We are searching for optimized hardware efficient algorithms for self-supervised fine-tuning of deep neural networks in our neuromorphic processor for optimized adaptivity in different applications.
What you will do
Since learning and adaptivity are among the main differentiators of neuromorphic technology, we would like to explore few applications domains (e.g. prediction of signals like audio/video, denoising, anomaly detection, etc.) and hardware efficient algorithms for online learning. Our vision is to start from an already trained neural network and perform a fine-tuning process  during inference in our neuromorphic processor. This fine-tuning results in higher accuracy for the specific task or more efficient inference by increasing spatio-temporal sparsity during inference. We are especially interested in exploring self-supervised learning algorithms  for DNN fine-tuning.
Our neuromorphic processor is flexible. It contains several RISC-V cores connected through an interconnected network. One of the main constraints of our hardware is its event-driven property. A Process in a RISC core only triggers with an event. Additionally, the different cores work independently from each other. These two constraints impose the implementation of an event-driven, distributed, and local learning mechanism.
The project duration is 9 to 12 months (preferable 12 months, by merging 3 months of internship and 9 months of M.SC. thesis). This project results in a demonstration of an application with online learning running on hardware. The outcome of the project may be published in high impact journals and may as well be patented.
We seek very motivated candidates with a relevant background, strong programming skills in Python (TensorFlow and/or PyTorch) and embedded C++ programming (for RISC-V programming). The target start date of the project is in summer 2021. The interested applicants should submit their CV, the academic transcripts (including the scores and the courses), and (if known) the name of the project supervisor from the university.
 Deep Learning using Transfer Learning (https://towardsdatascience.com/deep-learning-using-transfer-learning-python-code-for-resnet50-8acdfb3a2d38).
 Self-supervised learning: could machines learn like humans? (https://youtu.be/7I0Qt7GALVk).
- Literature review on neuromorphic architecture and relevant learning algorithms.
- Finding a suitable algorithm and applications for self-supervised learning.
- Implementation of the selected learning algorithm in python (TensorFlow/PyTorch).
- Implementation of the algorithm in the neuromorphic platform.
- Thesis writing and documentation in Imec Holst-Centre.
Who you are
- Msc in Electrical/Computer Engineering.
- Very good/excellent in python (TensorFlow/PyTorch) and embedded C (++) programming.
- Strong Knowledge of Deep Neural Networks training algorithms.
- A structured way of reporting, both orally and written.
- Motivated student eager to work independently and expand knowledge in the field.
- Good written and verbal English skill.
Click on 'apply' to submit your application. You will then be redirected to e-recruiting.