| KEYWORD |
Area Engineering
Neural Architecture Search for Spiking Neural Networks in PPG-Based Blood Pressure Estimation.
Thesis abroad
keywords ARTIFICIAL INTELLIGENCE, DEEP LEARNING, DEEP NEURAL NETWORKS, EMBEDDED SYSTEMS, ENERGY EFFICIENCY, LOW POWER, NEUROMORPHIC COMPUTING, SNNTORCH, SOFTWARE, SPIKING NEURAL NETWORKS
Reference persons ALESSIO BURRELLO, DANIELE JAHIER PAGLIARI
External reference persons Dr. Elisa Donati (INI Institute of Neuromorphic, Zurich)
Research Groups DAUIN - GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA, ELECTRONIC DESIGN AUTOMATION - EDA, GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA
Thesis type EMBEDDED SOFTWARE DEVELOPMENT, EXPERIMENTAL, RESEARCH, SOFTWARE DEVELOPMENT
Description The candidate will focus on designing and optimizing Spiking Neural Networks (SNNs) for the task of blood pressure estimation using Photoplethysmography (PPG) signals. This thesis will leverage the SNNtorch framework, with a preliminary phase aimed at understanding its functionality and applying it to blood pressure (BP) estimation tasks using benchmarks from four publicly available datasets (as cited in Nature).
Subsequently, the candidate will design and train a large-scale SNN architecture without constraints on memory usage to achieve state-of-the-art performance on PPG-based BP estimation. In the final phase, a Neural Architecture Search (NAS) algorithm tailored for SNNs will be developed to optimize the network for deployment on a multi-core RISC-V hardware platform, specifically GAP9, ensuring resource efficiency and performance alignment with the hardware constraints.
Key milestones:
• Mastering SNNtorch for bio-signal tasks and benchmarking performance on HR estimation.
• Propose and train a large, unconstrained SNN architecture for PPG-based blood pressure estimation.
• Developing a NAS approach to optimize the network for GAP9 hardware.
• Validating the optimized SNN on the target hardware platform with real-time simulations.
Useful reading to understand the topic:
- https://arxiv.org/abs/2109.12894
Required skills Proficiency in Python is required. Familiarity with Deep Learning, Spiking Neural Networks and the PyTorch library is a plus.
Notes Thesis in collaboration with Dr. Elisa Donati’s research group at the INI, Institute of Neuromorphic, Zurich. The thesis can be also carried out in Zurich.
Deadline 03/12/2025
PROPONI LA TUA CANDIDATURA