FR

EN

Homepage Inria website
  • Inria login
  • The Inria's Research Teams produce an annual Activity Report presenting their activities and their results of the year. These reports include the team members, the scientific program, the software developed by the team and the new results of the year. The report also describes the grants, contracts and the activities of dissemination and teaching. Finally, the report gives the list of publications of the year.

  • Legal notice
  • Cookie management
  • Personal data
  • Cookies



Section: Research Program

Decision support

Patterns have proved to be quite useful for decision-aid. Predictive sequential patterns, to give an example, have a direct application in diagnosis. Itemsets and contrast patterns can be used for interpretable machine learning (ML). In regards to diagnosis, LACODAM inherits, from the former DREAM team, a strong background in decision support systems with internationally recognized expertise in this field. This AI subfield is concerned with determining whether a system is operating normally or not, and the cause of faulty behaviors. The studied system can be an agro- or eco-system, a software system (e.g., a ML classifier), a living being, etc. In relation to interpretable machine learning (ML), this subfield is concerned with the conception of models whose answers are understandable by users. This can be achieved by inducing inherently white-box models from data such as rule-based classifiers/regressors, or by mining rules and explanations from black-box models. The latter setting is quite common due to the high accuracy of black-box models compared to natively interpretable models. Pattern mining is a powerful tool to mine explanations from black-box systems. Those explanations can be used to diagnose biases in systems, either to debug and improve the model, or to generate trust in the verdicts of intelligent software agents.

The increasing volumes of data coming from a range of different systems (ex: sensor data from agro-environmental systems, log data from software systems and ML models, biological data coming from health monitoring systems) can help human and software agents make better decisions. Hence, LACODAM builds upon the idea that decision support systems (an interest bequeathed from DREAM) should take advantage of the available data. This third and last research axis is thus a meeting point for all members of the team, as it requires the integration of AI techniques for traditional decision support systems with results from data mining techniques.

Three main research sub-axes are investigated in LACODAM:

  • Diagnosis-based approaches. We are exploring how to integrate knowledge found from pattern mining approaches, possibly with the help of interactive methods, into the qualitative models. The goal of such work is to automate as much as possible the construction of prediction models, which can require a lot of human effort.

  • Actionable patterns and rules. In many settings of “exploratory data mining”, the actual interestingness of a pattern is hard to assess, as it may be subjective. However, for some applications there are well defined measures of interestingness and applicability for patterns. Patterns and rules that can lead to actual actions –that are relevant to the user– are called "actionable patterns" and are of vital importance to industrial settings.

  • Mining explanations from ML systems. Interpretable ML and AI is a current trend for technical, ethical, and legal reasons  [28]. In this regard, pattern mining can be used to spot regularities that arise when a complex black-box model yields a particular verdict. For instance, one may want to know the conditions under which the control module of a self-driving car decided to stop without apparent reason, or which factors caused a ML-based credit assessor to reject a loan request. Patterns and conditions are the building blocks for the generation of human-readable explanations for such black-box systems.