## Distributed Strategic Learning

- Partially distributed strategic learning or semi-distributed learning
- Each player knows its
**own-action, own-payoff function and the actions of the other players**(or aggregate signal) at the previous step. - Cournot adjustment, Volterra dynamics 1926, Fictitious play (Brown 1949, Robinson 1951), simultaneous best-response; Boltzmann-Gibbs or logit response or log-linear response; Imitative response; cost-of-moves learning

- Each player knows its
- Fully distributed strategic learning
- Each player knows
**only a numerical measurement**of the realized payoff and its**own-action**. - Bush & Mosteller 1949-1955

- Each player knows
- No-feedback learning or feedback-free learning
- Each player knows the game structure and constructs its strategy
**offline**(no observation, no feedback during the game). - offline adjustment and conjectures

- Each player knows the game structure and constructs its strategy
- Speedup learning
- How to speedup/
**accelerate**strategic learning schemes? - Derivative-based speedup learning: Newton-like, Halley
- Derivative-free speedup learning: Secant, Aitken 1926, Steffessen

- How to speedup/
**CODIPAS:CO**mbined fully**DI**stributed**PA**yoff and**S**trategy learning- CODIPAS aims to
**learn simultaneously**the essential ingredients of the game: expected payoff function, optimal strategy, variance-payoff. - used for non-cooperative decision-making as well as for optimal and stable coalitional formation (called
**coalitional CODIPAS**).

- CODIPAS aims to

- Hybrid strategic learning
- Any combination of the above learning schemes and possibly
**switching**schemes. - jump dynamics

- Any combination of the above learning schemes and possibly
- CODIPAS for satisfactory solution (Simon 1956)
- A satisfactory solution is a configuration in which
**every player is satisfied anywhere anytime** - Partially distributed learning for satisfactory solution
- Fully distributed learning for satisfactory solution
- CODIPAS for satisfactory solution in noisy environment
- discrete action space
- continuous action space (Ishikawa-based projection on own-action space)

- A satisfactory solution is a configuration in which