Abstract
Sparse coding (SC) is a standard approach to relate neural response properties of primary visual cortex (V1) to the statistical properties of images. SC models the dependency between latent and observed variables using one weight matrix that contains the latents’ generative fields (GFs). Here, we present a novel SC model that couples latent and observed variables using two matrices: one matrix for component means and another for component variances. When training on natural image patches, we observe Gabor-like and globular GFs. Additionally, we obtain a second dictionary for each component’s variances. The double-dictionary model is thus the first to capture first- and second-order statistics of natural image patches using a multiple-causes latent variable model. If response probabilities of V1 simple cells are not restricted to first order statistics, the investigated model is likely to be more closely aligned with neural responses than standard SCs or independent component analysis (ICA) models.
S.H. Mousavi and J. Drefs—Authors contributed equally.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
References
Bornschein, J., Henniges, M., Lücke, J.: Are V1 simple cells optimized for visual occlusions? A comparative study. PLoS Comput. Biol. 9(6), e1003062 (2013)
Guiraud, E., Drefs, J., Lücke, J.: Evolutionary expectation maximization. In: GECCO. ACM (2018)
van Hateren, J.H., van der Schaaf, A.: Independent component filters of natural images compared with simple cells in primary visual cortex. Proc. R. Soc. Lond. 265, 359–366 (1998)
Hyvärinen, A., Oja, E.: A fast fixed-point algorithm for independent component analysis. Neural Comput. 9(7), 1483–1492 (1997)
Karklin, Y., Lewicki, M.S.: Learning higher-order structures in natural images. Netw.: Comput. Neural Syst. 14, 483–499 (2003)
Karklin, Y., Lewicki, M.S.: Emergence of complex cell properties by learning to generalize in natural scenes. Nature 457, 83–86 (2009)
Lücke, J., Sahani, M.: Maximal causes for non-linear component extraction. JMLR 9, 1227–1267 (2008)
Mousavi, S.H., Drefs, J., Hirschberger, F., Lücke, J.: Maximal causes for exponential family observables. arXiv preprint arXiv:2003.02214 (2020)
Olshausen, B.A., Field, D.J.: Sparse coding with an overcomplete basis set: a strategy employed by V1? Vis. Res. 37, 3311–3325 (1997)
Puertas, J., Bornschein, J., Lücke, J.: The maximal causes of natural scenes are edge filters. In: Advances in Neural Information Processing Systems (2010)
Sheikh, A.S., et al.: STRFs in primary auditory cortex emerge from masking-based statistics of natural sounds. PLoS Comput. Biol. 15(1), e1006595 (2019)
Shivkumar, S., Lange, R., Chattoraj, A., Haefner, R.: A probabilistic population code based on neural samples. Adv. Neural Inf. Process. Syst. 31, 7070–7079 (2018)
Wainwright, M.J., Simoncelli, E.: Scale mixtures of Gaussians and the statistics of natural images. Adv. Neural Inf. Process. Syst. 12, 855–861 (2000)
Zoran, D., Weiss, Y.: From learning models of natural image patches to whole image restoration. In: ICCV (2011)
Acknowledgments
We acknowledge funding by the German Research Foundation (DFG) in project 390895286 (cluster of excellence H4a 2.0, EXC 2177/1) and the German Ministry of Research and Education (BMBF) project 05M2020 (SPAplus, TP 3).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Mousavi, S.H., Drefs, J., Lücke, J. (2020). A Double-Dictionary Approach Learns Component Means and Variances for V1 Encoding. In: Nicosia, G., et al. Machine Learning, Optimization, and Data Science. LOD 2020. Lecture Notes in Computer Science(), vol 12566. Springer, Cham. https://doi.org/10.1007/978-3-030-64580-9_20
Download citation
DOI: https://doi.org/10.1007/978-3-030-64580-9_20
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-64579-3
Online ISBN: 978-3-030-64580-9
eBook Packages: Computer ScienceComputer Science (R0)