Multimodal Machine Learning: A Survey and Taxonomy

Tadas Baltrušaitis,Chaitanya Ahuja,Louis-Philippe Morency

2017-08-02

Abstract:Our experience of the world is multimodal - we see objects, hear sounds, feel texture, smell odors, and taste flavors. Modality refers to the way in which something happens or is experienced and a research problem is characterized as multimodal when it includes multiple such modalities. In order for Artificial Intelligence to make progress in understanding the world around us, it needs to be able to interpret such multimodal signals together. Multimodal machine learning aims to build models that can process and relate information from multiple modalities. It is a vibrant multi-disciplinary field of increasing importance and with extraordinary potential. Instead of focusing on specific multimodal applications, this paper surveys the recent advances in multimodal machine learning itself and presents them in a common taxonomy. We go beyond the typical early and late fusion categorization and identify broader challenges that are faced by multimodal machine learning, namely: representation, translation, alignment, fusion, and co-learning. This new taxonomy will enable researchers to better understand the state of the field and identify directions for future research.

Machine Learning

What problem does this paper attempt to address?

This paper primarily explores the research progress and challenges in the field of Multimodal Machine Learning and proposes a new classification system to organize the research outcomes in this area. Its core aim is to address how to enable artificial intelligence systems to understand and interpret data from different sensory inputs (such as vision, hearing, etc.).

Specifically, the research mentioned in the paper aims to address the following five key challenges:
1. **Representation**: How to effectively integrate data from multiple modalities to form complementary and redundant representations.
2. **Translation**: How to perform data conversion or mapping between different modalities.
3. **Alignment**: Identifying direct relationships between two or more modalities.
4. **Fusion**: Combining information from two or more modalities to complete prediction tasks.
5. **Co-learning**: How to transfer knowledge between modalities, including their representations and prediction models.

In addition, the paper reviews some important application areas of multimodal machine learning, including speech recognition, multimedia event detection, emotion recognition, media description, etc., and discusses the specific technical challenges behind these applications. For example, in the field of speech recognition, early research focused on audio-visual speech recognition; while in image description, it is necessary to address the issue of evaluating the quality of generated text descriptions.

In summary, the main goal of this paper is to provide a comprehensive overview and a new classification framework for the field of multimodal machine learning to promote future research and development.