Meta-Learning Strategies through Value Maximization in Neural Networks