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 [1].
Decentralized Projected Riemannian Stochastic Recursive Momentum Method for Nonconvex Optimization
Authors: Kangkang Deng, Jiang Hu
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Furthermore, we demonstrate the effectiveness of our proposed methods compared to state-of-the-art ones through numerical experiments on principal component analysis problems and low-rank matrix completion. |
| Researcher Affiliation | Academia | Kangkang Deng1, Jiang Hu2* 1Department of Mathematics, National University of Defense Technology, Changsha, 410073, China 2Department of Mathematics, University of California, Berkeley, CA 94720, US EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1: The DPRSRM for solving (1) Require: Initial point x0 = x0 M, s 1 = d 1 = 0, α, τ. 1: Sample ξi,0, let si,0 = di,0 = gradfi(xi,0, ξi,0). 2: vi,0 = PTxi,0M(si,0). 3: xi,1 = PM(Pn j=1 Wijxj,0 αvi,0). 4: for k = 1, 2, , K do 5: Update stochastic gradient estimator qi,k via (8) 6: Update the clipped gradient estimator di,k via (10). 7: Update Riemannian gradient tracking si,k via (11). 8: Project onto tangent space: vi,k = PTxi,k M(si,k). 9: Update new iterate xi,k+1 via (12). 10: end for |
| Open Source Code | No | The paper does not contain any explicit statements about the release of source code, nor does it provide links to a code repository. |
| Open Datasets | Yes | We also conduct numerical tests on the Mnist dataset (Le Cun 1998). The training images consist of 60000 handwritten images of size 32 32 and are used to generate Ai s. ... The MNIST database of handwritten digits. http://yann. lecun. com/exdb/mnist/. |
| Dataset Splits | No | For the Mnist dataset, the paper states: "We first normalize the data matrix by dividing 255 and randomly split the data into n = 8 nodes with equal cardinality." This describes how data is distributed among nodes, but it does not specify standard training, validation, and test splits (e.g., percentages or counts) for model evaluation. The synthetic dataset also lacks such details. |
| Hardware Specification | No | The paper makes general statements about computational resources in the introduction ("restricted computational resources") but does not provide specific hardware details (e.g., GPU models, CPU types, memory) used for running the experiments described in the "Numerical Experiments" section. |
| Software Dependencies | No | The paper describes the algorithms and experiments but does not provide specific version numbers for any software, libraries, or frameworks used (e.g., Python, PyTorch, TensorFlow, etc.). |
| Experiment Setup | Yes | In our study, we set the parameters as follows: m1 = . . . = mn = 1000, d = 10, and r = 5. ... The parameters γ and n are set to 0.8 and 8, respectively. We employ fixed step sizes for all algorithms. The step size is set to α = ˆβ/K with K being the maximal number of iterations. The grid search is utilized to find the best ˆβ for each algorithm. The momentum parameter is chosen as τ = 0.999. The batch size in each node is set as 10 and the maximum iteration is set K = 2000. The clipping constant is set as B = 108. ... For all algorithms, we use the fixed step sizes α = ˆβ/60000 with a best-chosen ˆβ, batch size 1500 and momemtum parameter τ = 0.999. |