On the Pros and Cons of Momentum Encoder in Self-Supervised Visual Representation Learning | Synapse