Efficient Projection-free Algorithms for Saddle Point Problems

Authors: Cheng Chen, Luo Luo, Weinan Zhang, Yong Yu

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate the effectiveness of our algorithms and verify our theoretical guarantees. We also conduct experiments on several real-world data sets for robust optimization problem to validate our theoretical analysis. The empirical results show that the proposed methods outperform previous projection-free and projection-based methods when the feasible set is complicated. In this section, we empirically evaluate the performance of our methods on the robust multiclass classification problem introduced in Section 2.3.
Researcher Affiliation Academia Cheng Chen1 Luo Luo2 Weinan Zhang1 Yong Yu1 1Shanghai Jiao Tong University 2The Hong Kong University of Science and Technology
Pseudocode Yes Algorithm 1 CGS Method for strongly convex functions, Algorithm 2 Procedure q+=Cnd G(r, q, β, η, Ω), Algorithm 3 Mirror-Prox Conditional Gradient Sliding, Algorithm 4 Procedure (x R, y R, v R)=Prox-step(f, x0, y0, z, v, γ, α, ζ, ϵ), Algorithm 5 Inexact STORC (i STORC), Algorithm 6 Mirror-Prox Stochastic Conditional Gradient Sliding, Algorithm 7 Procedure (x R, y R, v R)=Stochastic-Prox-step(f, x0, y0, z, v, γ, α, ζ, P, ϵ)
Open Source Code No The paper does not provide any concrete access to source code for the methodology described.
Open Datasets Yes We conduct experiments on three real-world data sets from the LIBSVM repository2: rcv1 (n = 15, 564, d = 47, 236, h = 53), sector (n = 6, 412, d = 55, 197, h = 105) and news20 (n = 15, 935, d = 62, 061, h = 20). 2https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/
Dataset Splits No The paper mentions using 'real-world data sets' but does not specify the train/validation/test splits, percentages, or methodology for data partitioning.
Hardware Specification No The paper does not provide any specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library or solver names with version numbers, needed to replicate the experiment.
Experiment Setup Yes We implement the mini-batch version of SVRE with batch size 100. The learning rate of SVRE is searched in {10 1, 10 2, . . . , 10 6}. On the other hand, the parameters of projection-free methods follows what the theory suggests.