Related Experiment Video
Updated: Sep 7, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.1K
Applying Meta-Learning and Iso Principle for Development of EEG-Based Emotion Induction System
Kana Miyamoto1,2, Hiroki Tanaka1,2, Satoshi Nakamura1,2
1Division of Information Science, Nara Institute of Science and Technology, Nara, Japan.
Frontiers in Digital Health
|June 23, 2022
Summary
This study developed personalized music generation using meta-learning for faster emotion prediction from electroencephalography (EEG) data. The new system effectively induces target emotions by adapting music to the listener
Area of Science:
- Neuroscience
- Music Technology
- Affective Computing
Background:
- Music is a common tool for emotion induction, but individual responses vary, necessitating personalized approaches.
- Previous personalized music generation systems relied on extensive electroencephalography (EEG) data for training emotion prediction models.
- Existing systems did not account for a listener's pre-existing emotional state when generating music.
Purpose of the Study:
- To address the need for personalized music generation with reduced EEG data requirements.
- To improve music-based emotion induction by incorporating the listener's baseline emotions.
- To evaluate a novel emotion prediction model using meta-learning and a music generation system adapting to listener emotions.
Main Methods:
- Trained emotion prediction models using a small amount of EEG data with a meta-learning approach, comparing it to other training methods.
- Developed a music generation system incorporating an iso-principle to gradually transition music from the listener's current emotions towards a target emotion.
- Assessed the performance of emotion prediction models based on Root Mean Square Error (RMSE) and evaluated the effectiveness of the music generation systems in emotion induction.
Main Results:
- Emotion prediction using meta-learning achieved the lowest RMSE compared to two other training methods (p < 0.016).
- Both the iso-principle-based music generation system and the conventional system demonstrated significantly more effective emotion induction than non-personalized music (p < 0.016).
Conclusions:
- Meta-learning enables accurate emotion prediction from limited EEG data, overcoming a major limitation of previous systems.
- The developed music generation system, adapting to listener emotions via the iso-principle, successfully induces target emotions.
- Personalized music generation, considering both EEG-based emotion prediction and baseline emotional states, offers a promising avenue for affective computing and therapeutic applications.
