Organisation/Company: CNRS
Department: Centre de Recherche sur la Conservation
Research Field: Physics
Researcher Profile: Recognised Researcher (R2)
Country: France
Application Deadline: 2 Dec 2024 - 23:59 (UTC)
Type of Contract: Temporary
Job Status: Full-time
Hours Per Week: 35
Offer Starting Date: 1 Jan 2025
Is the job funded through the EU Research Framework Programme? Not funded by a EU programme
Is the Job related to staff position within a Research Infrastructure? No
Offer Description The goal of this project is to use machine learning (ML) algorithms to model how the reflectance spectra of complex mixtures of colored materials from historical artworks, specifically illuminations and/or decorated letters from ancient manuscripts, depend on their chemical composition. Covering a wide range of energy from visible (VIS) to mid-infrared (mid-IR) coupled with X-ray fluorescence spectroscopy (XRF), a spectral database of reference inorganic and organic materials will be built. Subsequently, a supervised ML algorithm will be trained on the database, and the obtained ML model will be capable of simultaneously predicting the composition of the colored layer and the abundance of materials present in the mixture. Next, we will use the trained ML model to generate a greater variety of artificial reflectance spectra, thus expanding the existing data set. A second ML model will then be trained on the expanded data set to understand how the reflectance spectrum correlates with the composition of the pictorial mixture. This second model will be applied to the analysis of historical manuscripts. The candidate is expected to develop an approach that will offer an efficient and general method for onsite determination of the composition of historical illuminations based on their spectral response.
Key Responsibilities Creation of reference mock-up for colored material mixtures by acquiring reflectance spectra in the visible and near-infrared range, as well as XRF spectra.
Data augmentation: During this process, a neural network-based encoder-decoder model will learn how to generate synthetic spectra based on pigment concentrations. It will then be used to create an artificial dataset to complement the existing one.
Applications: The dataset created combining both real and artificial spectra will be used to train a neural network model capable of identifying pigments in mixtures. This model will be tested on datasets (hyperspectral VIS, NIR and XRF imaging) acquired on medieval manuscripts.
The postdoctoral fellow will work mainly at the CRC (Centre de Recherche sur la Conservation) at the Museum National d'Histoire Naturelle, Paris 5ème and at the Laboratoire de Chimie Physique - Matière et Rayonnement (LCPMR) at Sorbonne University, Paris 5ème. He will be supervised by Tsveta Miteva, specialist in machine learning and, in particular, deep learning, as well as Emeline Pouyet and Anne Michelin, physical chemist researchers, specialists in the characterization of heritage materials using non-invasive techniques.
Minimum Requirements • Python programming
• Knowledge in machine learning algorithms (most specifically encoder-decoder NN and transfer learning would be appreciated)
• Expertise in reflectance and X-ray fluorescence imaging spectroscopies (optional)
• Experience with multi-spectral dataset analyses (optional)
#J-18808-Ljbffr