Research internship in neuromorphic computing

Université de Sherbrooke, Canada

Updated: over 2 years ago
Location: Government of Canada Ottawa and Gatineau offices, ONTARIO
Job Type: FullTime
Deadline: 01 Oct 2022

Design of a memristor-based spiking neural network for anomaly detection with a low-power vibration sensor

Context:

Artificial intelligence makes it possible to perform complex tasks such as object detection and speech recognition. However, the high energy consumption of artificial neural networks based on deep learning prevents their integration into low-power sensors. Spiking neural networks offer an elegant solution to this problem. These systems work analogously to a biological brain, where neurons exchange electrical impulses through synapses. Since information is encoded as pulses of finite duration and exchanged asynchronously, the energy consumed by the circuit is several orders of magnitude less than conventional electronics.

We offer a 6-month internship project whose goal is to design a simulation environment for spiking neural networks capable of continuously analyzing the signal generated by a vibration sensor to detect anomalies indicating the imminent failure of a device. In this simulation, synaptic connections will be made using a behavioral model of TiO2 resistive memories, also known as memristors. These memories are under development at 3IT and will eventually be used as artificial synapses in computer hardware specifically optimized for bio-inspired computation.

Objectives:

  • Perform a literature review on artificial intelligence based on spiking neural networks for low-power sensor applications.
  • Develop a procedure to detect anomalies in a time series with the help of a spiking neural network.
  • Implement the procedure identified in point 2 with a theoretical neural network whose synaptic weights will be based on a behavioral mathematical model of resistive memory already validated at the University of Sherbrooke.
  • Determine the influence of the behavior of resistive memories on the performance of the spiking neural network.
  • Demonstrate anomaly detection with a network trained on real vibration data obtained from an industrial partner and predict the energy consumption for constant monitoring.

    • Supervision and working environment:

    This project will be carried out under the co-supervision of Dr Alexandre Juneau-Fecteau, Pr Dominique Drouin, Pr Serge Ecoffey, Pr Yann Beilliard, and Pr Luc Fréchette. The work will take place at the Institut Interdisciplinaire d’Innovation Technologique (3IT) of the University of Sherbrooke, a Canadian institute dedicated to cutting-edge research in the fields of energy, electronics, robotics, and health. The candidate will thus benefit from a highly interdisciplinary research environment where students, engineers, professors, and industrialists work in close collaboration.


    Similar Positions