PhD position in hardware-software co-design for bio-inspired ML models

In the last decade, ML models taking inspiration from the brain and biological neurons have emerged as an alternative to conventional ML models, promising a more efficient processing of temporal signals at the edge. This includes a wide range of signals, among which cortical activity for brain-computer interfaces, audio signals for keyword spotting and artificial cochleas, and tactile signals for robot perception.

Various types of bio-inspired mechanisms have been investigated in recent years to endow these bio-inspired models with efficient processing capabilities. Structural plasticity for example consists in letting the network’s sparse connectivity evolve throughout training, while delay plasticity consists in learning dendritic, synaptic, or axonal temporal delays to enrich the network’s spatiotemporal dynamics. This research project will thus explore how these different mechanisms can be optimally combined to deliver models with extremely constrained compute and memory footprints without compromising performance. This includes training spiking neural networks with multiple plasticities at different timescales, exploring trade-offs between different design choices, quantizing the models, and implementing the best-performing ones in custom digital hardware, on FPGA and/or in an application-specific integrated circuit. These results will pave the way for on-device edge intelligence, as well as for a better understanding of relevant bio-inspired mechanisms to be included in resource-constrained models at the edge.

While biomedical applications will be the primary focus of this work, the resulting ML models and hardware will be designed to be easily portable to a broad range of tasks based on temporal data (predictive maintenance, sensory processing for robot control, etc), in which efficient on-device processing is crucial.

We are looking for a highly motivated PhD candidate with an interest in hardware-software co-design of ML models taking inspiration from the brain. The applicant should:

• have a Master’s degree in Engineering with a background in Electrical Engineering, Computer Science, Artificial Intelligence, or a related field
• be ranked within the top 10% of their class at MSc and BSc level, with outstanding study results
• have a strong background in ML algorithms, with programming experience in Python and, more specifically, in common deep learning frameworks such as PyTorch and jax, for model training and inference
• have experience with embedded platforms such as FPGAs or RISC-V microcontrollers. Experience with digital integrated circuit (IC) simulation and design tools, such as Cadence XCelium, Genus, and Innovus, is a plus
• have qualities to carry out independent research, among which communication skills, critical thinking, and research ethics
• demonstrate an excellent command of the English language, both in spoken and written form
• show strong interpersonal skills and the ability to work in an international team

We can offer the PhD candidate:

• A PhD scholarship of up to four years and, if successful, a PhD in Engineering Technology
• A competitive salary and additional benefits such as health insurance, access to university sports facilities, etc
• The opportunity to be active in an exciting and international research environment, engage in research collaborations, and participate to international conferences
• Full-time employment for four years, with an intermediate evaluation after each year
• Excellent doctoral training at the Arenberg Doctoral School in an international environment at a top European university
• A flexible working culture with opportunity for up to 40% remote working

For more information, please contact the future supervisor Prof. dr. ir. Martin Lefebvre, mail: [email protected].