Stochastic Weakly Convex Optimization beyond Lipschitz Continuity

Authors: Wenzhi Gao, Qi Deng

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

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments demonstrate the efficiency and robustness of our proposed stepsize policies.
Researcher Affiliation Academia 1Institute for Computational and Mathematical Engineering, Stanford University 2Antai College of Economics and Management, Shanghai Jiao Tong University.
Pseudocode Yes Algorithm 1 Stochastic model-based optimization Input x1 for k = 1, 2,... do Sample data ξk and choose regularization γk > 0 xk+1 = arg min x fxk(x, ξk) + ω(x) + γk 2 x xk 2
Open Source Code No The paper does not provide any statement regarding the release of source code for the methodology described, nor does it include links to a code repository.
Open Datasets No The paper describes a data generation process consistent with a cited work (Deng and Gao, 2021) to create synthetic datasets for experiments, but it does not indicate that the generated dataset or the generation script is publicly available, nor does it use a pre-existing public dataset.
Dataset Splits No The paper describes parameters for generating synthetic data, such as m, n, κ, and pfail, but it does not specify any training, validation, or test dataset splits.
Hardware Specification No The paper does not provide any specific details regarding the hardware used to run the experiments, such as CPU, GPU models, or memory specifications.
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., programming languages, libraries, or frameworks like PyTorch or TensorFlow) used for the experiments.
Experiment Setup Yes Stopping criterion. We run algorithms for 400 epochs (K = 400m). Algorithms stop if f 1.2f(ˆx). Stepsize. We let γk = θ K for vanilla algorithms; γk = θ G( xk ) K for robust stepsize with known growth condition; γk = θ max{Lip(xk, ξ ), α} K for robust stepsize with unknown growth condition. θ [10 2, 101] serves as a hyper-parameter. Clipping. Clipping parameter α is set to 1.0.