Keynote Address on
"Disruptive Perspectives on
Biological and Machine Vision"

Tom Dean
Research Scientist, Google

We are machines that turn percepts into meaningful representations which are grounded in concrete tasks that facilitate our survival and reproductive success.  For humans, meaning flows from acting purposively, and, in the absence of a specific task, many computer vision problems are ill-posed. Contrary to intuition, perception is a constructive process that draws heavily upon memory and our ability to act in the world. Richard Gregory, a prominent British neuropsychologist, estimates that visual perception derives more than ninety percent of its constructed representations from memory and less than ten percent from the external stimuli impinging on our retina.

This talk explores perspectives on biological and computer vision with an emphasis on the role of large amounts of data and computation.  When you concentrate on concrete, visually-grounded applications, the availability and ability to handle very large collections of images encourages you to think about visual tasks in terms of memory.  We consider how new approaches to parallel computing and large-scale data storage are providing the foundation for a new generation of vision algorithms that borrow insights from biology and offer new paradigms for thinking about biological vision.