Table 7 Comparison between learnable and sinusoidal positional encodings on UTD-MHAD and MM-Fit datasets.

From: A tiny inertial transformer for human activity recognition via multimodal knowledge distillation and explainable AI

Positional encoding type

UTD-MHAD accuracy (%)

MM-fit accuracy (%)

Sinusoidal encoding (fixed)

97.89

97.62

Learnable positional embedding (ours)

98.71

98.55

Back to article page