At deepomatic, we believe artificial intelligence is the way to unlock the world of tomorrow. Our mission is to make it accessible to all, not just a few big technology companies. To fulfill this dream we are always looking for talented and creative engineers to help us along this journey.
What we are looking for
If you're passionate about the upcoming societal and economical impacts of AI, if you want to work with the latest technologies, if you have the skills to show for it, and most of all if you want to live an adventure; then there is a place for you!
We know that there's always more to someone than meets the eye so we appreciate atypical profiles: education, background, previous experiences, nationalities, etc., don't hold back!
This is a paid position for at least 6 months.
The field of deep learning has significantly advanced in recent years. One of the latest advancement is the exploration of hyper-complex neural networks, the complex-valued counterpart of existing real-valued networks. Indeed, recent research has shown that hyper-complex networks such as quaternion- or octonion-valued networks improve performances in image classification while simultaneously having a reduced number of parameters to train, hence resulting in faster training and inference times due to the smaller model size.
The aim of the internship is to, firstly, implement a Quaternion-Valued Neural Network in TensorFlow/Keras starting from the existing implementation by Chase Gaudet (DeepQuaternionNetwork) and evaluate its performances on real-life datasets. Secondly, targeted regularisation as proposed by Vecchi et al. can be added. Finally, the application to detection models should be investigated.