VOILA: Value-of-Information Guided Fidelity Selection for Cost-Aware Multimodal Question Answering

Rahul Atul Bhope, K. R. Jayaram, Vinod Muthusamy +3

Despite significant costs from retrieving and processing high-fidelity visual inputs, most multimodal vision-language systems operate at fixed fidelity levels. We introduce VOILA,…