Entanglement and differentiable information gain maximization