Accelerating CP Decomposition via WHT Sampling

The patent applies partial action by using random sampling to select only a subset of data points for each iteration of the CP decomposition process. Instead of processing the entire tensor data, the method samples a portion of the data to compute factor matrices, which reduces the computational burden while still achieving acceptable decomposition results. This is implemented through the sampling module that randomly selects indices from the tensor dimensions.

The patent segments the large-scale tensor decomposition problem into smaller, manageable sub-problems by dividing the data into sampled subsets. Each iteration works on a segmented portion of the data rather than the complete dataset, allowing the decomposition to proceed with reduced computational resources. The factor matrices are updated incrementally based on these segmented computations.

The patent reduces execution time by performing partial computations through random sampling. Instead of iterating over all data points in each ALS iteration, the method processes only a sampled subset, significantly decreasing the time required for each update step while maintaining the alternating least squares optimization framework for accuracy.

The patent performs preliminary random sampling to generate initial factor matrices before proceeding with the full decomposition process. This preliminary action creates a starting point that accelerates convergence, reducing the number of iterations needed to achieve accurate results and thereby decreasing overall execution time.