Student project: Event-based learning algorithm for asynchronous neuromorphic hardware 

Modeling of asynchronous inference and designing a learning rule to exploit the low latency feature of asynchronous neuromorphic processors.  

What you will do

GPUs are dominated by HW accelerators for training deep neural networks. Even though GPUs are very flexible, they still imposed constraints when one wants to train Spiking Neural Networks efficiently on GPUs. One of those constraints is the existence of time-step processing in SNNs instead of spike-by-spike processing. However, synchronized time-stepped SNNs cannot exploit the asynchronous inference feature of many neuromorphic hardware which results in ultra-low latency prediction.  

In this project, we would like to develop algorithms suitable for event-based inference and minimize inference latency and energy consumption. Our target application is small to medium size neural networks with simple datasets (e.g., IBM Hand Gesture dataset) and the focus is to implement the new learning rule in an efficient way to be optimized and deployed on imec neuromorphic processor. The KPI to benchmark the algorithm is latency and energy consumption as well as the prediction accuracy.  

Tasks:

• Literature study on asynchronous event-based inference. 
• Design, optimization, and implementation of the inference model and learning rule for best latency and energy consumption during inference.  
• Validation and demonstration. 
• Documentation. 

Reference:
[1] Ye, Mang, et al. "Deep learning for person re-identification: A survey and outlook." IEEE Transactions on Pattern Analysis and Machine Intelligence (2021). https://arxiv.org/pdf/2001.04193.pdf 
[2] Yousefzadeh et al. “SENeCA: Scalable Energy-efficient Neuromorphic Computer Architecture”, AICAS 2022
[3] https://www.sony-semicon.co.jp/e/products/IS/industry/product/evs.html   

What we do for you

Imec is one of the world's leading research institutes in micro and nano-electronics. The work of this project will fall under the scope of a European ECSEL project ("DAIS"), and the outputs may be published in high-impact journals/conferences (subject to the quality of the work). ImecNL provides the required equipment, access to lab facilities, a workplace in the Holst Centre at High Tech Campus, and a monthly allowance during the internship. 

Who you are

• M.Sc./Ph.D. students with a relevant background (non-European students are only eligible if they study in the Netherlands).
• Available for 9 months (the project can be extended up to 12 months).
• Have excellent programming skills in Python(TensorFlow/Keras) and C. 
• Have a good understanding of supervised learning rules in DNNs.
• Are in good command of spoken and written English. 
• Motivated student, good communicator, easy collaborator, and eager to work independently and expand knowledge in the field. 

Interested

Does this project sound like an interesting next step in your career at imec? Don’t hesitate to submit your application by clicking on ‘APPLY NOW’.
Should you have more questions about the project, you can contact Amirreza Yousefzadeh by mail amirreza.yousefzadeh@imec.nl.
Got some questions about the recruitment process? Marsha Loomans of the Talent Acquisition Team will be happy to assist you.