A Variational Perspective on High-Resolution ODEs

Authors: Hoomaan Maskan, Konstantinos Zygalakis, Alp Yurtsever

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Several numerical experiments compare and illustrate our stochastic algorithm with state of the art methods.
Researcher Affiliation Academia Hoomaan Maskan Umeå University hoomaan.maskan@umu.se Konstantinos C. Zygalakis University of Edinburgh k.zygalakis@ed.ac.uk Alp Yurtsever Umeå University alp.yurtsever@umu.se
Pseudocode No The paper defines algorithms using mathematical equations (e.g., (NAG), (23), (31), (35)) but does not provide structured pseudocode blocks.
Open Source Code No The paper does not provide any concrete statements about the release of source code or links to a code repository for the described methodology.
Open Datasets Yes Finally, we tackled the non-convex optimization problem of training a CNN with CIFAR10 dataset [Krizhevsky et al., 2009] using the SGD, SVRG, NNAG, and NNAG+SVRG methods.
Dataset Splits No The paper mentions 'training error' and 'validation accuracy' in the context of CIFAR10, but it does not specify the exact percentages or sample counts for training, validation, and test splits used to reproduce the data partitioning.
Hardware Specification Yes We gratefully acknowledge the support of NVIDIA Corporation with the donation of 2 Quadro RTX 6000 GPUs used for this research.
Software Dependencies No The paper does not provide specific software names with version numbers (e.g., 'Python 3.8', 'PyTorch 1.9') for its ancillary software dependencies.
Experiment Setup Yes For NNAG we used sk = 1/(Lk3/4) and β = L/10. All the perturbation was done using i.i.d. Gaussian noise with unit variance, and we conducted 100 Monte Carlo runs. The step-sizes for SGD and SVRG were set as 0.01, and for the NNAG and NNAG+SVRG algorithms we had c = 0.05, β = 1502 and c = 0.001, β = 1002/10, respectively.