ExpressIF

ExpressIF^® is a software suite that enables the production of explainable artificial intelligence based on reasoning and knowledge. It is the product of classical algorithms, the latest state-of-the-art technology, and an efficient implementation.
Over the years, various contributors have contributed their building blocks to bring you these advanced features.
Today, our research focuses primarily on new features to facilitate knowledge extraction from data, the usability of ExpressIF^®, and the confidence you can have in its decisions.
Below, we list a few research areas, which have relatively low TRLs but which you can access through collaborative projects.

Expressiveness

The expressiveness of a knowledge-based system is the entire vocabulary it can use. To ensure that the knowledge used by ExpressIF^® closely matches the knowledge we wish to express, we enrich ExpressIF^® with generic vocabulary.
To add "words" or relationships, we model them mathematically using various formal frameworks: morpho-mathematical, signal processing, logic, etc. This advanced vocabulary is then used in the explanations that ExpressIF can generate.

Expressiveness

The expressiveness of a knowledge-based system is the entire vocabulary it can use. To ensure that the knowledge used by ExpressIF^® closely matches the knowledge we wish to express, we enrich ExpressIF^® with generic vocabulary.
To add "words" or relationships, we model them mathematically using various formal frameworks: morpho-mathematical, signal processing, logic, etc. This advanced vocabulary is then used in the explanations that ExpressIF can generate.

Knowledge extraction

For each of the reasoning approaches we propose in ExpressIF^®, we develop knowledge-based model induction algorithms. Our work allows us to focus on frugal learning that approximates human capabilities. Our approaches benefit from the full vocabulary at our disposal to create interpretable models effortlessly. We are also working on coupling learning with existing knowledge, simulations, etc. Our methodology leads us to algorithms based on realistic assumptions: in particular, we avoid assumptions about data distribution or data quantity. Finally, we explore different paradigms to offer you the best of symbolic learning: active learning, reinforcement learning, etc.

Knowledge extraction

For each of the reasoning approaches we propose in ExpressIF^®, we develop knowledge-based model induction algorithms. Our work allows us to focus on frugal learning that approximates human capabilities. Our approaches benefit from the full vocabulary at our disposal to create interpretable models effortlessly. We are also working on coupling learning with existing knowledge, simulations, etc. Our methodology leads us to algorithms based on realistic assumptions: in particular, we avoid assumptions about data distribution or data quantity. Finally, we explore different paradigms to offer you the best of symbolic learning: active learning, reinforcement learning, etc.

Usability

In our research, we are interested in the usability of ExpressIF^® and how we can help you use it optimally. This includes the ergonomics of the user interface, as well as the rendering of decisions and their explanations. Incorporating state-of-the-art elements allows us to compare ourselves to existing systems using a panel of users.

Usability

In our research, we are interested in the usability of ExpressIF^® and how we can help you use it optimally. This includes the ergonomics of the user interface, as well as the rendering of decisions and their explanations. Incorporating state-of-the-art elements allows us to compare ourselves to existing systems using a panel of users.

Research

Expressiveness

Expressiveness

Knowledge extraction

Knowledge extraction

Usability

Usability