Human-in-the-Loop Feature Selection

Authors: Alvaro H. C. Correia, Freddy Lecue2438-2445

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our method in two scenarios: (i) a proofof-concept image classification task and (ii) a real-world project risk classification task.
Researcher Affiliation Collaboration Alvaro H. C. Correia ENSTA Paris Tech, Palaiseau, France Utrecht University, The Netherlands a.h.chaimcorreia@uu.nl Freddy Lecue Cort AIx Thales, Montreal, Canada Inria, Sophia Antipolis, France freddy.lecue@inria.fr
Pseudocode No The paper describes mathematical formulations and processes but does not include any pseudocode or algorithm blocks.
Open Source Code Yes The code for the image classification task can be found at github.com/Al Correia/ Human-in-the-loop-Feature-Selection.
Open Datasets Yes We tested our model on an augmented MNIST dataset (Le Cun et al. 1998) composed of cluttered images as in (Mnih et al. 2014).
Dataset Splits Yes Training, testing and validation sets correspond to 60/20/20% of the projects, respectively, with each class equally distributed in the validation and test sets.
Hardware Specification Yes All experiments were developed on top of the Tensorflow python API (Google Research 2015) and run on a single GPU Nvidia GEFORCE GTX 1080 Ti.
Software Dependencies No The paper mentions 'Tensorflow python API (Google Research 2015)' but does not provide a specific version number for TensorFlow or any other software libraries used.
Experiment Setup Yes The model based on the PD estimator was trained for 100 epochs with batches of 876 projects (1000 images)... the whole model was updated via Adagrad (Duchi, Hazan, and Singer 2011) with initial learning rate of .5. The hyperparameter λ in (7) was set to 1... For the SF estimator, the regularization parameters φ, λs and λv in (9) and (10) were set to .2, 1 and 1, respectively. For the PD estimator... it was set to 10 and decayed by 4% every 100 steps.