Spectra-Guided Neural Tucker Factorization
Researchers have introduced Spectra-Guided Neural Tucker Factorization (SG-NTF), a novel method for completing high-dimensional and incomplete tensors. This approach maps timestamps into a continuous spectral space to capture temporal patterns and employs a Spatio-Temporal Co-Gating mechanism to filter latent interactions. Evaluations indicate that SG-NTF achieves competitive accuracy while maintaining parameter efficiency. AI
IMPACT Introduces a novel technique for improving tensor completion accuracy and efficiency in high-dimensional datasets.