Perceptive Invariance and Associative Memory Between Perception and Semantic Representation USER a Universal SEmantic Representation Implemented in a System on Chip (SoC)

USER (Universal SEmantic Representation) is a bio-inspired module implemented in a system on a chip (SoC), which builds a link between multichannel perception and semantic representation. The input data are projected into a generic bioinspired higher dimensional non-linear semantic space with high sparsity. A pooling of these semantic representations (global, dynamic and structural) is done automatically by a set of dynamic attractors embedding spatio-temporal histograms, being drastically more efficient than back-propagation. A supervised learning is used to build the association between the invariant multimodal semantic representations (histogram results) and the labels (‘words’). The invariant recognition is achieve thanks to multichannel multiscale dynamic attractors and bilinear representations - imitating brain attentional processes. USER modules can be cascaded, allowing to work at different levels of abstraction (or complexity). Due to its low consumption, small size and minimal price, USER targets deep learning, robotics, and Internet of Things (IoT) applications.