
[Code] [Paper] [Poster]
Teaching

Alexander Tung Memorial Fellow

July 2018 - September 2018

I was the recipient of the Alexander Tung Memorial Fellowship from the Haas Center for Public Service at Stanford. I spent nine weeks first developing, then teaching an introductory Computer Science course in Curitiba, Brazil. My partner, Sabri, and I lectured to 300 high school students across 3 different schools. We taught in Brazilian Portuguese in four of our nine sections. All of our hand-crafted lecture notes, slides, and assignments can be found here.

CS106A/B Section Leader

December 2015 - December 2017

I was a member of the CS198 program, Teaching Computer Science, at Stanford. Over the course of two years, I served as a section leader for six sections of 8-15 students. Syllabi containing the material that I taught can be found here and here.