On active learning methods for manifold data

Abstract

Active learning is a major area of interest within the field of machine learning, especially when the labeled instances are very difficult, time-consuming or expensive to obtain. In this paper, we review various active learning methods for manifold data, where the intrinsic manifold structure of data is also incorporated into the active learning query strategies. In addition, we present a new manifold-based active learning algorithm for Gaussian process classification. This new method uses a data-dependent kernel derived from a semi-supervised model that considers both labeled and unlabeled data. The method performs a regularization on the smoothness of the fitted function with respect to both the ambient space and the manifold where the data lie. The regularization parameter is treated as an additional kernel (covariance) parameter and estimated from the data, permitting adaptation of the kernel to the given dataset manifold geometry. Comparisons with other AL methods for manifold data show faster learning performance in our empirical experiments. MATLAB code that reproduces all examples is provided as supplementary materials.

Beyond Tripeptides Two-Step Active Machine Learning for Very Large Data sets

“…Active learning is a rational sampling method that aims to identify the most informative data to label so that a supervised model trained on this data would perform better than a supervised model trained on an equivalent amount of labeled data chosen at random. 19 Active learning may also be known as sequential learning as it uses all measures up-to-date to inform the next-best candidate for labeling in an increasingly informed search for the optimal training set with minimal data. 20 Shmilovich et al have used this approach to traverse the chemical space of the DXXX-OPV3-XXXD molecular template, where OPV3 represents 1,4-distyrylbenzene and XXX represents variable tripeptides.…”

Optimal Design of Experiments on Riemannian Manifolds

“…Instead of using fixed values for regularization parameters, model selection criterion with theoretical justification might provide better learning performance. Similar work has been discussed by Li et al (2019), where they maximize the likelihood function to choose the values of λ A and λ I in a Gaussian Process model. Secondly, there are other optimality criterion than the D/G "alphabetic" criteria in the field of optimal design of experiments.…”

Cost-efficient unsupervised sample selection for multivariate calibration

“…Further research is recommended to refine this criterion of equivalence and to propose a sample selection algorithm aimed at optimizing this equivalence. This concept could be extended to other model architectures by identifying the mathematical components that could be controlled or evaluated in unsupervised sample selection [34].…”