A paper authored by Masaki Matsuoka (second-year master's student) as the first author has been accepted for publication in Expert Systems with Applications.
Masaki Matsuoka, Xinyue Niu, and Akira Furui
Guided Disentangled Representation Learning for Multi-Factor EMG Analysis
Expert Systems with Applications (accepted).
This study proposes a variational autoencoder (VAE)-based framework that learns disentangled representations of EMG signals by structuring the latent space around multiple generative factors—force level, forearm orientation, and movement class. By combining factor-specific predictors with adversarial gradient reversal, the method aligns each latent subspace with its designated factor while suppressing information leakage across subspaces. Experiments on two public EMG datasets demonstrate effective disentanglement, consistent improvements in multi-task recognition, and the potential for controllable EMG signal generation via latent space manipulation.
Congratulations!