Assembly pointers for variable binding in networks of spiking neurons

R. Legenstein, C. H. Papadimitriou, S. Vempala, and W. Maass

Abstract:

We propose a model for binding of variables such as the thematic role of a word in a sentence or episode (e.g., agent or patient), to concrete fillers (e.g., a word or concept). Our model is based on recent experimental data about corresponding processes in the human brain. One source of information are electrode recordings from the human brain, which suggest that concepts are represented in the medial temporal lobe (MTL) through sparse sets of neurons (assemblies). Another source of information are fMRI recordings from the human brain, which suggests that subregions of the temporal cortex are dedicated to the representation of specific roles (e.g.,subject or object) of concepts in a sentence or visually presented episode. We propose that quickly recruited assemblies of neurons in these subregions act as pointers to previously created assemblies that represent concepts. We provide a proof of principle that the resulting model for binding through assembly pointers can be implemented in networks of spiking neurons, and supports basic operations of brain computations, such as structured information retrieval and copying of information. We also show that salient featrues of fMRI data on neural activity during structured information retrieval can be reproduced by the proposed model.

Reference: R. Legenstein, C. H. Papadimitriou, S. Vempala, and W. Maass. Assembly pointers for variable binding in networks of spiking neurons. arXiv preprint arXiv:1611.03698, 2016.