Enhancing Model Representations through Heterogeneous Self-Supervised Learning

Heterogeneous self-supervised learning enables a base model to learn complementary characteristics from an auxiliary head with a heterogeneous architecture, enhancing the representation quality of the base model.