The final is comprehensive
Refer to the previous review sheets.
Multi-agent Learning
- Satisficing
- Aspiration update
- Algorithm
- Relaxation search
- The multi-agent social dilemma (MASD)
- Definitions of security and efficiency
- Properties of the Satisficing algorithm
- Security against exploiting opponents (and switching to a
pure strategy maximin if needed)
- Conditions of Pareto efficiency
- Critical events
- When is the S-algorithm efficient in self-play
- SALT
- Learning aspiration trembles
- Security
- Empricial trends in efficiency in self-play
Multi-agent Learning