I'm Drew Blount, a machine learning engineer. I've spent my career working on projects for the public interest.

I spent the last seven years at Wild Me, a computer vision wildlife nonprofit. My job was training ML models to identify individual animals in photographs--think face ID for whales. I published a lot, led workshops on three continents, gave a keynote address at the Humpback Whale World Congress, and met some amazing biologists who are friends for life. I trained hundreds of ML models as the lead of all ID work--the core business--at Wild Me. Then in late 2022, Wild Me fired all its engineers. C'est la vie.

I've also worked as a nuclear reactor operator, computer science research assistant, and as the programming and math guy in a couple ill-fated, socially progressive crypto startups.

Now, I'm looking for work while taking some time for myself. The same month Wild Me fired everyone, my dad suddenly died. C'est la vie.

My skills are in ML, pure math, full-stack development, and technical communication. If you're interested in my measurements, here's my CV.

Here's an account of some old projects. If you'd like to read about the more distant past, feel free to snoop around my GitHub or my CV.

Thesis with advisor Jim Fix

• Explored kriging-like methods for global optimization of expensive blackbox functions, using this paper (Springer) as a starting point
• Built an optimization pipeline in Python and NumPy

Evolution in the Patent Network with Mark Bedau and the Reed College Artificial Life Lab

• Investigating whether Darwinian evolution occurs in the citation network of US patents

• Database management in MongoDB, Python, and NumPy

• Experimenting with automated trait extraction from patent texts using tf-idf, LDA topics, and word2vec

• Source on GitHub

Machine Earning with Jacob Menick

• For fun, not profit
• Working on an automated trading system for cryptocurrencies
• Our goal is to utilize reinforcement learning both to make trading decisions and to optimize execution of buy/sell orders

Artificial Life Lab

In Mark Bedau's Philosophy of Biology class in Spring 2014, I wrote a paper proposing a test that would empirically identify adaptations in an evolving system. It became an independent research project, and I wrote a couple little papers.

• D. Blount, “Identifying Adaptations by Measuring the Trait Derivative of Fitness,” in ALife 14: Proc. of the 14th Intl. Conf. on the Synthesis and Simulation of Living Systems. Hiroki Sayama et. al., eds. MIT Press, 2014

• D. Blount, “A General Statistical Method for Identifying Adaptations by Parameterizing Trait Space,”accepted to Artificial Life. Mark Bedau, ed. MIT Press, 2015. Submitted to the special conference edition of the journal at the request of the ALife 14 Conference Committee.

Teuscher.:Lab

I worked at Christof Teuscher's lab at Portland State University in the summer of 2013, on a project to construct a neural network in an artificial chemistry—imagine simulated chemical interactions emulating neurons. We finally got a two-layer chemical feedforward neural network to learn XOR in late 2014, and along the way published a technical paper about our web-based simulation framework.

• P. Banda, D. Blount, and C. Teuscher, “COEL: A web-based chemistry simulation framework," in CoSMoS 2014: Proceedings of the 7th Workshop on Complex Systems Modelling and Simulation, S. Stepney and P. Andrews, Eds. Luniver Press, 2014.

• D. Blount, P. Banda, C.Teuscher, and D. Stefanovic, “Feedforward Chemical Neural Network: A Compartmentalized Chemical System that Learns XOR,” submitted to IEEE Trans. Neural Networks and Learning Systems. Derong Liu, ed. IEEE. Submitted Dec 2014.

I like to make art with code. The little mesh-y gems decorating this webpage are emergent simulated swarms. I implementated Craig Reynold's 'Boids' algorithm, which approximates the behavior of flocking animals like swifts or schools of minnows. Rather than drawing each individual in the swarm, I drew the connections between neighbors in the swarm, where information passes from one boid to another. The vertices in a network aren't as important as the edges.