ExpressIF® is a symbolic AI that can simulate human reasoning using knowledge representations like Boolean and fuzzy logic, for example. It makes explainable decisions based on imprecise knowledge and spatiotemporal data.
ExpressIF® can simulate human reasoning to either automate or assist with decision-making within a given domain.
It explains its reasoning based on various types of information provided by domain experts and modelled in the form of rules or constraints, for example.
It can make decisions from knowledge that is either imprecise or only partially verified and can handle complex concepts and work from incomplete observations.
It is also capable of processing spatial and temporal data and combining the two in its reasoning.
Unlike deep learning AI systems, ExpressIF® belongs to the family of symbolic AI systems capable of producing explainable results. This makes it ideal for use cases like diagnostic assistance, system control assistance, and crisis response, where experts need to understand the decisions made by the AI before they take action or audit decisions already made.
The tool comes with an intuitive rule editor allowing domain experts to create their own rules using natural language.
ExpressIF® was developed by CEA-List and is protected by several patents.
Main advantages:
ExpressIF®, with its unique spatial and temporal reasoning capabilities and ability to intelligibly process uncertain data, is a powerful crisis response tool. Firefighters in southwestern France have used ExpressIF® to improve their response to wildfires.
For this use case, two modules were activated. The first forecasted the fire’s progression in time and space in order to assess the risks of certain interventions and recommend appropriate actions. The reasoning was based on detailed geographic data and weather forecasts.
The second module used the tool’s temporal reasoning capabilities to monitor the temperatures of emergency vehicles and personnel on the ground, assess the associated risks, and raise alerts in the event of danger.
For more information, read the article A decision-assistance tool for crisis management
A machine learning module for classifying and annotating images—tagging objects or points of interest in a spectrum or signal, for example—was recently added to ExpressIF®.
To understand an image, a neural network first identifies specific areas that correspond to certain objects in the image. The new feature integrated into ExpressIF® then takes over, identifying objects using rules that allow it to position the objects clearly and annotate them. The main advantage is that only a few images—less than ten—are required for learning.
The first tests—carried out on abdominal MRI images—showed that the tool could not only automatically annotate the organs, but also explain its annotations with reasoning expressed in natural language. This represents a concrete step towards efficiently helping doctors interpret patient images.
In the news: Image recognition system can classify, annotate, and explain how
A Fuzzy Expert System Architecture For Data And Event Stream Processing, Poli J-P and Boudet (2018). L Fuzzy Sets and Systems. Vol. 343, pp. 20-34. Elsevier.
Natural Language Generation of Explanations of Fuzzy Inference Decisions, Baaj I and Poli J-P (2019). 2019 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE). pp. 1-6.
Spatial relation learning for explainable image classification and annotation in critical applications, Pierrard R, Poli J-P and Hudelot C (2021). Journal of Artificial Intelligence, 292.
Online Spatio-Temporal Fuzzy Relations, Poli J-P, Boudet L and Le Yaouanc J-M (2018). 2018 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE). pp. 1-8.
Material Classification from Imprecise Chemical Composition: Probabilistic vs Possibilistic Approach, Grivet Sébert A and Poli J-P (2018). 2018 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE). pp. 1-8.
