Statistical Dimension

For a time the Eluder dimension has been used to provide bounds for optimistic algorithms in function approximation regimes. We introduce the dissimilarity dimension that provides us with tighter bounds.