LUMIA: Linear Probing for Unimodal and MultiModal Membership Inference Attacks Leveraging Internal LLM States | Publicación