Colin Conwell (Coco)

Disclaimer: This site is (always) under construction! Please forgive any issues with rendering.

Biography

Friends call me Coco, and you can, too! I’m a hybrid academic and industry research scientist, working broadly in the areas of computational cognitive (neuro)science and machine learning (AI), but with a particular focus on the intersection of perception and language in the understanding (and reverse engineering) of human intelligence. As a graduate student in the Harvard Vision Sciences Lab, I worked mainly on understanding the role of rapid, subsymbolic perceptual computations in higher-order cognitive processes through the lens of emergent AI technologies. As an undergraduate, I studied comparative literature and international relations, with emphases on language learning, security, indigenous knowledge, and sustainable development.

The overarching question in my research these days is the question of how the mind makes meaning of the chaos that bombards our senses, and how we then communicate that meaning to other minds. I believe deeply that our ability to use language in service of effective communication is the cornerstone of our intelligence and our greatest hope for further human flourishing.

Outside of work (and sometimes at work), I live for the experience of beauty, embodied most particularly for me in two people speaking, with or without words; cookies and cream; well-designed computer programs; the moonlight reflecting off a lake onto the eyes of old friends sitting side-by-side on the dock; the Gricean maxims; Gordian knots; West African blues, Anatolian folk, deep dark techno, (any music with a saz, khora, microtonality, melisma, or thumping bass at 140BPM); calavera masks; stark desert still lifes; evolutionary spandrels; being a cat dad to a cat named Belle; adagio for strings; good neighbors with no fences; the wonderful, terrifying, seemingly limitless potential of the human spirit; and buffalo sauce.

A copy of my (likely outdated) curriculum vitae may be found at this link.

Tutorials

Here, you’ll find a collection of Google Colaboratory tutorials covering a range of topics, all with interactive Python code.

• Deep Dive into Brain + Behavior is an introduction to the DeepDive package, a codebase I and other collaborators have designed to facilitate transfer learning and read out from artificial deep neural networks to biological brains and behavior.
• Deep Learning Safari is broad introduction to various supervised, unsupervised, and multimodal deep learning pipelines in PyTorch.
• Deep Mouse Trap is a tutorial in using the features of deep nets to model neural data in the mouse brain (and other brains more generally).
• How to Lie with Statistics is both a foundational introduction to statistics and also a user’s guide on how to avoid falling for common mistakes in statistical literacy and reasoning.

Archives

Below is a random collection of curio from my life in science and art. I call them “archives”, but a better name might be something like “snapshots of a scatter-brain in science” or “mind fossils”. They’re basically a patchwork parchment of things I once thought, projects I began but never (really) finished, or ideas that never quite came to fruition. I decided to keep them here one day on a whim when (on the verge of deleting them in service of “an updated website”), I figured instead that fossils are informative not because they tell us how things should be, but because they tell us how things once were. Science is process, our minds evolve like bodies do, and ideas without history are like people without stories – (at best) very forgetful, (at minimum) very forgettable, and (at worst) forgotten altogether.

Here then for your viewing pleasure are some of my fossils:

Project Pages

Back in the day, I always loved reading publications from Distill, and once somehow thought I’d have the time to ‘distill’ all of my own projects in similar style… This didn’t quite go as planned for most of my projects, but here’s a sample of a few that made it:

• BlockBuster details an experiment comparing how well the responses of supervised and unsupervised deep neural networks predict human judgments in a classic intuitive physics task. The answer, in brief, and reinforced by machine-compatible psychophysics, is that they do pretty well – challenging views of intuitive physics that require simulation.
• Deep Mouse Trap is an exploration of which deep neural network models best predict neural responses in a large optical physiology dataset from the mouse visual cortex (courtesy of the Allen Brain Observatory).
• Deep Orientation asks where (if at all) orientation invariance emerges in the representational hierarchies of deep neural networks, and whether that invariance mirrors the invariance we observe in the human brain.
• Maritime Aesthetics reports the results of a collaboration with the Peabody Essex Museum, investigating the physiology of aesthetic experience and specific user interactions with artworks in the Museum’s catalogue.

… almost certainly more fossils to come as we continue to evolve further!

My People

Beyond anything else that makes me who I am, it is the people (family, friends, collaborators, mentors, mentees) that have taught me over the years.

Below is a list I’ve tried to make of at least some of those people I’ve been lucky enough to know and that I think you should know about, too! (I also try to keep a more exhaustive list here, but it’s not always as updated as I would like it to be.)

• The Harvard Vision Sciences Lab: Including my PhD advisors, George Alvarez and Talia Konkle, and my contemporaries there.
• Cognitive Science @ Johns Hopkins: Including my PostDoc advisors, Leyla Isik and Mick Bonner, and their students.
• Center for Brains, Minds, + Machines: An entire ecosystem of people, including my long-term collaborators and friends @ The MIT Info Lab, led by Andrei Barbu and Boris Katz.
• Greater Harvard / MIT Community: Including the many mentors of my PhD (Patrick Mair, Elizabeth Spelke, Susan Carey, Tomer Ullman, Josh Tenenbaum, Gabriel Kreiman).
• My Undergraduate Advisors: William Handley, Natania Meeker, Karen Huebner, Tara McPherson, and Justin Wood.
• Collaborators, Friends + Students Far too many to enumerate, but none forgotten! Check out the “Co-Authors” tag on my Scholar page for at least a somewhat automated list of people I’ve been fortunate to work with.)

And of course, my partner in science, art, and life alike: the inimitable Chelsea Boccagno.

Contact Info

Email: conwell[at]g[dot]harvard[dot]edu or colinconwell[at]gmail[dot]com