Title: Spectra-Guided Neural Tucker Factorization

Abstract: This study introduces Spectra-Guided Neural Tucker Factorization (SG-NTF), a novel approach designed for High-Dimensional and Incomplete (HDI) tensor completion. To overcome the constraints of discrete representations, SG-NTF transforms scalar timestamps into a continuous spectral space, thereby capturing temporal periodicities in an abstract manner. Simultaneously, the model employs a Spatio-Temporal Co-Gating (STCG) mechanism that utilizes multiplicative modulation on spatiotemporal contexts to explicitly filter latent interactions. Experimental results on real-world HDI tensors demonstrate that SG-NTF achieves competitive completion accuracy while maintaining high parameter efficiency.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC