Tag
论文提出MATCHA方法，通过显式可检查的多对多层映射和特征映射学习两个独立训练LLM之间的激活对齐，实现跨模型的探针和引导向量等干预迁移，在七个模型族的42对组合上验证了有效性。
This paper introduces a label-free method to find a valence axis from nine emotion examples that transfers across text, vision, audio, and brain modalities, achieving competitive sentiment classification with minimal labels.
ELMZip is a novel satellite image compression framework using Extreme Learning Machines (ELM) and domain decomposition to enable efficient onboard neural representation, transmitting only compact output weights to reduce downlink payload while maintaining high reconstruction fidelity.