[Stage M2] Representation Learning and Domain Adaptation (Strasbourg)

When:
15/02/2020 – 16/02/2020 all-day
2020-02-15T01:00:00+01:00
2020-02-16T01:00:00+01:00

Annonce en lien avec l’Action/le Réseau : aucun

Laboratoire/Entreprise : ICube, University of Strasbourg
Durée : 6 months
Contact : lampert@unistra.fr
Date limite de publication : 2020-02-15

Contexte :
Created in 2013, the laboratory brings together researchers from the University of Strasbourg, the CNRS (French National Center for Scientific Research), the ENGEES and INSA of Strasbourg in the fields of engineering and computer science, with imaging as the unifying theme.
With around 650 members, ICube is a major driving force for research in Strasbourg whose main areas of application are biomedical engineering and sustainable development.

SERTIT, a service platform of ICube, known for its ISO certified rapid mapping service, is seeking to accelerate its mapping activities through artificial intelligence. This service assists in post-crisis emergency management (e.g. ground rescue, reconstruction efforts …).

More information:
http://icube.unistra.fr/en/

accueil

Sujet :
You will exploit state-of-the-art advances in multi-modal and multi-domain representation learning made in the data science and knowledge research group (SDC) to detect objects in satellite images of different characteristics (resolution, bands, etc), i.e. modality, in collaboration with remote sensing experts in SERTIT.

These models have been developed with benchmarks and medical datasets in mind and need to be extended and refined to work with more complex, higher dimensionality data such as satellite imagery.

The work has two benefits: on the one hand, to reduce the burden of ground truth collection when sensors of different characteristics are used; and on the other to exploit the information contained in each data modality to learn representations that are more robust and general, i.e. to detect buildings/roads/trees in different countries that exhibit different characteristics.

Your contributions will be part of the global work of the SDC researchers who aim to propose and implement new generic methods and tools to exploit large sets of reference data from one domain/modality (sufficient to train an accurate detector) to train a multi-modal/domain detector that can be applied to imagery taken from another sensor for which there exists no reference data.

As such, the work tackles problems that are key to many machine learning and computer vision applications.

• You will join a transversal team of researchers, software engineers and geomatics specialists from SERTIT and SDC (Data Science and Knowledge research group)

• Collaborate with research teams to transfer deep learning models to applications in remote sensing

• Build deep learning pipelines for multi-modal domain adaptation

• Participate in a research and development team

• Develop experimental protocols

• Perform thorough evaluation of proposed solution

Further Reading:

[1]. J. Shen, Y. Qu, W. Zhang and Y. Yu, “Wasserstein Guided Representation Learning for Domain Adaptation,” In Proceedings of the AAAI Conference on Artificial Intelligence, 2018.

[2]. Y. Bengio, “Deep Learning of Representations for Unsupervised and Transfer Learning,” In Proceedings of the Conference on Advances in Neural Information Processing Systems, 2012.

[2]. K. Bousmalis, et al. “Domain separation networks,” In Proceedings of the Conference on Advances in Neural Information Processing Systems, 2016.

Profil du candidat :
2nd year of a Master’s in Computer Science degree or similar

Formation et compétences requises :
• Experience with the Python (numpy, keras, tensorflow, etc.)

• Interest/experience in deep learning

• Knowledge of machine learning workflows and techniques (e.g. best practices around training data management, understand basics of numerical optimisation)

• Familiarity with Linux environments

• Have excellent communication skills and a strong team player

• Good knowledge of English (French is not mandatory)

• Be enthusiastic!

Adresse d’emploi :
ICube
300 bd Sébastien Brant – CS 10413
F-67412 Illkirch Cedex
France

Document attaché : Intern_Ad_RL4MSD_ENG.pdf