Chris Rockwell

I am a second year Master's student at the University of Michigan in Computer Science and Engineering. I am currently researching in computer vision and machine learning with Prof. David Fouhey and previously worked with Prof. Jia Deng .

I previously worked as a trader at Citadel LLC , where I applied statistical arbitrage in fixed income. Earlier, I structured exotic options and created systematic hedging strategies at BNP Paribas . I also did my Bachelor's at the University of Michigan, where I briefly researched High Frequency Trading and AI under Prof. Michael Wellman.

Github  /  CV  /  LinkedIn

profile photo

Within computer vision, I'm most interested in using human pose to understand scenes, particularly in video. I also have research experience in meta-learning.

net1 Hourglass Networks with Top-Down Modulation for Human Pose Estimation
Graduate Research, Princeton Vision and Learning Lab. Summer 2018 - Winter 2019.
Advisor: Alejandro Newell and Prof. Jia Deng
[code: attention][pretrained model, 2HG attention]
[code: regularization][pretrained model, 8HG regularization]

We increase performance of Newell et al.'s Stacked Hourglass Networks on MPII using a simple additional bottleneck to decoder network used as attention. Using the 2HG model, this improved validation performance 0.4% and improved performance 0.7% when also using cutout and vertical flipping. On the 8HG model, we were also able to improve test performance from 90.9% to 91.3% by simply adding cutout and vertical flipping. Finally, we increased precision of network confidence and explored utilizing confidence for curriculum sampling of tail cases.

meta A Simple Baseline on Meta-Dataset
Graduate Research, Princeton Vision and Learning Lab. Spring 2019.
Advisor: Alejandro Newell and Prof. Jia Deng

We improve a simple fine-tuning baseline on Meta-Dataset to within 0.1 average rank (minimum reportable difference) of the authors' best meta-learning based method. Higher regularization on fine-tuning layer compared to backbone explains most of the improvement. Using Logistic Regression or an SVM instead of an FC layer for fine-tuning produce similar results.

clean-usnob Market Fragmentation and the Latency Arms Race
Undergraduate Research, Strategic Reasoning Group. Summer 2013.
Advisor: Erik Brinkman, Elaine Wah and Prof. Michael Wellman
[poster] [presentation]
I was featured in a UMSI Youtube video.

I assisted Erik Brinkman in expanding the agent-based latency arbitrage simulation of Wah and Wellman to model a prisoner's dilemma in the high-frequency trading space.

clean-usnob Replicating and Improving Stacked Hourglass Networks for Human Pose Estimation
Chris Rockwell, Uzziel Cortez, Eric Huang.
Course project, EECS 545 Machine Learning, Fall 2018.
Instructor: Prof. Clayton Scott.
[report] [code] [pretrained model, 8HG]

We replicate and improve upon validation accuracy from Stacked Hourglass Networks for Human Pose Estimation using Adam and larger batch size. I led implementation of the project and it was jointly useful for research. My pytorch implementation was published in the princeton-vl github.

clean-usnob Evaluating Scene Graph-Generated Images using Visual Question Answering
Chris Rockwell.
Course project, EECS 692 Advanced AI, Winter 2019.
Instructor: Prof. John Laird.
[presentation] [report]

I replicate and summarize Image Generation from Scene Graphs and evaluate replicated generated images, original generated images, and ground truths using VQA.

clean-usnob GTA Perception
Richard Higgins, Parth Chopra, Chris Rockwell, Sahib Dhanjal, Ung Hee Lee.
Course project, EECS 598/ROB 535 Self-Driving Cars, Fall 2018.
Instructors: Prof. Matthew Johnson-Roberson and Prof. Ram Vasudevan

Squeeze and Excitation ResNet finetuned to classify objects appearing in road-scene images in the Driving in the Matrix dataset. With improved sampling and data augmentation, we finished top 10 in the class. I helped with implementation and improved data augmentation.

clean-usnob Link Prediction on the Patent Citation Network
Samuel Chen, David Dang, Robert Macy, Chris Rockwell.
Course project, EECS 598 Advanced Data Mining, Winter 2019.
Instructor: Prof. Danai Koutra.
[poster] [report] [code]

We compare several methods for link prediction on partitions of the Patent Citation Network dataset for the first time. The very sparse nature of this graph yielded low performance compared to a more typical dataset, but methods performed well in a temporal setting. I led temporal experiments and adapted representation learning method SDNE for our setting.

um2 EECS 498/598 Deep Learning, Winter 2019 - Grader
Instructor: Prof. Honglak Lee
Guest blog post on (Michigan Economics) Professor Miles Kimball's blog Confessions of a Supply Side Liberal