TY - JOUR

T1 - More supervision, less computation

T2 - Statistical-computational tradeoffs in weakly supervised learning

AU - Yi, Xinyang

AU - Wang, Zhaoran

AU - Yang, Zhuoran

AU - Caramanis, Constantine

AU - Liu, Han

N1 - Publisher Copyright:
Copyright © 2019, The Authors. All rights reserved.
Copyright:
Copyright 2020 Elsevier B.V., All rights reserved.

PY - 2019/7/14

Y1 - 2019/7/14

N2 - We consider the weakly supervised binary classification problem where the labels are randomly flipped with probability 1 − α. Although there exist numerous algorithms for this problem, it remains theoretically unexplored how the statistical accuracies and computational efficiency of these algorithms depend on the degree of supervision, which is quantified by α. In this paper, we characterize the effect of α by establishing the information-theoretic and computational boundaries, namely, the minimax-optimal statistical accuracy that can be achieved by all algorithms, and polynomial-time algorithms under an oracle computational model. For small α, our result shows a gap between these two boundaries, which represents the computational price of achieving the information-theoretic boundary due to the lack of supervision. Interestingly, we also show that this gap narrows as α increases. In other words, having more supervision, i.e., more correct labels, not only improves the optimal statistical accuracy as expected, but also enhances the computational efficiency for achieving such accuracy.

AB - We consider the weakly supervised binary classification problem where the labels are randomly flipped with probability 1 − α. Although there exist numerous algorithms for this problem, it remains theoretically unexplored how the statistical accuracies and computational efficiency of these algorithms depend on the degree of supervision, which is quantified by α. In this paper, we characterize the effect of α by establishing the information-theoretic and computational boundaries, namely, the minimax-optimal statistical accuracy that can be achieved by all algorithms, and polynomial-time algorithms under an oracle computational model. For small α, our result shows a gap between these two boundaries, which represents the computational price of achieving the information-theoretic boundary due to the lack of supervision. Interestingly, we also show that this gap narrows as α increases. In other words, having more supervision, i.e., more correct labels, not only improves the optimal statistical accuracy as expected, but also enhances the computational efficiency for achieving such accuracy.

UR - http://www.scopus.com/inward/record.url?scp=85094259476&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85094259476&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:85094259476

JO - Free Radical Biology and Medicine

JF - Free Radical Biology and Medicine

SN - 0891-5849

ER -