search

hpi-sam / Robust-Multi-Agent-Reinforcement-Learning-for-SAS

Research project on robust multi-agent reinforcement learning (marl) for self-adaptive systems (sas)
MIT License
0 stars 0 forks source link

Emphasize on modular agent learning via divide and conquer #19

Open ulibath opened 2 years ago

christianadriano commented 2 years ago

Shops by agent is partition of the state-space, therefore, there might be better and worse partitions with respect to their ability to make learning within partition more sample efficient and quicker convergence to optimal policy. Learning optimal partitions might be a unsupervised learning problem that might also be influence by the types of perturbations that different partitions might suffer. Partitioning (or clustering) might be guided by types of usage (operational profile), types of failures (the corresponding propagations), and types of perturbations (changes in both usage and failure propagation).

jocodeone commented 2 years ago

Is this future work?

Could you elaborate more on the partitioning? How do you identify the types of usage? You could also give an example of this?