Katie Davey

Biomedical Engineering Department

Dr. Katie (Catherine) Davey is a Lecturer in the Biomedical Engineering Department at the University of Melbourne. Katie’s primary research areas are in functional MRI – the acquisition and analysis of a series of low resolution magnetic resonance images to understand brain function – and spike timing dependent synaptic plasticity (STDP) – the process by which connected neurons adapt connection strengths during learning. Katie’s research uses advanced signal processing methods, in conjunction with simulation and modelling techniques, to mathematically and programmatically model cortical processes and gain insight into how we perceive and process sensory information. Katie has additional collaborations, such as with Imperial College of London to investigate the encoding and storage of location information by place cells using calcium imaging, and with Florey Institute of Neurosciences in modelling the neural pathways for bowel disease. Katie completed her doctoral research in functional MRI connectivity, which is a field of research that analyses a series of low resolution MRI images to identify how brain regions cooperate to achieve sensory and perception tasks. After completing her Ph.D. Katie worked at the Defence Science Technology Organisation, modelling pilot cognition and aircraft control. She then worked in finance, modelling and predicting the movement of stock prices on the S&P500.

Contact details

cedavey@unimelb.edu.au

+61 3 8344 1899

Office G30

203 Bouverie St

University of Melbourne

Parkville, Victoria 3013

Australia

Plasticity in the visual cortex

Catherine E. Davey, Errol K.J. Lloyd, Levin Kuhlmann, Anthony N. Burkitt and Trichur R. Vidyasagar

Figure: Diagram of a three layered feed forward network, with neurons u, and v in the retina connecting to neurons n, and m, respectively, in layer LGN, while both neurons n and m in layer LGN connect to neuron j in layer V1. Each postsynaptic neuron has been colour coded to aid interpretation, such that neuron n has been coloured red, neuron m, blue, and neuron j, green. The synaptic connection density of a postsynaptic neuron determines the probability of a neuron in the presynaptic layer connecting to it, and is depicted by the density of that neuron’s colour in the presynaptic layer. Connection density is modelled using a Gaussian, with parameters constant across a layer, so that σRL denotes the radius, or standard deviation, of connections between the retina and the LGN, while σLV1 denotes the radius of connections between layers LGN and V1. The first and second standard deviations of the Gaussian connection probability for a postsynaptic neuron are represented by coloured, dashed concentric circles in the presynaptic layer of the neuron. The strength of a synaptic connection between two neurons, for example between neurons m, and n, is given by wLV1 . Not shown in the diagram are, NRL, and NLV1 , which mn represent the expected number of synaptic connections to each postsynaptic neuron in layers LGN, and V1, respectively.mechanistic and developmental problems, is based upon the fact that the sub-cortical stages of the visual pathway – the retina and LGN – have their own orientation biases, albeit with a lesser selectivity (Levick and Thibos, 1980; Vidyasagar and Urbas, 1982; Schall et al., 1986a; Shou and Leventhal, 1989). A simple sharpening of such broad biases during development could in principle lead to the typical sharp tuning for orientation seen in V1 cell responses. Here we are concerned not only with simulating the mechanistic solution to orientation selectivity, but also with how such selectivity emerges during development. We show how even experience-independent orientation selectivity development can occur if LGN inputs to V1 that share an orientation bias for the same or similar orientation correlate both in their spontaneous and visually evoked activities. Anatomical data shows that RGC projections to the LGN are both dominant, in that a single LGN cell is driven almost entirely by a single RGC (Lee et al., 1977) (but see Alonso et al. (2001)) and divergent, in that a single RGC projects to multiple LGN cells (Friedlander et al., 1981; Alonso et al., 2006; Dacey et al., 2003; Yeh et al., 2009). As the bias originates in the RGC cell, the scheme ensures correlation among similar V1 inputs. We also model how random spontaneous retinal activity and dominant and divergent retinal to LGN projections can lead to orientation selectivity in V1 through a well characterised and established Hebbian plasticity rule (Linsker, 1986a,b,c).

The neural basis of the selectivity for the orientation of a visual stimulus seen in the responses of V1 cells in mammals (Hubel and Wiesel, 1962, 1968) is still uner much debate (Vidyasagar and Eysel, 2015; Priebe, 2016; Sedigh-Sarvestani et al., 2017). In the earliest proposal (Hubel and Wiesel, 1962, 1968), a V1 neurons multiple inputs from the lateral geniculate nucleus (LGN) bear receptive fields (RFs) that are spatially dispersed along an axis, such that maximal input is generated by edges aligned along the same axis. However, such feedforward spatial convergence, fails to account for a number of behaviours of V1 neurons, leading to theories that revolve around what roles inhibition, intracortical horizontal networks or pre-existing sub-cortical selectivity may play (Vidyasagar and Eysel, 2015; Priebe, 2016). Experimental data show a seriously limited, if not absent, dispersion along the axis of the RFs (Creutzfeldt et al., 1974; Pei et al., 1994; Jin et al., 2008; Kremkow ;et al., 2016). A long standing alternative model, which posits congruent solutions to both the mechanistic and developmental problems, is based upon the fact that the sub-cortical stages of the visual pathway – the retina and LGN – have their own orientation biases, albeit with a lesser selectivity (Levick and Thibos, 1980; Vidyasagar and Urbas, 1982; Schall et al., 1986a; Shou and Leventhal, 1989). A simple sharpening of such broad biases during development could in principle lead to the typical sharp tuning for orientation seen in V1 cell responses. Here we are concerned not only with simulating the mechanistic solution to orientation selectivity, but also with how such selectivity emerges during development. We show how even experience-independent orientation selectivity development can occur if LGN inputs to V1 that share an orientation bias for the same or similar orientation correlate both in their spontaneous and visually evoked activities. University of Melbourne. Katie’s primary research areas are in functional MRI – the acquisition and analysis of a series of low resolution magnetic resonance images to understand brain function – and spike timing dependent synaptic plasticity (STDP) – the process by which connected neurons adapt connection strengths during learning. Katie’s research uses advanced signal processing methods, in conjunction with simulation and modelling techniques, to mathematically and programmatically model cortical processes and gain insight into how we perceive and process sensory information. Katie has additional collaborations, Anatomical data shows that RGC projections to the LGN are both dominant, in that a single LGN cell is driven almost entirely by a single RGC (Lee et al., 1977) (but see Alonso et al. (2001)) and divergent, in that a single RGC projects to multiple LGN cells (Friedlander et al., 1981; Alonso et al., 2006; Dacey et al., 2003; Yeh et al., 2009). As the bias originates in the RGC cell, the scheme ensures correlation among similar V1 inputs. We also model how random spontaneous retinal activity and dominant and divergent retinal to LGN projections can lead to orientation selectivity in V1 through a well characterised and established Hebbian plasticity rule (Linsker, 1986a,b,c).