Project Main Goals
Biological information processing systems (brains) overwhelmingly outperform their artificial counterparts. An engineering understanding of these computational principles of biology would permit the construction of a new generation of artificial systems for such tasks as vehicle navigation, prosthetics, and personal information systems. This knowledge would have enormous economic significance.
Biological brains are structured in layers of neurons, where neurons in a layer connect to a very large number (~104) of neurons in the following layer. The pattern of connectivity seems to follow a basic structure: each neuron in a layer connects to a “cluster of neurons” or “projective field” in the next layer. Mathematically, in most cases this can be approximated by computing two dimensional convolutions. Artificial bio-inspired software models have overwhelmed the specialized literature presenting many ways to perform bio-like (convolution-based) processing systems that outperform more conventionally engineered machines, although at extremely low speeds. For Real-Time solutions direct hardware implementations are required. But hardware engineers face a very strong barrier when trying to mimic the bio-inspired layered structure: the massive connectivity. In present day state-of-the-art integrated circuit technologies it is plausible to fabricate on a single chip many thousands (even millions) of artificial neurons (processing cells). However, it is not viable to connect physically each of them to even a few hundreds of other neurons. The problem gets worst for multi-chip multi-layer hierarchical bio-like systems. ‘Address-Event-Representation’ (AER) is an incipient bio-inspired spike-based technique capable of providing a hardware solution to the inter-chip massive connectivity problem.
The over-reaching objectives of CAVIAR are two-fold:
To develop a general AER infrastructure for constructing bio-inspired hierarchically
systems for sensing + processing + actuation (see Fig. 1).
• The implementation of a particular perceptive-action demonstrator vision system exploiting this infrastructure.
Fig. 1: A bio-inspired engineered autonomous system performing sensing+processing+actuation tends to have the following conceptual hierarchical structure: 1) A sensing layer, like an artifical retina (for vision) or cochlea (for audition), or any other artificial ‘sense’, or even a combination of them; 2) A set of low-level processing layers usually implementing proyection fields (similar to convolutions); 3) A set of high level processing layers that operate on extracted ‘abstractions’ and progressively concentrate information through (for example) dimension reduction, competition, and learning; 4) Once a very reduced set of signals (or decisions) is obtained they are conveyed to (usually mechanical) actuators.