Adaptive sensing and computing towards always-on context-awareness

This video program is a part of the Premium package:

Adaptive sensing and computing towards always-on context-awareness Video


  • IEEE MemberUS $11.00
  • Society MemberUS $0.00
  • IEEE Student MemberUS $11.00
  • Non-IEEE MemberUS $15.00
Purchase

Adaptive sensing and computing towards always-on context-awareness

7 views
  • Share
Create Account or Sign In to post comments
Future mobile electronic devices will be equipped with more and more sensors that require always-on operation, to bring continuous context-awareness to the mobile device. Enabling this at near-zero power budgets, is only feasible if the device can continuously tune its own performance and hardware configuration in function of the sensory context. This requires combining research on adaptive sensor interfaces, embedded machine learning, and reconfigurable computing. We propose two important self-adaptivity techniques which can be exploited both in the sensor interfaces, as well as in the subsequent machine learning processing layer: 1.) hierarchical activation, and 2.) precision scalability. Both techniques will be illustrated with practical silicon implementations to assess their benefits, and this for always-on acoustic sensing and always-on image recognition applications. The resulting hardware context-awareness will be crucial in achieving the necessary 10x energy improvement for further miniaturization of mobiles, wearables and the IoT.
Future mobile electronic devices will be equipped with more and more sensors that require always-on operation, to bring continuous context-awareness to the mobile device. Enabling this at near-zero power budgets, is only feasible if the device can continuously tune its own performance and hardware configuration in function of the sensory context. This requires combining research on adaptive sensor interfaces, embedded machine learning, and reconfigurable computing. We propose two important self-adaptivity techniques which can be exploited both in the sensor interfaces, as well as in the subsequent machine learning processing layer: 1.) hierarchical activation, and 2.) precision scalability. Both techniques will be illustrated with practical silicon implementations to assess their benefits, and this for always-on acoustic sensing and always-on image recognition applications. The resulting hardware context-awareness will be crucial in achieving the necessary 10x energy improvement for further miniaturization of mobiles, wearables and the IoT.