Curriculum Vitae

Applied ML for Climate.

Ev

an Austen Coleman

Google Scholar / LinkedIn / Twitter / Github / Resume

About me

I am a Principal at C1 Ventures and a Research Affiliate at the MIT Climate & Sustainability Consortium. I have a PhD in Theoretical High-Energy Physics from Stanford University, and I invest in and incubate ventures to decarbonize industry and address the impacts of climate change. My current research work involves the development of data-driven algorithms and hardware capable of performing context-aware, in-the-wild material characterization.

In my doctoral studies, I worked on fundamental problems in physics regarding the behavior of quantum theories of gravity, and how we might construct them in a manner consistent with the observed accelerated expansion of the universe. How can we describe inflating universes (metastable de Sitter vacua) in string theory without relying on, for example, the KKLT prescription? How do various mechanisms in string theory reinforce the causal nature of spacetime, or excise infinitely-curved regions of spacetime? How do we use lessons from non-perturbative frameworks like finite-volume AdS holography to gain insight into holographic descriptions of our universe?

My coding experience stems from my desire early in life to program game engines. As a high school student at the York school in Monterey, California, I coded competitive analysis algorithms for commercial relational databases, developed JavaScript API's for "Big Data" businesses, and hacked together an annotation interface for Mars images at the NASA Ames Research Center. I also co-founded York's FIRST Robotics team, Deus Ex Machina, which earned a spot at the 2014 FIRST World Championships in St. Louis.

During my undergraduate years at Brown, I utilized the University's lack of strict pre-requisites (the Open Curriculum) to specialize in High Energy Physics (HEP). Under the direction of Professor of Physics Meenakshi Narain, I brought my computer science knowledge to CMS, a CERN experiment operating at the Large Hadron Collider. I spent a good fraction of my undergraduate years at the Fermi National Accelerator Laboratory and at the CERN site in Switzerland, working on precision top quark lifetime measurements, particle classifiers, and future detector studies.

Away from my desk, I enjoy music of all kinds, and play classical guitar. I am also an avid endurance cyclist, and I enjoy rock climbing and judo as well. My older sister is an assistant director for some pretty amazing films. My younger sister is a pre-med, a practicing medical translator, and an undergraduate researcher in gerontology.

Read More

Projects

VisNIR Hyperspectrum of SOC Optimizing NMR Spectroscopy Pulse Sequencing with RL for Soil Atomic Abundance
Rohan Shenoy, Evan Coleman, Hans Gaensbauer, Elsa Olivetti
NeurIPS, 2024
Climate Change AI Workshop: "Tackling Climate Change with Machine Learning"
paper

Motivated by the lack of scalable and inexpensive techniques for in-situ soil health monitoring, we focus on low-voltage nuclear magnetic resonance (NMR) spectroscopy as a promising new approach and develop a reinforcement learning technique to modulate NMR pulses for rapid atomic abundance assessment of soils.

VisNIR Hyperspectrum of SOC Structured spectral reconstruction for scalable soil organic carbon inference
Evan Coleman, Sujay Nair, Xinyi Zeng, Elsa Olivetti
International Conference on Learning Representations, 2024
Climate Change AI Workshop: "Tackling Climate Change with Machine Learning"
project page / paper / poster / code

We present an approach to improve the generalizability of soil organic carbon (SOC) inference models while enabling label-free training. In addition, by using a physics-informed model, we extract an interpretable reference spectrum for SOC (left).

Transformer for sampling optimization Simulation of soil carbon stocks at farm-scale Sampling without stratification: end-to-end methods for farm-scale soil carbon monitoring
Jenny Moralejo; advised by Evan Coleman, Sherrie Wang
MIT M.Eng. Thesis, 2024
paper

This thesis simulates and analyzes field-level soil carbon data to train models which perform 3 tasks in an end-to-end fashion: (1) determining soil carbon stocks in unsampled areas based on sparse measurement data, (2) leveraging historical data to predict future changes to soil carbon levels, and (3) determining the optimal number and siting of samples for the next measurement cycle, to maximize the predictive power achieved per dollar spent on measurement.

Service

  • Reviewer, Climate Change AI @ NeurIPS2024
  • Reviewer, NSF Small Business Innovation Research Program, Phase I
  • (Apr. 2019 - Jul. 2022): Volunteer, Stanford Educational Farm
  • (Jan. 2022 - May 2022): Exam Proctor, Stanford Office of Accessible Education
  • (Jul. 2018): Himalayan Trek to Fight Child Sex Trafficking

Recognition

  • (2022) Impact Fellowship, MIT Climate & Sustainability Consortium
  • (2021) Paul H. Kirkpatrick Award for Teaching in Physics
  • (2020) Dr. HaiPing and Jianmei Jin Fellowship in Physics
  • (2018) Youth Philanthropist of the Year, National Philanthropy Day Committee of the Central Coast
  • (2018) R. Bruce Lindsay Prize for Excellence in Physics
  • (2018) National Science Foundation Graduate Research Fellowship
  • (2017) Goldwater Scholarship
  • (2017) Astronaut Scholarship
  • (2017) Sigma Xi

Teaching

  • (SP2020) PHYSICS121: Advanced Electricity and Magnetism
  • (FA2019) PHYSICS70: Introduction to Special Relativity and Quantum Mechanics
  • (WI2019) PHYSICS40: Introduction to Classical Mechanics