Stochastic Extragradient with Flip-Flop Shuffling & Anchoring: Provable Improvements

Authors: Jiseok Chae, Chulhee Yun, Donghwan Kim

NeurIPS 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We consider randomly generated quadratic problems... We ran the experiment on 5 random instances... results are plotted in Figure 1. ... Numerical computations are done using Num Py [24] and Sci Py [52], and the plots are drawn using Matplotlib [26].
Researcher Affiliation Academia Jiseok Chae Department of Mathematical Sciences KAIST Daejeon, Republic of Korea EMAIL Chulhee Yun Kim Jaechul Graduate School of AI KAIST Seoul, Republic of Korea EMAIL Donghwan Kim Department of Mathematical Sciences KAIST Daejeon, Republic of Korea EMAIL
Pseudocode Yes We present the pseudocode of the algorithms we consider in this paper in Algorithms 2, 3 and 4, with the pseudocode of the with-replacement stochastic methods in Algorithm 1. (Appendix A)
Open Source Code No The paper states 'We have also submitted the exact code that we used for our experiments as a supplemental material' in the Neur IPS Paper Checklist section, which is outside the main paper content. The main paper body does not contain an explicit statement or link to open-source code.
Open Datasets No We consider randomly generated quadratic problems of the form min x Rdx max y Rdy 1/n Pn i=1 fi(x, y). ... For an experiment for the monotone case, the random components are sampled as follows. We choose Bi so that each element is an i.i.d. sample from a uniform distribution over the interval [0, 1]... We repeat the exact same procedure for Ci as well.
Dataset Splits No We ran the experiment on 5 random instances of (13) with the stepsizes scheduled as ηk = η0/(1+k/10)0.34 where η0 = min{0.01, 1/L} for SEG-FFA, and αk = βk = ηk for SEG-US, SEG-RR, and SEG-FF.
Hardware Specification No The paper mentions 'Numerical computations are done using Num Py [24] and Sci Py [52], and the plots are drawn using Matplotlib [26]' but does not provide specific details about the hardware (e.g., CPU/GPU models, memory) used for the experiments.
Software Dependencies No Numerical computations are done using Num Py [24] and Sci Py [52], and the plots are drawn using Matplotlib [26].
Experiment Setup Yes We choose dx = dy = 20 and n = 40 for all the experiments. ... The stepsizes scheduled as ηk = η0/(1+k/10)0.34 where η0 = min{0.01, 1/L} for SEG-FFA, and αk = βk = ηk for SEG-US, SEG-RR, and SEG-FF.