Hello, my name is Chris Heinrich.
I recently finished a PhD in theoretical physics at the University of Chicago and am the co-founder of Triple, an augmented reality startup building software for the retail industry. I live in Pasadena, California with my wife Chen.
I’m interested in a broad range of topics in science and technology and have acquired an eclectic skill set by pursuing diverse interests over the years. I am equally at home chewing on hard research problems or taking on leadership roles and devising startup strategy. I have even done my fair share of sales and marketing. Read on to learn more about my interests and projects. You can view my full resume here.
I am the co-founder and CEO of Triple, a software startup that enables retailers to easily sell their products in augmented reality. We convert their existing product photography into a catalog of 3D models, and provide a white-labeled AR viewer which is used by their customers.
We have acquired a few high-profile customers in the furniture ecommerce space, including Burrow and Interior Define. As CEO I am responsible for nearly every role within the company from engineering to sales, marketing and fundraising. Our investors include the University of Chicago and Foundation Capital.
Knowvio publishes education apps on iOS for STEM subjects including physics, statistics, chemistry and more. I developed the curriculum for the original physics app, wrote a substantial part of the codebase and recruited and managed students to write content for the other apps. I built Knowvio as a side project while working on my PhD. All apps are live and continue to be downloaded and used by students across the globe.
I received a PhD in condensed matter theory, advised by professor Michael Levin, from the University of Chicago in 2017. My PhD work focused on studying topological phases in two dimensions. More specifically, we studied the interplay between local excitations in topological phases (anyons) and symmetries. To analyze these systems, which are generically strongly correlated and therefore quite complicated, we constructed exactly solvable models that captured the most important properties while still being amenable to analytic techniques.
One of the major results of my PhD was the ‘symmetry-enriched string net construction’. This work defined a concrete procedure for constructing exactly solvable models for certain symmetry enriched topological phases using only basic algebraic and topological data as input. For more info on this work see this paper. For a more lightweight overview of this work, as well as some following work, you can download the slides from my PhD defense here.
Fine-Grained Image Classification
I am one of the main contributors to deeptail, a project aimed at identifying individual hump-backed whales from images of their tails using deep convolutional neural networks. What makes this problem particularly challenging is that there are only a couple labeled images of each individual that can be used as training data for the classifier. We leveraged a pre-trained network and extensive data augmentation to achieve good accuracy on this challenging, and data limited, classification task.
Pixel-Wise Image Segmentation
In order to improve accuracy on the whale-tail classification task, we added image segmentation as a pre-processing step. By separating the whale-tail from the background (e.g. water, sky and horizon) we are hoping to improve classification accuracy.
For image segmentation we used SegNet, a deep encoder-decoder architecture for multi-class pixelwise image segmentation. I first applied unsupervised image segmentation algorithm to our dataset, and used the images where this performed well as training data for SegNet. Code for our segmentation pipeline can be found here: image segmentation
I have recently been experimenting with deep generative techniques, with a focus on image data using generative adversarial networks (GANs). Faceify is a small project I worked on to use an image generator, pretrained on the CelebA dataset, to generate an image that has the style of a celebrity face, but resembles the content of a user-supplied target image.
In my free time I enjoy playing music (violin, beatbox, guitar), learning languages (Mandarin, Italian) and exercising (running, swimming, hiking).