Skip to yearly menu bar Skip to main content


Poster

Adaptive Labeling for Efficient Out-of-distribution Model Evaluation

Daksh Mittal · Yuanzhe Ma · Shalmali Joshi · Hongseok Namkoong

East Exhibit Hall A-C #2206
[ ] [ Project Page ]
Fri 13 Dec 11 a.m. PST — 2 p.m. PST

Abstract:

Datasets often suffer severe selection bias; clinical labels are only available on patients for whom doctors ordered medical exams. To assess model performance outside the support of available data, we present a computational framework for adaptive labeling, providing cost-efficient model evaluations under severe distribution shifts. We formulate the problem as a Markov Decision Process over states defined by posterior beliefs on model performance. Each batch of new labels incurs a “state transition” to sharper beliefs, and we choose batches to minimize uncertainty on model performance at the end of the label collection process. Instead of relying on high-variance REINFORCE policy gradient estimators that do not scale, our adaptive labeling policy is optimized using path-wise policy gradients computed by auto-differentiating through simulated roll-outs. Our framework is agnostic to different uncertainty quantification approaches and highlights the virtue of planning in adaptive labeling. On synthetic and real datasets, we empirically demonstrate even a one-step lookahead policy substantially outperforms active learning-inspired heuristics.

Live content is unavailable. Log in and register to view live content