Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Stochastic Encodings for Active Feature Acquisition

About

Active Feature Acquisition is an instance-wise, sequential decision making problem. The aim is to dynamically select which feature to measure based on current observations, independently for each test instance. Common approaches either use Reinforcement Learning, which experiences training difficulties, or greedily maximize the conditional mutual information of the label and unobserved features, which makes myopic acquisitions. To address these shortcomings, we introduce a latent variable model, trained in a supervised manner. Acquisitions are made by reasoning about the features across many possible unobserved realizations in a stochastic latent space. Extensive evaluation on a large range of synthetic and real datasets demonstrates that our approach reliably outperforms a diverse set of baselines.

Alexander Norcliffe, Changhee Lee, Fergus Imrie, Mihaela van der Schaar, Pietro Lio• 2025

Related benchmarks

TaskDatasetResultRank
Active Feature AcquisitionCube-NM nc = 5, sigma = 0.1
Training Runtime (s)159
16
Active Feature AcquisitionCube-NM nc = 5, sigma = 0.2
Training Runtime (s)130
16
Active Feature AcquisitionSyn1
Training Runtime (s)316
16
Active Feature AcquisitionSyn3
Training Runtime (s)281
16
Active Feature Acquisitionconnect4
Training Runtime (s)460
16
Active Feature AcquisitionSplice
Training Runtime (s)15.8
16
Active Feature AcquisitionEngineFaultDB
Training Runtime (s)133
16
Active Feature AcquisitionMETABRIC
Training Runtime (s)106
16
Dynamic Feature SelectionCirrhosis 5-fold CV
AUAC-F150.67
13
Dynamic Feature SelectionWine (5-fold CV)
AUAC-F195.81
13
Showing 10 of 32 rows

Other info

Follow for update