On Adversarial Vulnerability of Vision-Language Models through the Lens of Intermediate Spectral Subspaces | Chethan Krishnamurthy Ramanaik et al. | ResearchPod