@@ -202,9 +202,10 @@ The Deep Bayesian Bandits library includes the following algorithms (see the
...
@@ -202,9 +202,10 @@ The Deep Bayesian Bandits library includes the following algorithms (see the
consists in randomly perturbing a point estimate trained by Stochastic
consists in randomly perturbing a point estimate trained by Stochastic
Gradient Descent on the data. The Parameter-Noise algorithm uses a heuristic
Gradient Descent on the data. The Parameter-Noise algorithm uses a heuristic
to control the amount of noise σ<sub>t</sub><sup>2</sup> it adds independently to the
to control the amount of noise σ<sub>t</sub><sup>2</sup> it adds independently to the
parameters representing a neural network: θ<sub>t</sub><sup>'</sup> where .
parameters representing a neural network: θ<sub>t</sub><sup>'</sup> = θ<sub>t</sub> + ε where
After using  for decision making, the following SGD