Visual saliency & attention
An information-theoretic account of what draws the eye (the AIM model), together with eye-tracking data, evaluation metrics and benchmarks for comparing models with human gaze.
About
I study how vision works, in brains and in machines, and what each can teach us about the other.
My research sits at the meeting point of computer vision, deep learning, human perception and computational neuroscience. Much of it asks a simple question in different guises: what information does a visual system keep, and why? That question has led to information-theoretic models of visual attention, to large-scale studies of human eye movements, and more recently to work on what deep neural networks actually encode, from absolute position to shape, texture and motion.
I completed a BSc (double major in Computer Science and Pure Mathematics) at the University of Guelph, an MASc in Systems Design Engineering at the University of Waterloo, and a PhD in Computer Science at York University. I held postdoctoral positions at INRIA in France and at Epson Canada, then served as Assistant and later Associate Professor at the University of Manitoba, and as a faculty member at Ryerson University (now Toronto Metropolitan University), before returning to Guelph.
Prospective students. I am always glad to hear from strong graduate and undergraduate students interested in vision, perception and deep learning. Please email me with a short note on your interests and a CV.
Research
Three threads that run through the work
What deep networks represent
Probing the hidden content of convolutional and spatiotemporal networks: how they encode absolute position, why padding matters, how they balance shape against texture, and what is static or dynamic in video models.
Dense prediction & applications
Segmentation and salient-object ranking, along with collaborations that carry vision and learning into human–computer interaction, security, medical physics and animal health.
Working papers
Attention, operators and the architecture of transformers
A monograph and four papers from a current research programme, released as working drafts in October 2026. Each has its own DOI on Zenodo, where later versions will appear. An overview of the programme is at mostlythoughtthrough.com.
- 2026 48 pp.
- 2026 74 pp.
- 2026 38 pp.
- 2026 72 pp.
- 2026 414 pp.
Publications
Selected papers
- 2005 1,637 cited
- 2009 1,244 cited
- 2020 576 cited
- 2024 122 cited
- 2017 298 cited
- 2018 162 cited
- 2021 114 cited
- 2021 57 cited
- 2022 35 cited
- 2015 117 cited
- 2020 245 cited
- 2019 24 cited
- 2024 6 cited
- 2023 12 cited
- 2016 76 cited
- 2015 11 cited
- 2020 7 cited
- 2009 17 cited
- 2013 77 cited
- 2008 51 cited
- 2021 82 cited
- 2016 86 cited
- 2011 13 cited
- 2018 628 cited
- 2018 199 cited
- 2018 56 cited
- 2014 75 cited
- 2025 9 cited
Citation counts from Google Scholar, October 2026. Full list on Google Scholar.
Writing
Mostly Thought Through
mostlythoughtthrough.comMy personal blog: longer-form thinking on research, perception, AI and whatever else is on my mind.