Self-Learning and Self-Adaptive Systems – SLSAS

Research Activity

Self-Learning Systems, inspired by neuro-science and human-brain learning mechanisms, aim at learning solutions typically by means of try-and-error interactions with the surrounding environments or directly from collected data, rather than being manually pre-programmed.
Self-learning capabilities allow equipping complex systems with self-adaptation when deployed in dynamic and distributed environments. In detail, the research activities focus on the:

  • development of novel self-learning methods by combining Inverse Reinforcement Learning and Forward Reinforcement Learning;
  • definition of innovative self-learning hierarchical approaches, especially concerning the solution of multi-objective problems
  • definition of self-learning distributed approaches via multi-agent systems
  • development of methods and tools for the optimization of QoS parameters dynamically self-learned, rather than statically pre-defined

These methods exploit and integrate paradigms such as Reinforcement Learning, Inverse Reinforcement Learning, Multi-Agent Reinforcement Learning, Hierarchical Learning, Continuous Learning, Deep Learning.

 

Goals

The research objectives concern the definition of novel self-learning and self-adaptation methodologies for complex systems such as cyber-physical systems, multi-agent systems, robotic systems, and edge nodes, without relying on ad-hoc pre-coded knowledge.

 

Application Fields

Main application domains are: Healthcare, Industry 4.0, Robotics, Ambient Intelligence.

Back To Top