Modified Frank Wolfe in Probability Space

Authors: Carson Kent, Jiajin Li, Jose Blanchet, Peter W Glynn

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

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our algorithm to a range of functionals arising from applications in nonparametric estimation. All simulations are implemented using Python 3.8 on a high performance computing server running Ubuntu 18.04 with a Gen10 Quad Intel(R) Xeon(R) Platinum 8268 CPU @ 2.90GHz processor.
Researcher Affiliation Academia Carson Kent Stanford University crkent@stanford.edu Jose Blanchet Stanford University jose.blanchet@stanford.edu Jiajin Li The Chinese University of Hong Kong gerrili1996@gmail.com Peter Glynn Stanford University glynn@stanford.edu
Pseudocode Yes Algorithm 1 Modified Frank Wolfe for (3)
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results? [Yes] See the supplemental material
Open Datasets No The paper mentions using a "2D Gaussian mixture" and "50 data points, Yi, sampled from a mixture of 7-Gaussians" and refers to "Student-Teacher Network problem" without providing concrete access information (link, DOI, repository, or formal citation with authors/year) for publicly available datasets.
Dataset Splits No The paper refers to a "validation dataset" in Figure 3 and states that "The detailed implementation set up is same as [3, Appendix G]" for the Student-Teacher Network problem. However, this paper does not explicitly provide specific dataset split information (exact percentages, sample counts, or detailed splitting methodology) within its main text.
Hardware Specification Yes All simulations are implemented using Python 3.8 on a high performance computing server running Ubuntu 18.04 with a Gen10 Quad Intel(R) Xeon(R) Platinum 8268 CPU @ 2.90GHz processor.
Software Dependencies Yes All simulations are implemented using Python 3.8
Experiment Setup Yes the number of particles is 200. The bisection method of Appendix C is used with tolerance set to 1e 3. For 50 data points, Yi, sampled from a mixture of 7-Gaussians, 200 particles are used in a non-parametric estimate the µi. the step size δ is 0.5. The number of particle is 200. In Figure 2(a), the tolerance for the ascent method is 1e 3 and the shaded bands show the standard derivation over 10 independent runs with random initializations.