What Makes Multi-modal Learning Better than Single (Provably)

Abstract

The world provides us with data of multiple modalities. Intuitively, models fusing data from different modalities outperform unimodal models, since more information is aggregated. Recently, joining the success of deep learning, there is an influential line of work on deep multimodal learning, which has remarkable empirical results on various applications. However, theoretical justifications in this field are notably lacking. Can multimodal provably perform better than unimodal? In this paper, we answer this question under a most popular multimodal learning framework, which firstly encodes features from different modalities into a common latent space and seamlessly maps the latent representations into the task space. We prove that learning with multiple modalities achieves a smaller population risk than only using its subset of modalities. The main intuition is that the former has more accurate estimate of the latent space representation. To the best of our knowledge, this is the first theoretical treatment to capture important qualitative phenomena observed in real multimodal applications. Combining with experiment results, we show that multimodal learning does possess an appealing formal guarantee.

Multimodal music emotion recognition in Indonesian songs based on CNN-LSTM, XLNet transformers

“…The fusion method achieves much higher results than audio classifiers or text classifiers. As in the [36] study, multi-modality outperforms single since the former has access to a better latent space From Tables 5 and 6, it can be concluded that MER text experiments consistently achieve higher accuracy than MER audio. This happens because the complexity of Mel-spectrogram data from audio is much higher.…”

Mpox-AISM: AI-mediated super monitoring for mpox and like-mpox

“…Given that mpox is an infectious disease, it is recommended that primary healthcare providers combine epidemiological survey results when using the mpox-AISM app to make more accurate diagnoses and guide patients to refer them to professional institutions. 44 In addition, considering that models based on multimodal inputs have better learning ability compared to single modal input models, 45 we are planning to develop a multimodal model that combines rash images and epidemiological survey results, aiming to improve the diagnostic accuracy and stability of mpox-AISM.…”

Exploring the influence of transformer-based multimodal modeling on clinicians’ diagnosis of skin diseases: A quantitative analysis

“…Additionally, the multimodal model closely aligns with the diagnostic behavior of clinicians and offers better interpretability. 36 In this context, we utilized the soft voting method in ensemble learning to construct the multimodal DL model for verifying its diagnostic assistance to clinicians of varying expertise levels during subsequent testing. We selected the voting method because it closely approximates the decision-making process of clinicians, prioritizing more important image types.…”