Nikita S. Serov

Soft Non-diagonality Penalty Enables Latent Space-Level Interpretability of Parameter-Efficient Peptide LM at No Performance Cost.

This work introduces a soft weight matrix nondiagonality penalty and a one-hot encoded sequence clustering-based contrastive loss to develop compact models with interpretable embeddings while maintaining competitive performance, and demonstrates that domain-tailored architectural design can yield highly parameter-effic...