Motivating Bilevel Approaches To Filter Learning: A Case Study