Seeing the light: Why vision is mathematically impossible
To understand how we see we must first understand what we see. Colour – arguably our most basic visual sensation – suggests an answer to this question. Colour shows us that context is everything when it comes to our perceived truths … that our mental images are not accurate depictions of the world but useful behaviours towards a world. Colour shows us that the brain evolved to continually redefine normality by learning and re-learning the relationships within and between the patterns of light that hit the eye that enabled survival in the past. In this talk I’ll focus less on data and more on the framework that guides our research. In doing so I’ll describe what colour really is and why it’s in principle impossible to see, and yet our most simple sensory construct. Along the way we’ll hear the music of light, see the world as an insect might see it and watch in real-time the evolution of artificial life agents that survive by adapting to ‘eat’ colour. The subtext of the presentation is to use colour (and vision generally) as a way of understanding the computational principles by which biological systems generate robust behaviour towards a world they can never know.

Seeing the light: How the interplay of brain mechanisms and the statistics of natural scenes illuminates human vision
To understand how we see we must first understand why we see. We see so that we can act appropriately in complex visual environments. This places computational constraints on any models we build of human vision; i.e. on the output we would like the model to produce. Other constraints come from the data (the statistical structure of natural scenes) and the hardware (the available neural machinery). I will show how, when taken together, these three constraints can help us to understand complex visual problems like brightness perception: our singular ability to determine the reflectance/lightness of surfaces under large variations in illumination. This is a tough/under-constrained problem since a surface will return more light depending both on its lightness and how intensely it is illuminated; i.e. we infer two pieces of information from one. To do it the visual system makes assumptions about the world. I’ll show how a knowledge of natural scene statistics, and of the physiology of the human visual system, leads to a simple computational model of brightness perception that accords with (a) behaviour, (b) patterns of brain-activation and © our experience of a wide-range of powerful visual illusions.