Exploratory machine learning with unknown unknowns

DOI10.1016/J.ARTINT.2023.104059MaRDI QIDQ6152674zbMATH OpenOpenAlexFDO

Authors Peng Zhao, Jia-Wei Shan, Yujie Zhang, Zhi-Hua Zhou

Publication date 13 February 2024

Published in Artificial Intelligence (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2002.01605

zbMATH Keywords

robustnessunknown unknownsexploratory machine learningrobust AI

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05)

Abstract: In conventional supervised learning, a training dataset is given with ground-truth labels from a known label set, and the learned model will classify unseen instances to the known labels. In this paper, we study a new problem setting in which there are unknown classes in the training dataset misperceived as other labels, and thus their existence appears unknown from the given supervision. We attribute the unknown unknowns to the fact that the training dataset is badly advised by the incompletely perceived label space due to the insufficient feature information. To this end, we propose the exploratory machine learning, which examines and investigates the training dataset by actively augmenting the feature space to discover potentially unknown labels. Our approach consists of three ingredients including rejection model, feature acquisition, and model cascade. The effectiveness is validated on both synthetic and real datasets.

Recommendations

• Generalized mixture models, semi-supervised learning, and unknown class inference
• Unsupervised supervised learning. I: Estimating classification and regression errors without labels
• Learning a classifier when the labeling is known
• Anomaly and novelty detection for robust semi-supervised learning
• Semi-supervised novelty detection

Cites work

• A survey on concept drift adaptation
• Action elimination and stopping conditions for the multi-armed bandit and reinforcement learning problems
• Active learning
• Advanced Lectures on Machine Learning
• Bagging predictors
• Classification methods with reject option based on convex risk minimization
• Classification with a reject option using a hinge loss
• Classification with reject option
• Ensembling neural networks: Many could be better than all
• Fast classification rates without standard margin assumptions
• Foundations of machine learning
• Handwritten digit recognition by combined classifiers.
• Improving classification performance through selective instance completion
• Learning with Rejection
• Mixed rational soliton solutions of two differential-difference equations in Casorati determinant form
• On optimum recognition error and reject tradeoff
• On reject and refine options in multicategory classification
• Reinforcement learning. An introduction

This page was built for publication: Exploratory machine learning with unknown unknowns

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6152674)