Hi! I’m Eric, and I’m a Senior Deep Learning Engineer at Skydio working on training and deploying efficient edge neural networks running on drones. You can download my CV here, and find me on google scholar here and LinkedIn here.

In 2021 I graduated from McGill University with a PhD in computer science, supervised by Joelle Pineau, where I worked on reinforcement learning and building VAEs with structured (object-like) latent representations. The main focus of my PhD was object discovery, i.e. given a dataset of images/videos/interactions, how can we identify common objects in the dataset, learn to detect them, and maybe even reason about them. More concretely, I focused on how to build deep probabilistic neural networks that can learn to detect and track objects in the visual stream without supervision. I built systems that can discover objects in images, videos, and, most recently, 3D worlds. I also explored how to build systems that can reason in terms of objects in ways that exploit their compositionality.

In 2014 I completed a Masters degree in Computer Science in the Computational Neuroscience Research Group at the University of Waterloo. I was supervised by Chris Eliasmith, and worked on a biologically plausible model of human knowledge representation. I also wrote an MPI implementation of the nengo neural simulator. In 2012 I obtained a BMATH(CS) degree, also from Waterloo, and spent my co-op terms working on a GPU implementation of nengo using CUDA.

When not working I like to travel, hike, play sports (squash, running and ultimate currently), play board games, and read books, especially sci-fi and non-fiction. My favorite authors are Kim Stanley Robinson, Neal Stephenson, Greg Egan, and Dan Dennett.

News

My paper Learning 3D Object-Oriented World Models from Unlabeled Videos received an Outstanding Paper Award at the Object-Oriented Learning workshop at ICML 2020.

Conference / Journal Articles

Exploiting Spatial Invariance for Scalable Unsupervised Object Tracking.
Eric Crawford and Joelle Pineau.
AAAI (2020). [code] [project] [supplementary] [arxiv]

Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks.
Eric Crawford and Joelle Pineau.
AAAI (2019). [code] [supplementary]

BanditSum: Extractive Summarization as a Contextual Bandit.
Yue Dong, Yikang Shen, Eric Crawford, Herke van Hoof and Jackie C.K. Cheung.
EMNLP (2018).

Modeling interactions between speech production and perception: speech error detection at semantic and phonological levels and the inner speech loop.
Bernd J. Kroger, Eric Crawford, Trevor Bekolay and Chris Eliasmith.
Frontiers in Computational Neuroscience (2016).

Biologically plausible, human-scale knowledge representation.
Eric Crawford, Matthew Gingerich and Chris Eliasmith.
Cognitive Science (2015). [code]

Biologically plausible, human-scale knowledge representation.
Eric Crawford, Matthew Gingerich and Chris Eliasmith.
Conference of the Cognitive Science Society (2013). [code]

Workshops, Preprints, Theses, Reports

Learning Object-Oriented Models of the Visual World.
Eric Crawford. PhD Thesis, McGill University (2021).

Learning 3D Object-Oriented World Models from Unlabeled Videos.
Eric Crawford and Joelle Pineau.
ICML Workshop on Object-Oriented Learning (2020). Outstanding Paper Award.

Spatially Invariant, Label-free Object Tracking.
Eric Crawford and Joelle Pineau.
NeurIPS Workshop on Perception as Generative Reasoning (2019). Spotlight. [code]

Self-supervised Learning of Distance Functions for Goal-Conditioned Reinforcement Learning.
Srinivas Venkattaramanujam, Eric Crawford, Thang Doan, and Doina Precup.
arXiv preprint arXiv:1907.02998 (2019).

Spatially Invariant Attend, Infer, Repeat.
Eric Crawford and Joelle Pineau.
NeurIPS Workshop on Modeling the Physical World (2018). [code] [poster]

Sequential Coordination of Deep Models for Learning Visual Arithmetic.
Eric Crawford, Guillaume Rabusseau and Joelle Pineau.
arXiv preprint arXiv:1809.04988 (2017).

Policy Gradient Methods for Reinforcement Learning.
Eric Crawford. Ph.D. Comprehensive Exam, McGill University (2015).

Biologically plausible, human-scale knowledge representation.
Eric Crawford. Master of Mathematics Thesis, University of Waterloo (2015). [code]

Learning large-scale heteroassociative memories in spiking neurons.
Aaron Voelker, Eric Crawford and Chris Eliasmith.
Unconventional Computation and Natural Computation (2014).