PhD - Antwerpen | More than two weeks ago
Traditional deep learning algorithms (DNN) have gotten very powerful over the last few years ( ) but also very data and power hungry. This limits their usage in edge scenario’s where computational power and power budgets are limited. Additionally deep learning networks tend to operate in a black box manner with limitations towards and uncertainty definition. It is infeasible to understand how comes to its conclusions in massive neural network. it is not able to express whether the answer that is given is true or false and how certain it is about its answers.
For use cases that require more insight in algorithmic behavior (such as medical applications) other learning algorithms are preferred, either replacing or complementing a deep neural network. Examples of such algorithms are Bayesian neural networks, Bayesian networks, Markov random fields... These algorithms can operate on limited amounts of data and ingest expert knowledge to arrive at “explainable” conclusions.
To be feasible in edge scenario’s, additionally to lower data consumption/smaller model size, the computational power budget needs to be reduced as well. Training/executing machine learning algorithms tend to be very inefficient on general hardware. Custom AI accelerators are created, optimizing the hardware for specific algorithm needs (massive parallel compute, sampling...).
In this project you will investigate (probabilistic) accelerator architectures for hardware/software co-design using novel devices (MRAM, 2D ) and analyzing/optimizing the impact on algorithms and applications.
You will join the imec AI group (), which focusses on research and engineering in the domain of Edge AI. The team is multidisciplinary and highly international, composed of talent with skills in ML algorithms, sensor fusion techniques, pipelines and general application development. is currently a team of around 100 people, operating from the imec offices in Ghent, Antwerp and Leuven. Its main emphasis is on software/hardware co-design and your work will be part of an ongoing effort for disruptive innovation through creative collaboration between the hardware and software departments at imec.
Required background: Computer science, Engineering Technology
Type of work: 20% literature, 10% experimental, 70% modeling/simulation
Supervisor: Steven Latré
Daily advisor: Ben Stoffelen, Tanguy Coenen
The reference code for this position is 2023-154. Mention this reference code on your application form.