• Math 171: Fundamental Concepts of Analysis
• Math 143: Differential Geometry
• Math 120: Groups and Rings
• EE 276: Information Theory
• EE 364M: Mathematics for Convex Optimization Lecture Notes

We suggest an approach to select which tasks should train together in multi-task learning models. Our method determines task groupings in a single training run by co-training all tasks together and quantifying the effect to which one task's gradient would affect another task's loss.

Restructuring Vector Quantization with the Rotation Trick
Christopher Fifty, Ronald G. Junkins, Dennis Duan, Aniketh Iyengar, Jerry W. Liu, Ehsan Amid, Sebastian Thrun, Christopher Ré
International Conference on Learning Representations (ICLR), 2025 (Oral Award)
arXiv / code / video

Context-Aware Meta-Learning
Christopher Fifty, Dennis Duan, Ronald G. Junkins, Ehsan Amid, Jure Leskovec, Christopher Ré, Sebastian Thrun
International Conference on Learning Representations (ICLR), 2024
arXiv / code / video

Efficiently Identifying Task Groupings for Multi-Task Learning
Christopher Fifty, Ehsan Amid, Zhe Zhao, Tianhe Yu, Rohan Anil, Chelsea Finn
Neural Information Processing Systems (NeurIPS), 2021 (Spotlight Award)
arXiv / code / blog

Step-size Adaptation Using Exponentiated Gradient Updates
Ehsan Amid, Rohan Anil, Christopher Fifty, Manfred K Warmuth
International Conference on Machine Learning (ICML), 2020
Beyond First Order Methods in Machine Learning Systems Workshop (Spotlight Award)
arXiv

Simplifying Graph Convolutional Networks
Felix Wu, Amauri Souza, Tianyi Zhang, Christopher Fifty, Tao Yu, Kilian Weinberger
International Conference on Machine Learning (ICML), 2019
arXiv / code

I founded a startup that transforms any picture into a work of art artificial intelligence. To save on GPU cloud computing costs, I built and host the deep learning server from my bedroom configured with a static IP from my ISP. The deep learning server communicates with a Firebase webserver via streaming GRPC and protobufs, and I wrote an improved neural style transfer algorithm that reduces runtime from 60 to 10 seconds. PJ deserves credit for coding up the Firebase webserver, front-end, and transaction processing logic. The logic of a customer going from their first stylization to the canvas arriving at their doorstep is fully automated.

I create education machine learning content focusing on TensorFlow, hardware, and Natural Language Processing. My channel has around 75 subscribers, over 10,000 views, and around 330 hours of watch time.

• CS 229: Machine Learning.
• CS 224n: Natural Language Processing with Deep Learning.
• CS 4820: Introduction to Analysis of Algorithms.
• CS 4740: Natural Language Processing.

• NeurIPS 2020-2023 (1x Outstanding Reviewer Award)
• ICML 2019-2025
• ICLR 2022-2025