Supervisors and Institutions
Paleontological data are essential to obtain insights on changes in paleobiodiversity through the geological time and the biotic response to past events of global change. Investigations are usually carried out through the analysis of large databases, in which the record of species level occurrences is of prime importance. However, the collection of numerous data at species level is labour intensive; also, taxonomic occurrence data may be influenced to some extent by the opinion of taxonomists. The use of neural networks may be the solution to study big datasets more rapidly and objectively. This PhD project will apply neuromorphic modelling in radiolarian identification. The project will be jointly supervised by Prof. Taniel Danelian, a radiolarian micropaleontologist, and Prof. Pierre Boulet, a computer scientist specialist of neuromorphic computing.
The project will focus in exploring two main scientific questions:
(1) How can neural network learning achieve equal accuracy in the identification of Middle Eocene radiolarian species as an Eocene specialist in radiolarian taxonomy?
(2) How does neural network identification compare with the results of classical morphometric studies?
We will compare deep learning and spiking neural networks for these tasks. Deep learning is the mainstream technology and spiking neural networks have the potential for unsupervised learning and ultra low power hardware implementation allowing the building of autonomous classifying microscopes.
The possibility exists for a month's visit at the Natural History Museum in Berlin and possibly a month's internship in the Biostratigraphy Department of TOTAL. The PhD student will also be involved in the humAIn artificial intelligence alliance of the Hauts-de-France region.
We expect the candidate to have a strong background in paleontology/biology and an interest in computer science.