L. Bai, “Frozen Language Models Are Gradient Coherence Rectifiers in Vision Transformers”, AAAI, vol. 39, no. 2, pp. 1817–1825, Apr. 2025.