The Secret Life of Kernels: Reconsolidation in Flexible memories

Abstract

Event Abstract Back to Event The Secret Life of Kernels: Reconsolidation in Flexible memories In this paper we are presenting memory reconsolidation in kernel associative memories. Reconsolidation is an important process in memory dynamics that is observed both in neurophysiological and psychological studies, as well as modelled in various artificial neural systems. The memory tracks changes in the environment or in associations among objects. Recent models predict that memory representations should be sensitive to learning order, consistent with psychophysical studies of face recognition and electrophysiological experiments on hippocampal place cells. Our goal is to show that such reconsolidation effects are possible in more flexible environment, dealing with large-scale data. As memory model we introduce a special neural system that while relying on some of the analysis developed by Hopfield has memory attractors that do not lie in the input or neural space but rather in an abstract unbounded high-level kernel space. For reconsolidation we choose the principle of global memory update. It is more stable than updating the attractor closest to current input, and also this technique enables direct analogy to existing reconsolidation methods in classical Hopfield networks. We establish a metric in between kernel associative memories. In the feature space this metric is equivalent to the distance between the weight matrices of two networks, but in input space it is a Riemannian metric. Subsequent application of this procedure implies dynamic reconsolidation of memories that stay always consistent to the changing environment. Basing on this metric we construct an elementary update procedure - one step of reconsolidation. Memory is shifted in the direction of new input along a geodesic curve in corresponding Riemannian space. Our first experiment was made using the MNIST database of handwritten digits. We examined network's ability to track images gradually varying in time. For this purpose a learning set of rotating digits was created. The learning set contained 9000 images. They were obtained from 100 original digits (10 per class) by rotating them counterclockwise on angle from 0 to 180o.Classification was tested on the set of 1000 images closest to the final state. Obtained recognition quality was 96.4%. The second experiment is similar to one previously made in humans. Attractor tracking was investigated using sequences of morphed faces from Productive Aging Lab's Face Database. When the learning order follows image order in the morphing sequence, attractors changed gradually and consistently. The ability to recognize the initial set of images gradually decreased when attractors tended to the final set. In case of random learning order attractors quickly got messy, with no significant ability to distinguish faces. This experiment also demonstrates efficiency of reconsolidation in kernel memories for high-dimensional data. Based on these results we conclude that reconsolidation in kernel memories is both computationally efficient and biologically plausible, and it can model phenomena previously observed in human memory. Conference: Computational and systems neuroscience 2009, Salt Lake City, UT, United States, 26 Feb - 3 Mar, 2009. Presentation Type: Poster Presentation Topic: Poster Presentations Citation: (2009). The Secret Life of Kernels: Reconsolidation in Flexible memories. Front. Syst. Neurosci. Conference Abstract: Computational and systems neuroscience 2009. doi: 10.3389/conf.neuro.06.2009.03.271 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: 04 Feb 2009; Published Online: 04 Feb 2009. Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract Supplemental Data The Authors in Frontiers Google Google Scholar PubMed Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.