Interpretable lightweight transformer via unrolling of learned graph smoothness priors

Published in Submitted for Advances in Neural Information Processing Systems (NeurIPS 2025), 2024