Presentation information

General Session

General Session » GS-2 Machine learning

[2G1-GS-2d] 機械学習:シーケンシャルデータの処理

Wed. Jun 9, 2021 9:00 AM - 10:40 AM Room G (GS room 2)

座長:林 知樹(名古屋大学)

10:20 AM - 10:40 AM

[2G1-GS-2d-05] An emotion estimation model with domain adaptation using a small amount EEG of target user

〇Shoya Furukawa1, Takuto Sakuma1, Shohey kato1,2 (1. Nagoya Institute of Technology, 2. Frontier Research Institute for Information Science, Nagoya Institute of Technology)

Keywords:Deep Learning, Emotion estimation, Domain Adaptation, Electroencephalogram(EEG)

In recent years, making computers understand the emotions of users is necessary because emotions are an important factor in human communication. Among many methods of recognizing emotions, EEG is widely used because it has high temporal resolution and it is impossible to disguise intentionally. However, it is necessary to acquire new user data and construct the personal Emotion recognition model since EEG varies widely among individuals. The conventional methods lack practicality because it builds a different model for every new user data. Our method builds a model in a single training using new user data. To reduce the number of new user data for training and to relieve the burden of EEG measurement, we adapt existing user data. We absorb the individual differences in EEG between a new user and existing users by domain adaptation using a small number of new user data and construct a model by deep neural networks. From the experiments, we confirmed that the proposed method performs as well as the conventional methods, even though it is built with single training using new user data. Also, experimental results with new user data measured at a later date suggest that repetition of stimuli may cause habituation and fade out target emotion features.

Authentication for paper PDF access

A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.