The U2iS laboratory of ENSTA Paris at Institut Polytechnique de Paris is looking for internship students interested in the theory of deep learning.

You should have basic knowledge of Deep Learning and machine learning. Computer science/ or mathematician profiles are also welcome to apply. The future applicant

Nowadays, the computer vision community relies on deep learning algorithms that have been trained to perform various tasks and learned the necessary concepts on large amounts of data. However, these Deep Neural Network (DNN) are not reliable.

Gao et al [1] has demonstrated that DNNs are often overconfident. Hence, we cannot rely on their predictions. The classical DNN will classify these pixels, and the confidence score related to these predictions will be high, contrary to a strategy that we develop in OVNNI [2].

The goal of the internship will be to propose a solution to quantify the uncertainty of DNN for computer vision tasks.

References :

[1] Guo, C., Pleiss, G., Sun, Y., \& Weinberger, K. Q. (2017). On calibration of modern neural networks. arXiv preprint arXiv:1706.04599.

[2] Franchi, G., Bursuc, A., Aldea, E., Dubuisson, S., \& Bloch, I. (2020). One Versus all for deep Neural Network Incertitude (OVNNI) quantification. arXiv preprint arXiv:2006.00954.