Large-Margin Metric Learning for Constrained Partitioning Problems

Authors: Rémi Lajugie, Francis Bach, Sylvain Arlot

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments show how learning the metric can significantly improve performance on bioinformatics, video or image segmentation problems.
Researcher Affiliation Academia R emi Lajugie REMI.LAJUGIE@ENS.FR Sylvain Arlot SYLVAIN.ARLOT@ENS.FR Francis Bach FRANCIS.BACH@INRIA.FR D epartement d Informatique de l Ecole Normale Sup erieure, (CNRS/INRIA/ENS), Paris, France
Pseudocode Yes Algorithm 1 Dynamic programming for maximizing Tr(AM) such that M Mseq
Open Source Code No The paper does not provide an explicit statement or link to the source code for the methodology described.
Open Datasets Yes On the data from the Neuroblastoma dataset (Hocking et al., 2013), some caryotypes with changes of distribution were manually annotated. We consider the task of segmenting images of the Weizmann horses dataset (Borenstein & Ullman, 2004), using N = 20 training images with colour and dense SIFT features. In Table 2, we present analogous results for the Oxford flowers (Nilsback & Zisserman, 2006) dataset, for which the training set size is bigger: 150 images.
Dataset Splits Yes Using 4 shows for train, 3 for validation, 3 for test, we report below the test errors for each test show with the loss ℓ(smaller is better).
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions general software or techniques but does not specify software dependencies with version numbers.
Experiment Setup No The paper provides details on features used (GIST, MFCC, etc.) and states that the structured SVM parameter was adjusted using a validation set, but it does not specify concrete hyperparameter values or comprehensive training configurations like learning rates, batch sizes, or optimizer details.