Understanding new tasks through the lens of training data via exponential tilting
This addresses the problem of distribution shift for practitioners needing to adapt models to new tasks, but it is incremental as it builds on existing reweighting and exponential tilt methods.
The paper tackles the challenge of deploying machine learning models to new tasks by reweighting training data to better represent target distributions, using an exponential tilt model to learn importance weights that minimize KL divergence, and demonstrates effectiveness on Waterbirds and Breeds benchmarks.
Deploying machine learning models to new tasks is a major challenge despite the large size of the modern training datasets. However, it is conceivable that the training data can be reweighted to be more representative of the new (target) task. We consider the problem of reweighing the training samples to gain insights into the distribution of the target task. Specifically, we formulate a distribution shift model based on the exponential tilt assumption and learn train data importance weights minimizing the KL divergence between labeled train and unlabeled target datasets. The learned train data weights can then be used for downstream tasks such as target performance evaluation, fine-tuning, and model selection. We demonstrate the efficacy of our method on Waterbirds and Breeds benchmarks.