Hei Shing Cheung

Eric Xie, Hei Shing Cheung

Surface electromyography (sEMG) enables continuous hand pose estimation on wearable devices, but models trained on multi-user corpora degrade on unseen individuals due to inter-user variability in anatomy and electrode placement. We propose REACT, a lightweight conditioning framework that personalizes a frozen pretrained EMG-to-pose backbone at inference time using only a handful of calibration recordings. REACT learns a compact user embedding from calibration data and applies Feature-wise Linear Modulation (FiLM) to adapt the shared encoder's feature space, requiring no gradient updates at deployment. On the large-scale EMG2POSE benchmark, REACT improves over the state-of-the-art baseline across all three generalization splits in both regression and tracking modes, reducing angular error by up to 3.9% with minimal parameter overhead and under 45 seconds of per-user calibration.

Hei Shing Cheung, Boya Zhang, Jonathan H. Chan

We present a lightweight latent diffusion model for vocal-conditioned musical accompaniment generation that addresses critical limitations in existing music AI systems. Our approach introduces a novel soft alignment attention mechanism that adaptively combines local and global temporal dependencies based on diffusion timesteps, enabling efficient capture of multi-scale musical structure. Operating in the compressed latent space of a pre-trained variational autoencoder, the model achieves a 220 times parameter reduction compared to state-of-the-art systems while delivering 52 times faster inference. Experimental evaluation demonstrates competitive performance with only 15M parameters, outperforming OpenAI Jukebox in production quality and content unity while maintaining reasonable musical coherence. The ultra-lightweight architecture enables real-time deployment on consumer hardware, making AI-assisted music creation accessible for interactive applications and resource-constrained environments.