Prediction With Expert Advice
Prediction with expert advice is a subfield of competitive on-line prediction in which the strategies in the benchmark class are free agents. When understood in a wide sense (as in Cesa-Bianchi and Lugosi 2006), prediction with expert advice is synonymous with competitive on-line prediction.
When strategies are free agents, or experts, the protocol of a game of prediction with expert advice is
Players: Experts, Forecaster, Reality
Protocol:
Initialize
⚠ $L_0=0, L_0(k)=0, k=1,2,\dots$
FOR
⚠ $t=1,2,\dots$
:
Experts announce
⚠ $\gamma^k\in\Gamma, k=1,2,\dots$
Forecaster announces
⚠ $\gamma\in\Gamma$
Reality announces
⚠ $\omega\in\Omega$
Update cumulative loss
⚠ $L_t=L_{t-1} + \lambda(\omega,\gamma)$
, ⚠ $L_t(k)=L_{t-1}(k) + \lambda(\omega,\gamma_k), k=1,2,\dots$
END
These are some open problems:
- Find the n-experts separation curve.
Bibliography
- Nicolo Cesa-Bianchi and Gabor Lugosi. Prediction, Learning, and Games. Cambridge: Cambridge University Press, 2006.