A high-capacity model for one shot association learning in the brain

• Main Authors: Hafsteinn eEinarsson, Johannes eLengler, Angelika eSteger
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2014-11-01
• Series: Frontiers in Computational Neuroscience
• Subjects: Association Learning; Hetero-associative Network; Memory capacity; One shot learning; Relation Learning; Bootstrap Percolation and Propagation
• Online Access: http://journal.frontiersin.org/Journal/10.3389/fncom.2014.00140/full
• ISSN: 1662-5188

We present a high-capacity model for one-shot association learning(hetero-associative memory) in sparse networks. We assume that basic patternsare pre-learned in networks and associations between two patterns are presentedonly once and have to be learned immediately. The model is a combination of anAmit-Fusi like network sparsely connected to a Willshaw type network. Thelearning procedure is palimpsest and comes from earlier work on one-shotpattern learning. However, in our setup we can enhance the capacity of thenetwork by iterative retrieval. This yields a model for sparse brain-likenetworks in which populations of a few thousand neurons are capable of learninghundreds of associations even if they are presented only once. The analysis ofthe model is based on a novel result by Janson et. al. on bootstrappercolation in random graphs.