Feature Binding within a Spiking Neural Network Model of the Primate Ventral Visual Pathway
Dr Simon Stringer (Director, Theoretical Neuroscience and Artificial Intelligence Laboratory, Department of Experimental Psychology, University of Oxford)
Senior Common Room, Level 2 (2D17 ), Priory Road Complex
Abstract: The feature binding problem refers to the ability of the visual brain to not only represent a hierarchy of features such as edges and objects but also to represent the relationships between these features. This is necessary for the brain to be able to interpret and make sense of complex natural scenes. Solving this problem is an important step towards the development of artificial general intelligence and machine consciousness. We have found a new approach to solving feature binding by incorporating key characteristics of the cortex into biologically realistic neural network simulations. In particular, our simulations implement the following: (i) the cortex has bottom-up, lateral and top-down synaptic connections, (ii) real neurons in the brain communicate by emitting electrical pulses called action potentials or ‘spikes’, (iii) synaptic plasticity depends on the relative timings of spikes emitted by pre- and post-synaptic neurons, and (iv) action potentials take several milliseconds to pass between neurons. When these biological characteristics are incorporated into computer simulations we observe the emergence of rich spatiotemporal patterns of spiking activity in the higher layers – a phenomenon known as polychronization. Such a subpopulation of neurons is referred to as a polychronous neuronal group (PNG). We have found that PNGs emerge robustly within the model even when neurons in the input layer represent visual stimuli with spikes that are randomized in time. Most importantly, within these spatiotemporal spike patterns are embedded binding neurons that encode the hierarchical relationships between lower and higher level visual features. The resulting hierarchical representation of visual scenes, including the representation of hierarchical binding relations between lower and higher level visual features, is consistent with the hierarchical phenomenology of primate vision.
All Welcome. Tea, coffee and biscuits will be available after the seminar.
For more information see https://mindandmachine.blogs.bristol.ac.uk/seminars/.