Minding the Gaps for Block Frank-Wolfe Optimization of Structured SVMs

Authors: Anton Osokin, Jean-Baptiste Alayrac, Isabella Lukasewitz, Puneet Dokania, Simon Lacoste-Julien

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we provide an exhaustive empirical evaluation of all our methods on four structured prediction datasets.
Researcher Affiliation Academia 1INRIA Ecole Normale Sup erieure, Paris, France 2INRIA Centrale Sup elec, Chˆatenay-Malabry, France
Pseudocode Yes Algorithm 1 Block-Coordinate Frank-Wolfe (BCFW) algorithm for structured SVM
Open Source Code Yes The code and datasets are available at our project webpage.2 2 http://www.di.ens.fr/sierra/research/gap BCFW/
Open Datasets Yes We evaluate our methods on four datasets for different structured prediction tasks: OCR (Taskar et al., 2003) for handwritten character recognition, Co NLL (Tjong Kim Sang & Buchholz, 2000) for text chunking, Horse Seg (Kolesnikov et al., 2014) for binary image segmentation and LSP (Johnson & Everingham, 2010) for pose estimation.
Dataset Splits No The paper mentions using a 'test performance' value for λ, but does not provide specific train/validation/test split percentages or sample counts for any of the datasets.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU/GPU models, memory, or cloud instance types).
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used in the experiments.
Experiment Setup Yes We report the results of each method on 6 datasets (including 3 sizes of Horse Seg) for three values of the regularization parameter λ: the value leading to the best test performance, a smaller and a larger value. For each setup, we report the duality gap against both number of oracle calls and elapsed time. We run each method 5 times with different random seeds influencing the order of sampled objects and report the median (bold line), minimum and maximum values (shaded region).