English

Agnostic learning with unknown utilities

Machine Learning 2021-04-20 v1 Machine Learning

Abstract

Traditional learning approaches for classification implicitly assume that each mistake has the same cost. In many real-world problems though, the utility of a decision depends on the underlying context xx and decision yy. However, directly incorporating these utilities into the learning objective is often infeasible since these can be quite complex and difficult for humans to specify. We formally study this as agnostic learning with unknown utilities: given a dataset S={x1,,xn}S = \{x_1, \ldots, x_n\} where each data point xiDx_i \sim \mathcal{D}, the objective of the learner is to output a function ff in some class of decision functions F\mathcal{F} with small excess risk. This risk measures the performance of the output predictor ff with respect to the best predictor in the class F\mathcal{F} on the unknown underlying utility uu^*. This utility uu^* is not assumed to have any specific structure. This raises an interesting question whether learning is even possible in our setup, given that obtaining a generalizable estimate of utility uu^* might not be possible from finitely many samples. Surprisingly, we show that estimating the utilities of only the sampled points~SS suffices to learn a decision function which generalizes well. We study mechanisms for eliciting information which allow a learner to estimate the utilities uu^* on the set SS. We introduce a family of elicitation mechanisms by generalizing comparisons, called the kk-comparison oracle, which enables the learner to ask for comparisons across kk different inputs xx at once. We show that the excess risk in our agnostic learning framework decreases at a rate of O(1k)O\left(\frac{1}{k} \right). This result brings out an interesting accuracy-elicitation trade-off -- as the order kk of the oracle increases, the comparative queries become harder to elicit from humans but allow for more accurate learning.

Keywords

Cite

@article{arxiv.2104.08482,
  title  = {Agnostic learning with unknown utilities},
  author = {Kush Bhatia and Peter L. Bartlett and Anca D. Dragan and Jacob Steinhardt},
  journal= {arXiv preprint arXiv:2104.08482},
  year   = {2021}
}

Comments

30 pages; published as a conference paper at ITCS 2021

R2 v1 2026-06-24T01:16:18.507Z