Active feature acquisition reduces cost while maintaining accuracy
Active Feature Acquisition by Distilling an Oracle Teacher
Machine LearningArtificial Intelligence
Summary
Measuring all information about something can be expensive or slow, so it's better to decide which details to get first. The authors created a way to teach a computer model exactly which features are most helpful for making good predictions by showing it an all-knowing example during training. This approach helps the model learn which features to acquire and when to stop, saving money or time while keeping the accuracy high. Their method works well on medical diagnosis data and other tests, getting most of the benefits of buying all features but at much less cost.
What this means in practice
- •For hospital data teams: Optimize which medical tests to order per patient to save costs without compromising diagnostic accuracy.
- •For mobile app developers: Develop apps that request user information selectively to minimize privacy risks and data collection costs while maintaining performance.
Authors
Yang Meng, Zixin Ding, Robert L. Grossman, Yuxin Chen
Abstract
Acquiring every feature of an instance is often costly: a measurement can carry a monetary price, a time delay, or a privacy cost. Active Feature Acquisition (AFA) addresses this by learning a policy that decides, per instance, which features to acquire under a budget. Existing AFA methods train this policy by optimizing a self-constructed utility target rather than distilling an explicit teacher. We construct this teacher explicitly. At training time the label and the unobserved feature values are available, and together they determine the teacher in closed form: an oracle that ranks every candidate by its effect on the downstream predictor's error. OracleImitator distills this ranking into a deployable policy by soft-target imitation. Across multimodal medical diagnosis and the AFABench benchmark, OracleImitator matches or outperforms existing AFA methods across the budget range, recovering 90% of the attainable accuracy gain at under a quarter of the acquisition cost. Because the teacher is an explicit algorithmic object rather than an implicit optimization target, we can also measure directly how much of its ranking is recoverable by a policy restricted to what is visible at deployment.