Simple Weak Coresets for Non-decomposable Classification Measures
DOI:
https://doi.org/10.1609/aaai.v38i13.29341Keywords:
ML: Evaluation and Analysis, ML: Classification and Regression, ML: Dimensionality Reduction/Feature Selection, ML: Scalability of ML Systems, ML: Learning Theory, ML: Optimization, ML: Learning on the Edge & Model Compression, SO: Sampling/Simulation-based SearchAbstract
While coresets have been growing in terms of their application, barring few exceptions, they have mostly been limited to unsupervised settings. We consider supervised classification problems, and non-decomposable evaluation measures in such settings. We show that stratified uniform sampling based coresets have excellent empirical performance that are backed by theoretical guarantees too. We focus on the F1 score and Matthews Correlation Coefficient, two widely used non-decomposable objective functions that are nontrivial to optimize for and show that uniform coresets attain a lower bound for coreset size, and have good empirical performance, comparable with ``smarter'' coreset construction strategies.
Downloads
Published
2024-03-24
How to Cite
Malaviya, J., Dasgupta, A., & Chhaya, R. (2024). Simple Weak Coresets for Non-decomposable Classification Measures. Proceedings of the AAAI Conference on Artificial Intelligence, 38(13), 14289-14296. https://doi.org/10.1609/aaai.v38i13.29341
Issue
Section
AAAI Technical Track on Machine Learning IV
