Internship: Combining Uncertainty Quantification and Explainable AI for More Reliable Explanations

Stack

Responsibilities

• ▹Analyse existing uncertainty estimation methods in machine learning, in particular those based on uncertainty decomposition (aleatoric and epistemic)
• ▹Design a counterfactual explanation generation algorithm that integrates constraints on uncertainty
• ▹Implement the proposed method and evaluate the quality of the generated explanations on datasets of different natures (tabular, images)
• ▹Compare the developed method with one or two recent state-of-the-art competing methods

Requirements

• ▹Bac+5 (five years of higher education): engineering school or master's degree, in AI/data science
• ▹Python programming skills (PyTorch/TensorFlow, scikit-learn)
• ▹Knowledge of statistics, mathematics and statistical learning
• ▹Interest in the topic

About the company

Orange Innovation Recherche steers Orange's research work, drawing on the Orange Innovation Data-AI, Networks, IT and Services, and Marketing & Design teams. The Trust, Security and Transactions research area addresses security challenges in a constantly evolving ecosystem.