Geoffrey Hinton released last month a document titled “Dynamic Routing Between Capsules“ and the entire Deep Learning community was shaken by this article.
This publication compares the new algorithm with the latest CNNs. The authors hypothesize that the human brain has modules called “capsules”. These capsules are particularly suitable for handling different types of visual stimuli and coding things like pose (position, size, orientation), deformation, speed, etc. It is assumed that the brain has a mechanism for “routing” low-level visual information to the “capsule” best suited to managing that data.
The new architecture proposed by Hinton seems to allow CapsNet networks to learn internal representations (such as posing of an object) to better generalize the class for 3D objects.
The beginning of this dissertation will consist of a research work in literature on both the convex networks and the new theories of Capsule Networks. Work will continue with code implementation and objective measurement of network performance and comparison with the results obtained from the latest CNN implementations.
The company’s interest is to identify the benefits and potential CapsNet practical applications for artificial vision. This includes the use of CapsNet for image segmentation and the reduction of the number of instances needed during the training phase.
It requires that the candidate has considerable mathematical backgrounds, analytical skills and good programming skills especially in Python.
The candidate will be supported by highly qualified staff and will be able to access business computing resources (IBM PowerAI equipped with NVIDIA P100 and V100 cards) when needed.
Who we’re looking for
Students that are about to get their master degree in: Computer Science, Mathematics
Skills: Python, C/C++, math, preferred knowledge of TensorFlow / PyTorch / BigDL
Duration of this Projects: 6-8 months