Learning invariance in visual perception.
-
1
Chemnitz University of Technology, Computer Science, Germany
The human visual system has the remarkable ability to largely recognize objects invariant of their position, rotation and scale. In part, this is likely achieved step by step from early to late areas of visual perception. But how could this invariance be achieved by the visual system? One key point is the stable world, which, however, is perceived through rapid variations of the image on the retina. Thus, the visual system has to learn to represent this rapid varying information as stable as the entity of the world is (cf. Berkes, 2005).
Obviously, exploiting the fact that the world is temporally coherent could be an appropriate neural learning strategy by making use of a trace in the learning rule (e.g. Földiak, 1991). We show that such a strategy could be implemented in a biologically plausible way and could be used to learn invariant representations from natural scene inputs. Furthermore, we show how one concept of neuronal learning can be applied to build a model for invariant object recognition being able to recognize and invariantly represent object properties of different complexities.
Therefore, we propose a learning rule based on the conceptual design of the previously developed learning rule for simple cells (Wiltschut and Hamker, 2009). The previous rule has been developed on the ideas of normalized Hebbian learning, covariance learning, and anti-Hebbian decorrelation. It leads to largely independent responses of V1-simple cells when trained on natural scenes. Our new rule expands these ideas by incorporating aspects of BCM learning and the calcium dependency of learning processes (Shouval et al., 2002), particularly by using the level of calcium rather than neural activity in the learning rules.
We have implemented this rule in a model learning the invariance properties of so-called V1-complex cells. It is important to generate appropriate image transformations simulating the rapidly varying retinal image. For this purpose, we use small shifts of the input simulating fixational eye movements. For the learning of invariance representations in higher visual areas it is required to simulate more complex and realistic object transformations.
The rule enables us to learn neurons sharing the invariance properties of V1-complex cells (Teichmann et al., 2012). We are able to show that the degree of temporal information used in the learning influences the development of invariant and temporally slowly fluctuating neuronal responses. Importantly, the derived learning rule is not special for this particular purposes: the same learning rule differing only in the length of the trace allows neurons to learn simple-cell like receptive fields.
References
Berkes, P., and Wiskott, L. (2005). Slow feature analysis yields a rich repertoire of complex cell properties. Journal of vision 5, 579-602.
Shouval, H.Z., Castellani, G.C., Blais, B.S., Yeung, L.C., and Cooper, L.N. (2002). Converging evidence for a simplified biophysical model of synaptic plasticity. Biological cybernetics 87, 383-91.
Teichmann, M., Wiltschut, J., and Hamker, F.H. (2012). Learning invariance from natural images inspired by observations in the primary visual cortex. Neural computation 24, 1271-96.
Wiltschut, J., and Hamker, F.H. (2009). Efficient coding correlates with spatial frequency tuning in a model of V1 receptive field organization. Visual neuroscience 26, 21-34.
Keywords:
complex cells,
computational model,
Hebbian Learning,
invariant representations,
natural scenes,
Simple cells,
trace learning,
Visual Perception
Conference:
Bernstein Conference 2012, Munich, Germany, 12 Sep - 14 Sep, 2012.
Presentation Type:
Poster
Topic:
Sensory processing and perception
Citation:
Teichmann
M and
Hamker
FH
(2012). Learning invariance in visual perception..
Front. Comput. Neurosci.
Conference Abstract:
Bernstein Conference 2012.
doi: 10.3389/conf.fncom.2012.55.00173
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
11 May 2012;
Published Online:
12 Sep 2012.
*
Correspondence:
Mr. Michael Teichmann, Chemnitz University of Technology, Computer Science, Chemnitz, 09111, Germany, michael.teichmann@informatik.tu-chemnitz.de