Generalized Smooth Variational Inequalities: Methods with Adaptive Stepsizes

Authors: Daniil Vankov, Angelia Nedich, Lalitha Sankar

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present numerical experiments that support our theoretical guarantees and highlight the efficiency of proposed adaptive stepsizes.
Researcher Affiliation Academia 1Department of Electrical and Computer Engineering, Arizona State University Tempe, Arizona, USA.
Pseudocode Yes Algorithm 1 Korpelevich Method with Backtracking
Open Source Code No The paper does not contain any statements about making its source code publicly available, nor does it provide any links to a code repository.
Open Datasets No We present the experiments on training GAN for the 2D Ring dataset, a mixture of 8 equal-prior Gaussian distributions, with mean cos(2πi/8), sin(2πi/8) for i {1, . . . , 8} and variance 10 4. The paper describes how the dataset was generated but does not provide access information (link, DOI, formal citation) to a publicly available version of this dataset.
Dataset Splits No The paper mentions running methods for 100 epochs with a batch size of 128 but does not provide explicit training, validation, or test dataset splits (e.g., percentages, sample counts, or references to predefined splits).
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory specifications, or types of computing resources used for the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software libraries, frameworks, or programming languages used in the experiments.
Experiment Setup Yes We set parameters of the problem to be {(α 0.090., p = 2.1), (α 0.66, p = 4.0), (α 0.86, p = 8.0)}. We run methods for 100 epochs with a batch size of 128. γk = 10 3 for methods without clipping, and γk = 10 3 min{1, 1 F (hk) } for methods with clipping. Korpelevich method with clipping and backtracking (Algorithm 1) with β = 1.0 and q = 0.75. γk = βk min{1, 1 F (hk) , 1 ( uk hk 1 +1)α/(1 α) }, with the same βk = a b+k for all methods, where a = b = 100.