positions

Task: Sparse on-device training, (WP1)

Host institution: TU Graz

Country: Austria

Supervisor: Prof. O. Saukh [TU Graz]

Co-supervisors: Prof. J. Champati [IMDEA]; Dr. P. Priller [AVL]

Objectives: 1) To explore different strategies of connection sampling and connection ranking to keep the cost of connection update low; 2) To enable training for quantised low bitwidth networks; 3) To propose new sparse training methods and optimisation techniques, e.g., for forward-forward training methods in the context of on-device learning and in combination with sparsity.

Expected Results: 1) Efficient connection sampling methods for dynamic sparse training; 2) A framework for training quantised low bitwidth networks; 3) New methods for sparse on-device training.

PhD enrolment: Doctoral School of TU Graz

Planned secondments: 

• AVL (4 months, M16-M19): Sparse on-device training implemented on AVL platforms, with Dr. P. Priller (KPI: joint paper)

• IMDEA (4 months, M28-M31): Novel sparse on-device training methods designed for low-power smart farming sensors, with Prof. J. Champati (KPI: joint paper)

Candidate profile: computer science, computer engineering, applied mathematics

Desirable skills/interests: machine learning, optimization, tinyML, embedded systems/AI (the applicant should be proficient in at least two of the skills)

Application Deadline: February 14, 2025, AoE

Descriptions of all the 18 DC Positions

Submit Your Application HERE