Research

Visual neuroscience

For years, the main focus of my research was in visual neuroscience. I studied in how we recognise things visually. Surely we must know all that already?! No, not by a long way.

We know a lot about the eye and we do a reasonable job of making devices (digital cameras) that do a roughly similar job. But imagine a piece of computer software that took your digital photographs and identified what was happening in them. “This is a photo of your father, sitting on a wooden chair with a carving of a rose on it. Next to him is a Yorkshire terrier chewing a bone.”

For you it’s trivial because of your visual neurons, but imagine trying to tell a computer what rules it should follow to understand the contents of a photograph. Imagine the range of lighting changes and hairstyles that a person can have without you ever failing to recognise that person. Despite the advances of computer AI your visual recognition is so much better than any automated computer detection system, which is why computer "Captcha" challenges to test if you're a robot still largely rely on visual recognition challenges.

We need to know how the brain does such a remarkable job of perceiving the world. That is the endeavour of visual neuroscience.

Research Methods (meta-science)

I wrote a package called PsychoPy to help my own research efforts and as that got popular my interests in how to improve research methods, as an area of research, grew. My patricular interests in the area are as follows

I care about precision. Software to measure human performance, for psychology, neuroscience or linguistics, needs to present stimuli and measure responses precisely. I spend a lot of time trying to understand where performance may be poor and also where that matters (there are also times when people worry about levels of precision they don't need).

I care about reproducibility. Scientists should be able to reproduce their studies and analyses but most, surprisingly, couldn't. It's all too easy to forget some detail of how you ran the study, or analysed the data. "How exactly did I calculate the position of the stimulus?", "Was this version of the graph before or after I found that better way to normalise the baseline?". I believe strongly in reproducible experiment scripts and data pipelines where all the details are kept for re-use.

I care about making things easy. I think a lot of errors occur in science because tools are too hard and dont automatically check for errors. I believe easier, more informative, tools make for better science.

I care about sharing. Science is better if we share (isn't everything?!). Partly that makes studies more reproducible but it also accelerates the rate at which we can develop/run studies. I try to encourage this by a) making it easy to share studies using the tools I develop and b) by setting a good example by sharing my own materials (i.e. open-sourcing tools) as much as possible.