DC7: Light-weight JCAS analytics for resource-constrained embedded devices
Task: Light-weight JCAS analytics for resource-constrained embedded devices (WP3)
Host institution: TU Delft
Country: The Netherlands
Supervisors: Prof. Q. Wang [TU Delft]
Co-supervisors: Prof. J. Yang [TU Delft], Dr. A. Mueller [Bosch]
Objectives: To design self-optimising algorithms for allocation of on-board computational and communication resources for JCAS; To employ knowledge distillation techniques to create performant and low-complexity models for embedded devices; To implement and evaluate the proposed solutions using facilities of the embedded-ML lab at TUDelft.
Expected Results: Learning algorithms that can adapt the allocation of resources on embedded devices for different JCAS requirements; Learning algorithms that are significantly compressed but incur negligible performance loss.
PhD enrolment: Doctoral School of TU Delft
Planned secondments:
-
Bosch (3 months, M10-M12): Design of low-complexity ML algorithm for sensor fusion, with A. Mueller (KPI: joint conference paper)
-
UNITN (3 months, M18-M20): Model compression for high-speed object detection tasks, with P. Casari (KPI: joint conference paper)
Candidate profile: Telecommunications engineering, Embedded systems, Computer engineering, or equivalent disciplines
Desirable skills/interests: Deep learning, tiny machine learning (TinyML), TensorFlow Lite micro, embedded systems, wireless communications (the applicant should be proficient in at least two of the skills)