Using inspiration from the brain, our goal is to develop novel and unique neuromorphic algorithms, sensors and processing hardware, to create machines which are able to continuously and autonomously learn about the real world, and use this knowledge for predictive decision making.
Our systems will seamlessly integrate learning in deep neural networks with inference over distributed graph-based knowledge representations to facilitate continuous "one-shot" learning and inference and the transfer of learnt knowledge to new domains.
Our neuromorphic learning and inference algorithms are designed to be executed on massively parallel, many core distributed computer systems which use discrete event-based, asynchronous, sparse computation for speed, efficiency and low energy.
To achieve this, our neuromorphic algorithms have the following main characteristics:
