Improved Algorithms for Agnostic Pool-based Active Classification

Authors: Julian Katz-Samuels, Jifan Zhang, Lalit Jain, Kevin Jamieson

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we demonstrate that our algorithm is superior to state of the art agnostic active learning algorithms on image classification datasets.
Researcher Affiliation Academia 1University of Wisconsin, Madison, WI 2Paul G. Allen School of Computer Science and Engineering, University of Washington, Seattle, WA.
Pseudocode Yes Algorithm 1 ACED (Active Classification using Experimental Design). Algorithm 2 Fixed Budget ACED.
Open Source Code Yes Code can be found at https://github.com/jifanz/ ACED.
Open Datasets Yes MNIST 0-4 vs 5-9 (Le Cun et al., 1998). SVHN 2 vs 7 (Netzer et al., 2011). CIFAR Bird vs Plane (Netzer et al., 2011). Fashion MNIST T-shirt vs Pants (Xiao et al., 2017).
Dataset Splits No The paper describes using active learning from a pool of unlabeled examples and retraining models, but it does not specify explicit train/validation/test splits with percentages or counts for the entire dataset.
Hardware Specification No Computational resources from Amazon Web Services were generously gifted as part of an Amazon Research Award. No specific hardware (e.g., GPU/CPU models, memory) is mentioned.
Software Dependencies No We used the logistic regression implementation in Scikit-learn (Pedregosa et al., 2011). No version numbers for Scikit-learn or Vowpal Wabbit are provided.
Experiment Setup Yes Detailed hyperparameters considered for the baselines are included in Appendix M.