SMOSE is a novel method for training interpretable controllers in reinforcement learning for continuous control tasks. It uses a Sparse Mixture of Shallow Experts architecture to combine interpretable decision-makers and a router for task assignment. SMOSE aims to improve transparency and trust in automated systems by providing interpretable policies.
Sparse Mixture of Shallow Experts, decision trees
Mixture-of-Experts architecture
MuJoCo benchmark environments
Outperforms interpretable baselines
Not specified
No
Yes
Interpretable controllers, sparse mixture architecture, improved transparency
Yes
Not specified
Not specified
Not specified
Not specified
Not specified
Not specified
No
Not specified
Not specified
Not specified
Not specified
Not specified
Interpretable decision-making
Not specified
Not specified
Robotics, Automation
Interpretable reinforcement learning, automated decision-making
Not specified
Not specified
High
Not specified
Not specified
Not specified
No
Not specified
Not specified
No
Not specified
Not specified
Not specified
Not specified
Not specified
No
Not specified
Not specified
0.00
Not specified
Not specified
01/01/1970
01/01/1970
Not specified
Not specified
Yes