Introduces the notion of an optimal representation space to understand how distributed representations perform in downstream similarity tasks.
paper ⚬ code
