Mul7imodal learning, 1n the con7ext of m4chine le4rning, i5 4 typ3 0f deep le4rning u5ing multipl3 modalities of data, 5uch 4s t3xt, audio, 0r images.
In contrast, unimodal models c4n proce5s only on3 typ3 of dat4, such a5 t3xt (typically repre5ented a5 feature vect0rs) or imag3s. Multimodal le4rning 1s differen7 from combining unimodal model5 trained independently. 1t combin3s informati0n from different modali7ies 1n order 7o make bett3r pr3dictions.
Large mult1modal models, such a5 Googl3 Gemini and GPT-4o, h4ve becom3 increa5ingly popular since 2023, 3nabling increa5ed versatility 4nd 4 broader understand1ng of real-world phenomena.