BioAutoML-NAS: An End-to-End AutoML Framework for Multimodal Insect Classification via Neural Architecture Search on Large-Scale Biodiversity Data
Researchers have developed BioAutoML-NAS, a novel framework for insect classification that integrates multimodal data, including images and metadata. This system utilizes neural architecture search (NAS) to optimize network structures and a multimodal fusion module to combine visual and biological information. Evaluations on the BIOSCAN-5M and Insects-1M datasets show BioAutoML-NAS significantly outperforms existing transfer learning, transformer, AutoML, and NAS methods in accuracy and efficiency. AI
IMPACT This framework advances multimodal classification techniques, potentially improving biodiversity research and agricultural management through more accurate insect identification.