Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Non-Convex Tensor Recovery from Tube-Wise Sensing
Authors: Tongle Wu, Ying Sun
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results validate the effectiveness of GD in solving the proposed local TCS model. Simulation result shown in Fig. 2 corroborates our findings. 5 Experimental results. We validate our theoretical findings on synthetic tensors (real-data results appear in Appendix C). |
| Researcher Affiliation | Academia | Tongle Wu The Pennsylvania State University EMAIL Ying Sun The Pennsylvania State University EMAIL |
| Pseudocode | Yes | Algorithm 1 Gradient descent with spectral initialization for minimizing (6). Input: Operator Z and measurements Y Rn1 n2 m, rank r, step size η, total iteration Tmax. Spectral Initialization: 1: X 0 := Zc(Y) m 2: [A0, B0] = T-SVD-Spec(X 0, r) Gradient Descent: 3: for t = 0, . . . , Tmax 1 do 4: Update At+1 = At η AL(At, Bt), Bt+1 = Bt η BL(At, Bt) based on (7), (8). 5: end for Output: Recover tensor X Tmax = ATmax (BTmax) . Algorithm 2 Truncated t-SVD based spectral initialization: [A, B] = T-SVD-Spec(X, r) Input: X Rn1 n2 n3, r. |
| Open Source Code | Yes | We include anonymized code and run scripts with instructions. |
| Open Datasets | No | We validate our theoretical findings on synthetic tensors (real-data results appear in Appendix C). ... We test the proposed local TCS model on the highway video sequence, which is reshaped into third-order tensor with size 144 150 176. |
| Dataset Splits | No | For each (n, m) pair, we simulate 10 test trials and assess a trial to be successful if the recovered ˆX satisfies ˆX X F / X F 0.001. |
| Hardware Specification | Yes | All experiments were conducted using MATLAB R2022a on a Windows system equipped with a 12th Gen Intel(R) Core(TM) i7-12700 CPU at 2.10 GHz and 16.0 GB of RAM. |
| Software Dependencies | Yes | All experiments were conducted using MATLAB R2022a on a Windows system equipped with a 12th Gen Intel(R) Core(TM) i7-12700 CPU at 2.10 GHz and 16.0 GB of RAM. |
| Experiment Setup | Yes | To validate convergence, we set (n1, n2, n3) = (100, 100, 50), tubal rank r = 5, and condition number κ = 1 for ground-truth tensor X ... We set the η = 0.2 and tube-wise sample complexity m = 10. ... For α = 1, we select the largest η = 0.0002 to guarantee convergence. For α = 0.001, the η = 0.2 is the same as in the Fig. 1(a) setting. ... We fix the tube dimension as n3 = 50 and consider the two settings: (1) r = 5; (2) r = 10. ... For the real-data experiments, we set r = 3 in our algorithm. For this r, we select the two tube-wise sample sizes as m = 20, 30... We have to select small step sizes (0.00005) to guarantee the convergence |