Bring Us MacGyver Predictor: Towards a Deep Learning-Based Mechanism to Design Emergent Behaviors in Systems-of-Systems

• Kanan Castro Silva UFABC
• Flávio Horita UFABC
• Valdemar Vicente Graciano Neto UFG

Resumo

Systems-of-Systems (SoS) involve independent systems called constituents that, together, achieve a set of goals by means of emergent behaviors. Those behaviors can be deliberately planned as a combination of the individual functionalities (herein named as features) provided by the constituents. Currently, SoS stakeholders heavily rely on the creativity of engineers to combine the features and draw the behaviors. The limitation of human perception in complex scenarios can lead to engineering sub-optimized SoS arrangements, offering global behaviors that are limited to the engineer’s abilities and prior experience, potentially causing waste of the resources, sub-optimal services and reducing quality.In that sense, the main contribution of this paper is introducing MacGyver Predictor, a deep learning-based mechanism for inferring/suggesting emergent behaviors that could be designed over a given set of constituents. An initial dataset was elaborated from a systematic mapping to feed the mechanism. We expect that our mechanism can extrapolate the human capabilities and glimpse global behaviors, hopefully revealing unpredicted behaviors that could be offered by the SoS and supporting engineers to architect SoS with (i) more diversified behaviors and (ii) enhanced SoS overall quality.

Palavras-chave: System-of-Systems, deep learning, multi-label classification, emergent behavior