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..
Guarantees for Alternating Least Squares in Overparameterized Tensor Decompositions
Authors: Dionysis Arvanitakis, Vaidehi Srinivas, Aravindan Vijayaraghavan
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our theoretical results on overparameterization are also supported by empirical evaluations in Appendix D. In our experiments we investigate whether this overparameterization factor is also observed in practice, and what the leading constant in the dependence is. |
| Researcher Affiliation | Academia | Dionysis Arvanitakis Department of Computer Science Northwestern University Evanston, IL 60208 EMAIL |
| Pseudocode | Yes | Algorithm 1 Alternating Least Squares (ALS) for order-3 tensor decomposition |
| Open Source Code | Yes | We provide python code to run both experimental setups as part of the supplementary material. |
| Open Datasets | No | For each trial we generate 3 random n r factor matrices (each entry is an independent Gaussian) to make up our ground truth tensor. |
| Dataset Splits | No | For each trial we generate 3 random n r factor matrices (each entry is an independent Gaussian) to make up our ground truth tensor. |
| Hardware Specification | Yes | We used a regular laptop. |
| Software Dependencies | No | To evaluate the parallel-update version of ALS (Algorithm 1), we implemented a non-optimized version using the scipy least squares solver. ... To evaluate standard ALS, we used the parafac method from the Tensor Ly library [KPAP19], which provides an optimized version of the standard (sequential) ALS method. |
| Experiment Setup | Yes | For all n = 200 and n = 500 we set the maximum number of iterations to be 20, due to computational constraints. ... For n = 500 we set the maximum number of iterations to be 100, and for n = 1000 we set the maximum number of iterations to be 20... We initialize the factors of our model to be fully random n k matrices. For each trial we generate 3 random n r factor matrices (each entry is an independent Gaussian) to make up our ground truth tensor. |