Predictive models that generalize well under distributional shift are often
desirable and sometimes crucial to building robust and reliable machine
learning applications. We focus on distributional shift that arises in causal
inference from observational data and in unsupervised domain adaptation. We
pose both of these problems as prediction under a shift in design. Popular
methods for overcoming distributional shift make unrealistic assumptions such
as having a well-specified model or knowing the policy that gave rise to the
observed data. Other methods are hindered by their need for a pre-specified
metric for comparing observations, or by poor asymptotic properties. We devise
a bound on the generalization error under design shift, incorporating both
representation learning and sample re-weighting. Based on the bound, we propose
an algorithmic framework that does not require any of the above assumptions and
which is asymptotically consistent. We empirically study the new framework
using two synthetic datasets, and demonstrate its effectiveness compared to
previous methods.
Description
[1802.08598] Learning Weighted Representations for Generalization Across Designs
%0 Journal Article
%1 johansson2018learning
%A Johansson, Fredrik D.
%A Kallus, Nathan
%A Shalit, Uri
%A Sontag, David
%D 2018
%K causal-analysis robustness
%T Learning Weighted Representations for Generalization Across Designs
%U http://arxiv.org/abs/1802.08598
%X Predictive models that generalize well under distributional shift are often
desirable and sometimes crucial to building robust and reliable machine
learning applications. We focus on distributional shift that arises in causal
inference from observational data and in unsupervised domain adaptation. We
pose both of these problems as prediction under a shift in design. Popular
methods for overcoming distributional shift make unrealistic assumptions such
as having a well-specified model or knowing the policy that gave rise to the
observed data. Other methods are hindered by their need for a pre-specified
metric for comparing observations, or by poor asymptotic properties. We devise
a bound on the generalization error under design shift, incorporating both
representation learning and sample re-weighting. Based on the bound, we propose
an algorithmic framework that does not require any of the above assumptions and
which is asymptotically consistent. We empirically study the new framework
using two synthetic datasets, and demonstrate its effectiveness compared to
previous methods.
@article{johansson2018learning,
abstract = {Predictive models that generalize well under distributional shift are often
desirable and sometimes crucial to building robust and reliable machine
learning applications. We focus on distributional shift that arises in causal
inference from observational data and in unsupervised domain adaptation. We
pose both of these problems as prediction under a shift in design. Popular
methods for overcoming distributional shift make unrealistic assumptions such
as having a well-specified model or knowing the policy that gave rise to the
observed data. Other methods are hindered by their need for a pre-specified
metric for comparing observations, or by poor asymptotic properties. We devise
a bound on the generalization error under design shift, incorporating both
representation learning and sample re-weighting. Based on the bound, we propose
an algorithmic framework that does not require any of the above assumptions and
which is asymptotically consistent. We empirically study the new framework
using two synthetic datasets, and demonstrate its effectiveness compared to
previous methods.},
added-at = {2019-05-23T04:15:00.000+0200},
author = {Johansson, Fredrik D. and Kallus, Nathan and Shalit, Uri and Sontag, David},
biburl = {https://www.bibsonomy.org/bibtex/236c921553c031d93d42937c8ba839c54/kirk86},
description = {[1802.08598] Learning Weighted Representations for Generalization Across Designs},
interhash = {1731fb14d27ec88287b6a8d9e58ef61a},
intrahash = {36c921553c031d93d42937c8ba839c54},
keywords = {causal-analysis robustness},
note = {cite arxiv:1802.08598},
timestamp = {2019-05-23T04:15:00.000+0200},
title = {Learning Weighted Representations for Generalization Across Designs},
url = {http://arxiv.org/abs/1802.08598},
year = 2018
}