Action, Logical Reasoning and Spiking networks – ALoRS

This project involves the design of a multiagent hybrid architecture. It approaches in a multidisciplinary way the study of conceptual neural networks based on spikes called ``spiking neural networks’’ (SNN), which have the advantage not only of being biologically credible, but favorable to be modeled using formal tools. Thus, the neural network will represent the long-term memory of an artificial agent, while its working memory and its rules of reasoning and planning will be represented using logic.
Our approach is first to implement a large-scale SNN in order to obtain optimal internal functioning from a computational time point of view. This includes a parallel calculation of the reaction of the output neurons to the stimulus entering the network. We will then formalize the link that exists in human cognition between a conceptual network (its long-term memory) and the central cognition where reasoning (working memory) occurs. This study will also make it possible to fix the relationships that can exist between knowledge updating operations as they are known in symbolic AI, and the updating of conceptual neural networks in the image of what is happening in our brain. Finally, we will produce a hybrid architecture of an intelligent artificial agent with different modules representing its long-term memory (SNN), its working memory (knowledge base), its ability to update its memories, as well as its ability to reason and plan.
This project therefore aims to propose a model reconciling the symbolic and connectionist approaches of AI by proposing such a hybrid architecture. In addition, it also has the ambition to create a bridge between computer science, neuroscience and psychology, by basing this architecture on our knowledge of human cognition and a biologically credible model of neural network. Finally, the link it establishes between abstract operations of the evolution of a knowledge base in logic, and the updating of conceptual neural networks, also provides elements for bringing the brain and knowledge together.
In addition to theoretical publications in conferences and journals in AI and neurosciences, several implementations are planned, including two prototypes (TRL4) that will be distributed in open access to the greatest number.