Depuis sa création, ANITI organise des focus techniques à destination de ses partenaires industriels. Retrouvez la liste des librairies produites par ANITI et accessibles en open source.
Explainability and Model Robustness
– XPLIQUE is a Neural Networks Explainability toolbox, composed of various modules implementing various methods such as Attribution Methods, Feature Vizualization, Concepts methods (replay – slides – github). The 2024 version “XPLIQUE UNLEASHED” incorporates new features : new operators, improved attribution methods, automatic concepts extraction and visualization methods(replay – slides – github – tutorials)
– GEMS.AI is a A Python package for AI fairness and interpretability. The approach is to build counterfactual distributions that permit answering “what if?” scenarios. The key principle is that we stress one or more variables of a test set and we then observe how the trained machine learning model reacts to the stress (website – replay – slides – github)
– Build 1-Lipschitz Neural Networks to guarantee robustness with DEELLIP (replay – slides – github)
– PUNCC a library to predict and quantify the uncertainty of the result of a ML model (replay – slides – github)
– INFLUENCAE a library for tracing the influence back to the data points (replay – slides – github)
Computer Vision performance
– Advanced tools for Computer Vision using bio-inspired feedback mechanisms (replay – slides – github)
– Improve your Computer Vision with PREDIFY a brain-inspired framework (replay – slides – github)
– An evolutionary testing simulation with Out Of Distribution Images, SimODD (replay – slides – github)
Optimization
– Toolbar2 : un optimiseur efficace pour les modèles à variables secrètes (replay – slides – github)
– Skicit-decide : an AI toolbox for reinforcement Learning, Automated planning and Scheduling (replay – slides – github)
Operational domain / domaine d’emploi
– OODEEL: Oodeel allows the use of state-of-the-art methods from the Out-Of-Distribution (OOD) detection literature for any image classifier in PyTorch and Tensorflow. It focuses on Post-Hoc OOD methods that apply to already trained models and, therefore, do not require further training (replay – slides – github)
French 
English