Cohort-Based Active Modality Acquisition

Researchers have introduced Cohort-based Active Modality Acquisition (CAMA), a novel framework designed to optimize multimodal machine learning in resource-constrained environments. Addressing the challenge of missing or costly data modalities, CAMA focuses on test-time, cohort-level acquisition rather than traditional per-sample methods. The study proposes imputation-based strategies that estimate the expected utility of acquiring additional data, outperforming entropy-based guidance and random selection in experiments involving up to fifteen modalities. To demonstrate real-world applicability and scalability, the authors applied CAMA to guide the acquisition of proteomics data for disease prediction within the UK Biobank, a large prospective cohort. This approach enables more effective resource allocation by prioritizing samples where additional modality acquisition yields the highest predictive value. The work highlights significant advancements in handling incomplete multimodal datasets, offering a practical solution for improving model performance in healthcare and other fields where data collection is expensive or limited.