1. What is the purpose of the catalog?

The catalog is a web application for browsing the results of the Confiance.ai program; it offers sorting and search features to make it easier to identify the assets that are most relevant for a specific usage.


2. What is the Confiance.ai program?

The objective of the Confiance.AI program is to respond to the challenges and issues posed by the integration of artificial intelligence into critical systems, from specification to maintenance in operational conditions. To do so it delivers a set of software components whose interoperability will make it possible to compose tool chains within existing engineering workbench which, supported by methodological guidelines, make it possible to design, validate, deploy and maintain critical systems based on artificial intelligence.


3. What are the results of the Confiance.ai program?

The results of the Confiance.ai program take several forms, three in the first instance:

  • Methods,
  • Software components,
  • Functional sets.


4. What is a method?

The methods are documentary deliverables, they explain and specify how to address the issues specific to AI-based systems engineering, particularly with regard to the trust that one wishes to obtain in these systems. Some of the published methods cover large areas of AI-based systems engineering, such as the :

Other methods, on the other hand, focus on specific issues, such as the characterization of robustness, domain adaptation, adversarial training, etc.


5. What is a software component ?

There are three types of software components among the results available in the catalog:

  • Python libraries,
  • Containerised images,
  • Demonstrators,

Most of the available components are python libraries, they are packaged and can be installed either via the public library with a simple command pip install libraryname, or by retrieving the package using the same credentials as those that will allow access to the catalogue. Whether it is one or the other method, the information is available on the component page.

Containerised images, in this case Docker images, are used to simplify the deployment and use of components that can be complicated to deploy: they are usually complete applications and not just a library. As with libraries, Docker images can be retrieved via the public docker directory, by doing a docker pull applicationname or by connecting to the irtsysx image repository, always with the credentials used to access the catalogue. When the license of use and distribution allows it, the source code of the containerised application is also made available.

Finally, demonstrators, although they are rather rare, correspond to the implementation of methods on specific problems, for example image processing, time-series, among other topics. If they are not function libraries, the demonstrators often use them to illustrate their use. They may be used to be modified and applied over different set of data or models.


6. What is a functional set ?

AA functional set is a set of methods and software components dedicated to a particular topic: for instance, the robustness of a machine learning or the quality of the data used to train one. To date (May 2023) no functional set is available, but the program is moving towards the consolidation and delivery of coherent functional sets, i.e. with documentation, tutorials and videos that explains their use and composition, and a set of components that can be used together or in isolation. At present, the seven functional sets planned are the following, but it is likely that new sets will be identified :

  • End to End
  • Data Life Cycle
  • Model Component Life Cycle
  • Evaluation
  • Deployment
  • Operation
  • Uncertainty
  • Robustness
  • Embeddability


7. What do one need to know to correctly navigate through the program results ?

In addition to their distribution across methods and components, program results are also identifiable by means of certain properties:
Trust attributes, which identify to which trust dimensions an asset in the catalog is relevant such a :

  • Robustness,
  • Reliability,
  • Safety,
  • Availability,
  • Accountability,
  • Maintainability.
  • Confidentiality.
  • Replicability.
  • Integrity.
  • Security.

Engineering activities, which with a strong machine learning coloration, indicates at which stages of the design process the result (library or application) is intended to be used :

  • Perform Machine Learning Model training,
  • Monitor an AI-based system,
  • Specify, develop, implement online monitoring of inputs,
  • Test machine learning model robustness,
  • Select a strategy for machine learning evaluation,
  • Evaluate data trustworthiness.

Use-case Typology which designate the type of data the library or the application is relevant to work with :

  • Visual inpection
  • Vision
  • Time-Series
  • Object reidentification
  • Surrogate Model
  • NLP
  • Object Tracking


8. What information is available for each result?

A result shows the following information:

On the upper side,

  • A description of the component in a few lines followed by the different properties indicated in point 7.

  • A list of the files included in the component

  • The version of the component, followed by the name of the collaborator responsible for making the component available. Attention: this is not necessarily the person who created the component, it may simply be the person who is best able to provide information on its use,

  • Then there is a first link to the documentation for using the result. Depending on the nature of the software component, this may be only the readme available on the repository or a more extensive documentation.

  • When it exists, a link to the official site of the component is then proposed.

  • Then a link to download the component directly, or to retrieve it via a 'pip install' or 'docker pull' command.

On the lower side,
  • the usage and distribution licence associated with the entity holding the rights over the intellectual property,
  • The "functional set" field indicates to which functional set this result belongs.
  • The maturity field should be considered with care: it indicates the maturity as assessed within the Confiance.ai program. Maturity is considered through two dimensions: technological maturity (to what extent the component is usable) and functional maturity (is the component useful?). As the evaluation and maturation activities are in progress (2023), the available assets still need to be strengthened.
  • Finally, the type appears, which provides information on the nature of the component itself: is it a library, a docker image or a demonstrator?


9. Why are some results licensed ?

The confiance.ai program is not intended to develop, by itself, all the functionalities necessary for the design of AI-based trust systems. Many of the proposed results are derived from existing methods outside the programme but which have been tested and evaluated within it. Indeed, these methods are often licensed.


10. What is a Confiance.ai license or owner ?

The developments carried out exclusively within the framework of the Confiance.ai programme are covered by the consortium agreement which governs the programme and its partners. This means that the intellectual property of this work is currently shared between all the industrial partners and that their valorisation will have to be discussed.


11. I need help, who can help me?

For each result, a contact person is listed, along with their email address. This is the person, a member of the confiance.ai programme, who is best placed to answer any questions. Please note, however, that although they can provide help, these people are not responsible for providing support.


12. I would like to give some feedback, how do I do this ?

If it is about a specific component you can contact the responsible person, indicated on the component result page. If it is a more general feedback or doesn't fit within a specific component, you can send a feedback email Avoid using this email adress to get support as the responsiveness may be poor.