Interoperability Between Heterogeneous Multi-agent Systems Recommended by FIPA: Towards a Weakly Coupled Approach Based on a Network of Recurrent Neurons of the LSTM type
Noureddine El Abid Amrani, Mohamed Youssfi, Omar Bouattane, Oum El Kheir Abra · 2020
In this paper, we propose an approach that combines a CNN convolution neural network and an LSTM-type recurrent neural network to ensure interoperability between heterogeneous multi-agent systems recommended by FIPA (Foundation for Intelligent Physical Agents). FIPA has proposed a set of standards and specifications to maximize interoperability between heterogeneous multi-agent platforms. These FIPA standards and specifications focus on the communication aspect between agents based on the FIPA-ACL language, which is not sufficient to ensure interoperability between heterogeneous agents. Message content, and higher semantic levels that are more domain-specific (e.g. ontologies), must also be standardized while maintaining separation from the communicative level. Similarly, the lower level and the message transport must also be interoperable. In this context, we propose in this paper to integrate in the architecture of these systems, in a first step, a Broker that relies on the AMQP (Advanced Message Queuing Protocol) protocol, such as RabbitMQ, to ensure interoperability in the message transport level between agents. Secondly, a neural network composed of two layers: the CNN layer to extract the main characteristics of a message sent by an agent and send them to the second layer; a recurrent neural network of the LSTM type, which will process them characteristic by characteristic in order to interpret the content of the message.