Neural mechanisms of object recognition

SupervisorsRamakrishna Chakravarthi, Søren Andersen

Project description:

Recognising objects is one of the central functions of the human visual system and is essential for adaptive behaviour. Hence, a fundamental goal of visual neuroscience is uncovering the neural and computational mechanisms underlying this ability. Towards this end, the project will use visual crowding, a phenomenon where recognisable objects are rendered unidentifiable by nearby clutter [1], as a tool. Crowding is strongly dependent on the spacing between objects and the current consensus argues for a spacing-dependent measure of crowding, the critical spacing. However, we recently discovered [2] that crowding is, instead, better characterised by critical density – the number of objects that the brain can resolve within a unit area of the visual field.

In this project, we propose to explore the exciting new window that this finding has opened up to determine the workings of object recognition. We will bring together three distinct approaches – psychophysics, neural recordings and computational modelling to develop a quantitative model of crowding and hence object recognition. This project requires a solid experimental base with strong mathematical, programming and data analytical abilities. It is therefore suitable to those with a background in experimental psychology, neuroscience, or those in the physical, mathematical or computational sciences looking to apply their skills in neuroscience. The successful candidate can expect to receive extensive training on techniques complementary to their skills.

Project: As a first step, we will reanalyse existing datasets and results in the literature to verify if critical density can characterise human crowding behaviour better than traditional approaches. Second, we will determine, using psychophysics, an optimal method to estimate critical density for a given set of objects. One example will be to ascertain the number of objects that can be identified within a specified region of visual space and thereby estimate the critical density. Next, we will use this optimal method and examine the dynamics of neural mechanisms that create the cortical bottleneck in object recognition. To do so, we will analyse Steady State Visual Evoked Potentials (SSVEPs), an Electroencephalographic (EEG) tool that can reveal the neural resources used by each individual object separately at a high temporal resolution. Finally, we plan to use these data to develop a computational model of crowding that takes into account critical density. We will implement a population coding model [3], which posits that the activity of ensembles of neurons, not single neurons, represent objects. This approach is well suited for testing the idea that the density of objects limits object recognition.

This set of studies is aimed at shedding light on one of the fundamental processes in the human brain. It also provides a conducive environment for the student to learn several cutting-edge research techniques. The results have the potential to also inform modern recognition algorithms used in face and object recognition in pictures and videos and products from cameras, self-driving cars to Artificial Intelligence.

[1] Levi DM (2008) Vis Res 48:635-54. [2] Soo L, et al. (submitted) Curr Bio. [3] Harrison WJ Bex PJ (2015) Curr Bio 25:1-7.