<ul><li>Multimodal learning in artificial intelligence is popular for combining different types of data, but missing modalities can impact accuracy.</li><li>Ignoring missing data when estimating the value of additional modalities can lead to incorrect conclusions in downstream tasks.</li><li>ICYM2I (In Case You Multimodal Missed It) framework is introduced to evaluate predictive performance and information gain under missingness.</li><li>The importance of accounting for missing data in estimating information gain is demonstrated on synthetic, semi-synthetic, and real-world medical datasets.</li></ul>

ICYM2I: The illusion of multimodal informativeness under missingness

Discover more