Minibatch and Momentum Model-based Methods for Stochastic Weakly Convex Optimization

Authors: Qi Deng, Wenzhi Gao

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we examine the empirical performance of our proposed methods through experiments on the problem of robust phase retrieval.
Researcher Affiliation Academia Qi Deng1 Wenzhi Gao2 School of Information Management and Engineering Shanghai University of Finance and Economics 1qideng@sufe.edu.cn 2gwz@163.shufe.edu.cn
Pseudocode Yes Algorithm 1 Stochastic Model-based Method with Minibatches (SMOD) and Algorithm 2 Stochastic Extrapolated Model-Based Method (SEMOD)
Open Source Code Yes Code is supplied in the supplemental materials
Open Datasets Yes We consider zipcode, a dataset of 16x16 handwritten digits collected from [21].
Dataset Splits No The paper describes how synthetic data is generated and specific instances of the zipcode dataset are used for 'testing cases', but it does not specify any explicit training, validation, or test dataset splits (e.g., percentages, counts, or predefined splits) for reproducibility.
Hardware Specification No Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [No] The main goal of experiments is to demonstrate our theoretical foundings, thereby only showing the iteration complexity of algorithms.
Software Dependencies No The paper does not provide specific software dependencies or library version numbers (e.g., Python, PyTorch, TensorFlow versions) used for implementation or experimentation.
Experiment Setup Yes 2) Initial point. We set the initial point x1(= x0) N(0, Id) for synthetic data and x1 = x + N(0, Id) for zipcode; 3) Stopping criterion. We set the stopping criterion to be f(xk) 1.5 ˆf... 3) Stepsize. We set the parameter γ = α 1 0 p K/m where m is the batch size; For synthetic dataset, we test 10 evenly spaced α0 values in range [10 1, 102] on logarithmic scale, and for zipcode dataset we set such range of α0 to [101, 103]; 4) Maximum iteration. We set the maximum number of epochs to be 200 and 400 respectively for minibatch and momentum related tests; 5) Batch size. We take minibatch size m from the range {1, 4, 8, 16, 32, 64};