Event-based LEarning, Vision, And conTrol for Embodied systems (ELEVATE)

The research of the ELEVATE network aims at impacting the efficiency of machine intelligence by acknowledging and leveraging the event-based nature of animal intelligence. Specifically, we will address three fundamental limitations of current digital technology:

(i) Digital technology is densely pixelated and clocked, meaning that any task that requires to integrate data over long temporal and spatial scales relative to the resolution of the (spatial) pixel and (temporal) clock leads to prohibitive storage and size requirements;

(ii) Computational models are trained offline and centrally, far away from the machines that use them. Timely adaptation to changing environments necessitates prohibitive data transfer and communication requirements, prone to security and privacy issues that themselves generate additional algorithmic requirements of increasing size and complexity;

(iii) Processing information with clocks rather than with events prevents many rhythmic features of animal intelligence from being translated to machine intelligence. As a consequence, scientific progress in neuroscience has become increasingly difficult to translate into technological progress in machine intelligence.