Robust Neural Tucker Factorization with Bias Correction and Adaptive Initialization
Researchers have developed a new method called Kabin (Kaiming initialization and bias correction) to improve the accuracy of Neural Tucker factorization for high-dimensional incomplete tensor completion. This technique addresses issues with parameter initialization and output mapping bias that can hinder performance in conventional linear reconstruction frameworks. Experiments on real-world datasets demonstrate that Kabin offers superior performance compared to the original NeuTucF model with minimal added computational cost. AI
IMPACT Improves accuracy in modeling complex spatio-temporal data, potentially benefiting applications in traffic and climate science.