Power-efficient Autonomous AI

With breakthroughs in AI based on deep learning in deep neural networks, applications are bountiful. For many applications however, AI has to be ‘always on’ and, when applied in autonomous settings, energy-efficiency is of paramount concern.

With breakthroughs in AI based on deep learning in deep neural networks, applications are bountiful. For many applications however, AI has to be ‘always on’ and, when applied in autonomous settings, energy-efficiency is of paramount concern.

Spiking neural networks approximate the sparse and power-efficient computation that biological brains achieve. Applications range from always on AI on cell phones to highly power-efficient intelligence in drones. Future autonomous automotive solutions are likely to similarly benefit from power-efficient deep neural networks.

The research of our Machine Learning group is delivering novel deep-spiking neural networks that compute deep neural networks with very few computational and power resources, matching performance with standard deep neural networks for vision, speech recognition and sequential prediction tasks. Applications may be of interest to companies as diverse as Google, Qualcomm, NVIDIA, automotive manufacturers and the developers of cell phone apps or drone software.

Contact person: Sander Bohte
Research group: Machine Learning (ML)
Research partners: UvA, TU Delft