Stochastic Frank-Wolfe for Constrained Finite-Sum Minimization

Authors: Geoffrey Negiar, Gideon Dresdner, Alicia Tsai, Laurent El Ghaoui, Francesco Locatello, Robert Freund, Fabian Pedregosa

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

Reproducibility Variable Result LLM Response
Research Type Experimental Benchmarks on several datasets highlight different regimes in which the proposed method exhibits a faster empirical convergence than related methods. Finally, we provide an implementation of all considered methods in an open-source package. ... These experiments show that the proposed method converges at least as fast as previous work, and notably faster on several such instances.
Researcher Affiliation Collaboration 1Berkeley AI Research, University of California, Berkeley, CA, USA 2Department of Computer Science, ETH Zurich, Switzerland 3Sum Up Analytics 4Max-Planck Institute for Intelligent Systems, T ubingen, Germany 5MIT Sloan School of Management 6Google Research.
Pseudocode Yes Algorithm 1 Stochastic Frank-Wolfe 1: Initialization: w0 C, α0 Rn, r0 = X α0 2: for t = 1, 2, . . . , do 3: Sample i {1, . . . , n} uniformly at random. 4: Update α(i) t = 1 nf i(x i wt 1) 5: Update α(j) t = αj t 1, j = i 6: rt = rt 1 + (α(i) t α(i) t 1)xi 7: st = LMO(rt) 8: wt = wt 1 + 2 t+2(st wt 1) 9: end for
Open Source Code Yes Our implementation is available in the C-OPT package.1 https://github.com/openopt/copt
Open Datasets Yes We consider ℓ1 constrained logistic regression problems on the BREAST CANCER and RCV1 datasets, and an ℓ1 constrained least squares regression problem on the CALIFORNIA HOUSING dataset, all from the UCI dataset repository (Dua & Graff, 2017).
Dataset Splits No The paper mentions using specific datasets but does not provide explicit details about train, validation, and test splits (e.g., percentages or sample counts).
Hardware Specification No The paper describes the experimental setup and datasets but does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions that its implementation is available in the C-OPT package but does not list specific software dependencies with version numbers (e.g., Python, PyTorch, or other libraries with their versions).
Experiment Setup Yes We use batches using 1% of the dataset at each step, following Lu & Freund (2018). Within a batch, data points are sampled without replacement.