How it works
About GEMS AI
Go to Github
What about GEMS AI?
Machine Learning Algorithms are trained on database to provide accurate predictions. But the main issue is how far can we trust them ? Can we certify the behaviour of these algorithms ? How can we be sure that the outcome of the algorithms will be reliable and unbiased ? This need for explanation of AI-based algorithms is at the core of GEMS-AI.
The Research project GEMS for AI aims at understanding the limitations of a machine based predictive model by analyzing its response under stress conditions. Using mathematics, we provide a certified framework for explainability of algorithms by pushing them to their limits and studying how they react to perturbations on the input.
Our goal is twofold :
– Provide a tool to analyse the reasons why an automatic decision was taken ;
– Explain the decision rules by understanding the effects of the variables that change the decisions, and thus detecting possible biases or worse unethical behaviour of the algorithm.
GEMS AI for what ?
Model interpretability is a difficult issue when confronted to the complexity of modern black box algorithms, especially deep neural networks.
Explaining and Understanding the outcome of an algorithm is necessary to provide trust in the use of Artificial Intelligence in real world applications, especially when the decisions may impact and reshape the world.
A special application is to prevent reproduction and generalisation of unfair human discrimination that probably reflects in the learning data. Model fairness consists in making sure a decision was not based on protected attributes (e.g. gender, race… for a bank loan). Hence understanding the specific contribution of each variable in the decision is an important step for enhancing fairness in AI based systems.
For several communities
to find an free certified code/app
to work on this program
DSI (direction du système d’information) / information system management ?
to understand and explain automatics and find a certified code/app to theirs problematics
who want to understand mathematics solutions to AI fairness and more
How IT WORKS?
In this notebook, we explain the maths behind GEMS-AI and how it is implemented. It is not mandatory to start using the package. To so so, please read the Getting started tutorial. Let’s assume you have a dataset and a model that generates predictions on this dataset. For the sake of the example, we’ll be wortking with the Adult dataset.
What YOU CAN DO?
Presentation of main characteritics of the toolbox and their properties.
Understanding how to use the toolbox through the study of some examples.
Guidelines and tutorials to install the API.
Some ideas or concerns on the toolbox: please come and join by contacting the research team.
Join our team to ask questions, make comments and tell stories about.
Who WORKS ON?
This work is done at the Toulouse
Mathematics Institute from Université Toulouse 3.
It is supported and funded by the Centre
National de la Recherche
and in collaboration with the
Artificial and Natural Intelligence
Toulouse Institute (ANITI) project.
This software is released under the
Code source on GitHub
You can find onGitHub : theIntroduction, Installation guide, User guide (Measuring model influence, Evaluating model reliability, Support for image classification), Authors and License.
118 Rte de Narbonne
31400 Toulouse - FRANCE