Learning and Adaptation in the Primary Visual Cortex

This project aims to use computational neuroscience to understand how the primary visual cortex (V1) adapts to the changing visual statistics of its environment, and why. It is within the EPSRC's Biological Informatics research area.

In the mammalian visual cortex, certain cells are "orientation selective", meaning that (within the region of the visual field they are sensitive to), they respond optimally to alternating black and white stripes at a particular orientation (Hubel and Wiesel 1959). When observing natural scenes, the orientations are represented uniformly by V1; however, when the frequency of a particular orientation is increased, the representation is adjusted to maintain a homeostasis in firing rates (Benucci et al. 2013). This adaptation keeps both first and second order firing statistics approximately constant, and therefore cannot be explained by a mechanism at the level of individual neurons.

Recent progress in this area has come in the form of a descriptive model of this cortical activity (Westrick et al. 2016). This model aims to understand the adjustment in terms of normalisation, a proposed canonical computation performed by V1 (Carandini and Heeger 2011). However, there is not yet a mechanistic level explanation for these dynamics. It has previously been demonstrated that the Stabilized Supralinear Network (SSN) model is able to perform normalisation computations (Ahmadian et al. 2013). We will therefore investigate the possibility that by supplementing the SSN with appropriate plasticity rules we can formulate a mechanistic model of adaptation. We also aim to understand what (if any) computational motivation there is for this adaptation, using recent theories of efficient representation (Ganguli and Simoncelli 2014).

References:
[1] D. H. Hubel and T. N. Wiesel, 'Receptive fields of single neurones in the cat's striate cortex', J Physiol, vol. 148, no. 3, pp. 574-591, Oct. 1959.
[2] A. Benucci, A. B. Saleem, and M. Carandini, 'Adaptation maintains population homeostasis in primary visual cortex', Nat Neurosci, vol. 16, no. 6, pp. 724-729, Jun. 2013, doi: 10.1038/nn.3382.
[3] Z. M. Westrick, D. J. Heeger, and M. S. Landy, 'Pattern Adaptation and Normalization Reweighting', Journal of Neuroscience, vol. 36, no. 38, pp. 9805-9816, Sep. 2016, doi: 10.1523/JNEUROSCI.1067-16.2016.
[4] M. Carandini and D. J. Heeger, 'Normalization as a canonical neural computation', Nat Rev Neurosci, vol. 13, no. 1, pp. 51-62, Jan. 2012, doi: 10.1038/nrn3136.
[5] Y. Ahmadian, D. B. Rubin, and K. D. Miller, 'Analysis of the stabilized supralinear network', Neural Comput, vol. 25, no. 8, pp. 1994-2037, Aug. 2013, doi: 10.1162/NECO_a_00472.
[6] D. Ganguli and E. P. Simoncelli, 'Efficient Sensory Encoding and Bayesian Inference with Heterogeneous Neural Populations', Neural Computation, vol. 26, no. 10, pp. 2103-2134, Oct. 2014, doi: 10.1162/NECO_a_00638.