Robust Low-Tubal-Rank Tensor Completion under Cross-Concentrated Sampling (Q7284591)

We propose R-ItCUR, a tensor-native algorithm for robust low-tubal-rank tensor completion from partially observed tensors contaminated by sparse outliers via t-CCS. We apply adaptive blockwise Welsch correction and projected blockwise gradient descent to update the low-rank component directly on the sampled cross, avoiding full tensor reconstruction and achieving substantial memory and computational savings.