LESSL: Can LEGO sampling and collaborative optimization contribute to self-supervised learning?