TMAK; Emotion Estimation, Mental Health Application, Information Recommendation - Application for Stress Estimation Model and Mild Cognitive Impairment Detection

Abstract

 TMAK (Trustworthy Multimodal Affective Intelligence and Knowledge Engineering Laboratory) is advancing research on several key themes. First, we are researching methods to estimate human emotions from multimodal information (audio, language, images) and apply this to support the diagnosis of mental disorders, mental health conditions, and dementia. Second, we are researching techniques to analyze interest and reputation information from diverse web reviews and social media posts and utilize it for information recommendation. Third, we are developing empathetic dialogue systems by leveraging the rapidly advancing large-scale multimodal language models. All these research areas require feature extraction from large-scale data, making 
big data analysis platforms indispensable. Our laboratory is also advancing research on lightweight AI models, developing language models, emotion estimation, and stress estimation algorithms capable of running on local edge devices. This presentation will introduce past research examples, outline solutions using our proprietary emotion estimation technology based on multi-stage fine-tuning, and discuss future prospects such as cognitive function prediction.