Early and Reliable Event Detection Using Proximity Space Representation

Authors: Maxime Sangnier, Jerome Gauthier, Alain Rakotomamonjy

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental studies provide compelling results on toy data, presenting the trade-off that occurs when aiming at accuracy, earliness and reliability. Results on real physiological and video datasets show that our proposed approach is as accurate and early as state-of-the-art algorithm, while ensuring reliability and being far more efficient to learn.
Researcher Affiliation Academia Maxime Sangnier MAXIME.SANGNIER@TELECOM-PARISTECH.FR LTCI, CNRS, T el ecom Paris Tech, Universit e Paris-Saclay, 75013, Paris, France Jerome Gauthier JEROME.GAUTHIER@CEA.FR LADIS, CEA, LIST, 91191, Gif-sur-Yvette, France Alain Rakotomamonjy ALAIN.RAKOTO@INSA-ROUEN.FR Normandie Universit e, UR, LITIS EA 4108, Avenue de l universit e, 76801, Saint-Etienne-du-Rouvray, France
Pseudocode Yes Algorithm 1 Algorithm to learn a an early detector. 1: A random sample of indexes between 1 and m 2: repeat 3: Q yik(X(i) 1..T , pj) j A, 1 i n 4: β solve Problem (4) 5: j arg min 1 j m ((Qβ)j + µj) 6: j+ arg max 1 j m ((Qβ)j µj) 7: if µj λ (Qβ)j and (Qβ)j+ µj+ then 8: convergence 9: else if µj λ > (Qβ)j then 10: A A {j } 11: else 12: A A {j+} 13: end if 14: until convergence
Open Source Code Yes Matlab R code ran on a single core of an Intel R Xeon R E5-2630 CPU, operating at 2.4 GHz with GNU/Linux and 144 Gb of RAM. In addition, this code is available on the authors websites.
Open Datasets Yes For this experiment, we used the publicly available dataset 2a from BCI competition IV (Brunner et al., 2008).As a video-based experiment, we consider the extended Cohn-Kanade dataset (CK+) (Lucey et al., 2010).
Dataset Splits Yes C is obtained through a 5-fold crossvalidation (maximizing the AUROC) on the following grid [2 2, 20, . . . , 210].
Hardware Specification Yes Matlab R code ran on a single core of an Intel R Xeon R E5-2630 CPU, operating at 2.4 GHz with GNU/Linux and 144 Gb of RAM.
Software Dependencies No The paper mentions 'Matlab R code' and 'GNU/Linux', and mentions 'lpsolve (Berkelaar et al., 2004)' as a tool, but does not provide specific version numbers for Matlab or lpsolve, or any other key software components used for the experiments.
Experiment Setup Yes The parameter γ that defines the landmarking space is set to 2 1 (default value for normalized data).C is obtained through a 5-fold crossvalidation (maximizing the AUROC) on the following grid [2 2, 20, . . . , 210].