Novel deep-network architectures for studying primary visual cortex.

• Thesis title in Czech: Novel deep-network architectures for studying primary visual cortex.
• Thesis title in English: Novel deep-network architectures for studying primary visual cortex.
• Key words: cortex; vision; neuroscience; machine learning; convolutional neural networks; deep learning
• Academic year of topic announcement: 2018/2019
• Thesis type: diploma thesis
• Department: Department of Software and Computer Science Education (32-KSVI)
• Supervisor: Mgr. Ján Antolík, Ph.D.
• Author: hidden - assigned and confirmed by the Study Dept.
• Date of registration: 14.01.2019
• Date of assignment: 14.01.2019
• Confirmed by Study dept. on: 28.01.2019

Guidelines

A common approach for studying the function of early sensory systems is to determine the relationship between sensory inputs and associated (experimentally recorded) neural responses. In the past, mostly linear [1], or shallow non-linear techniques were utilized, leading to limited predictive and consequently explanatory power of models fitted in this way. More recently, the popular deep convolutional architectures were successfully tested on the neural data [2,3]. These general, machine-learning motivated models, however, ignore the known anatomical and functional architecture of visual system. Recently, we have presented a multi-stage model of primary visual cortex which reflected some of the most prominent features of the early visual system [4], and demonstrated that such incorporation of cortical biology can improve performance in comparison to state-of-the-art models. In this project student will build upon these early results, and develop novel deep-architectures inspired by the deep convolutional networks, but enriched by biologically inspired elements. The student will be responsible for designing, implementing and subsequently testing the new models on neural population recordings from cat primary visual cortex. This project will be undertaken in collaboration with experimental lab of Yves Fregnac, CNRS, France, and computational lab of Dan Butts, University of Maryland. Prior experience in machine learning is recommended.

References

[1] Willmore, B. D. B., & Smyth, D. (2003). Methods for first-order kernel estimation: simple-cell receptive fields from responses to natural scenes. Network, 14(3), 553–77. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/12938771
[2] Vintch, B., Movshon, J. A., & Simoncelli, E. P. (2015). A Convolutional Subunit Model for Neuronal Responses in Macaque V1. Journal of Neuroscience, 35(44), 14829–14841. https://doi.org/10.1523/JNEUROSCI.2815-13.2015
[3] Cadena, S. A., Denfield, G. H., Walker, E. Y., Gatys, L. A., Tolias, A. S., Bethge, M., & Ecker, A. S. (2017). Deep convolutional models improve predictions of macaque V1 responses to natural images. bioRxiv, 201764. https://doi.org/10.1101/201764
[4] Antolík, J., Hofer, S. B., Bednar, J. A., & Mrsic-Flogel, T. D. (2016). Model Constrained by Visual Hierarchy Improves Prediction of Neural Responses to Natural Scenes. PLoS Computational Biology, 12(6). https://doi.org/10.1371/journal.pcbi.1004927