Post-doc Informatique

L'UTC présente un modèle de formation où les sciences de l'ingénieur, les sciences humaines et sociales, les sciences économiques et politiques sont intégrées harmonieusement au service de l'éducation de l'ingénieur, du scientifique, du manager du futur, innovant, humaniste, apte à maîtriser les enjeux de la complexité dans une société de l'information et de la communication. 

This post doc project aims to apply machine learning to analyze 300 to 500 data elements for patients suspected of long-term infections, including vector-borne. The data consists of clinical signs, biochemical blood analyses, serological tests, MR and CT scan text reports, and epidemiological data. The goal is to develop classification algorithms that can highlight the effects of long-term infections with a single vector or multiple vectors.

The first phase of the project will address the challenge of obtaining reliable, complete datasets from standard online forms filled out by patients under the supervision of medical assistants. The machine learning algorithms will be adapted to process subjective clinical signs, which are nuanced according to amplitude, frequency, and evolution in time.

In the following phase, treatment data will be added to the patient database, which will involve processing data captured at different timestamps. The learning dataset will be constantly updated with data from new confirmed patients. The classification methods used should remain simple.

The project will also explore reinforcement learning to identify the most relevant features and parameters for the considered pathology. We will assess its use in reducing the duration of expertise transfer during the supervised learning phase of the machine learning system. The project will also evaluate how patients perceive this assisted diagnosis process.

Future work will address the ability of a machine learning system to justify its decisions, an issue known as “explainability”, particularly tricky in deep learning implementations.

The applicant should have good knowledge about data science, and data mining tools:

- Python, pytorch, scikit-learn, Tensor Flow, etc.

- Design a patients' database, and/or interfaces between different databases

- Design a machine learning algorithm to assist the diagnosis by analyzing health data