wip

tutorials/BayesianOptimizationDesign.md

@@ -19,7 +19,7 @@ In this tutorial, we look at the design of spidr's Bayesian optimization functio
 
 ## A Bayesian optimization refresher
 
-Techniques such as stochastic gradient descent excel at optimizing functions whose value and gradient are cheap to evaluate. But when function evaluations are expensive and the gradient is unknown, these techniques can be infeasible. In Bayesian optimization we approach such problems by placing a probabilistic model over historical function evaluations, and optimizing over this surrogate model instead. Model evaluations are cheap, and the gradients are known. A probabilistic model captures the uncertainty in the objective function as a posterior _distribution_, rather than a point. Bayesian optimization algorithms can make use of this predictive distribution in their own way in order to find the objective optimum, for example choosing a particular balance of exploration and exploitation. For a much more in-depth discussion, we recommend [this tutorial](https://arxiv.org/abs/1807.02811).
+Techniques such as stochastic gradient descent excel at optimizing functions whose value and gradient are cheap to evaluate. But when function evaluations are expensive and the gradient is unknown, these techniques can be infeasible. In Bayesian optimization we approach such problems by placing a probabilistic model over historical function evaluations, and optimizing over this surrogate model instead. Model evaluations are cheap, and the gradients are known. A probabilistic model captures the uncertainty in the objective function as a posterior _distribution_, rather than a point. Bayesian optimization algorithms can make use of this predictive distribution in their own way in order to find the objective optimum, for example choosing a particular balance of exploration and exploitation. For an in-depth discussion of Bayesian optimization, we recommend [this tutorial](https://arxiv.org/abs/1807.02811).
 
 ## Bayesian optimization design