Regularization

Demonstrates that compressing LLM embeddings can improve performance on noisy regression tasks like financial prediction by reducing overfitting, suggesting compression acts as a regularization mechanism.