Abstract
Predictive models can fail to generalize from training to deployment environments because of dataset shift, posing a threat to model reliability in practice. As opposed to previous methods which use samples from the target distribution to reactively correct dataset shift, we propose using graphical knowledge of the causal mechanisms relating variables in a prediction problem to proactively remove variables that participate in spurious associations with the prediction target, allowing models to generalize across datasets. To accomplish this, we augment the causal graph with latent counter-factual variables that account for the underlying causal mechanisms, and show how we can estimate these variables. In our experiments we demonstrate that models using good estimates of the latent variables instead of the observed variables transfer better from training to target domains with minimal accuracy loss in the training domain.
| Original language | English (US) |
|---|---|
| Title of host publication | 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018 |
| Editors | Ricardo Silva, Amir Globerson, Amir Globerson |
| Publisher | Association For Uncertainty in Artificial Intelligence (AUAI) |
| Pages | 947-957 |
| Number of pages | 11 |
| Volume | 2 |
| ISBN (Electronic) | 9781510871601 |
| State | Published - Jan 1 2018 |
| Event | 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018 - Monterey, United States Duration: Aug 6 2018 → Aug 10 2018 |
Other
| Other | 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018 |
|---|---|
| Country | United States |
| City | Monterey |
| Period | 8/6/18 → 8/10/18 |
Fingerprint
ASJC Scopus subject areas
- Artificial Intelligence
Cite this
Counterfactual normalization : Proactively addressing dataset shift using causal mechanisms. / Subbaswamy, Adarsh; Saria, Suchi.
34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018. ed. / Ricardo Silva; Amir Globerson; Amir Globerson. Vol. 2 Association For Uncertainty in Artificial Intelligence (AUAI), 2018. p. 947-957.Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
}
TY - GEN
T1 - Counterfactual normalization
T2 - Proactively addressing dataset shift using causal mechanisms
AU - Subbaswamy, Adarsh
AU - Saria, Suchi
PY - 2018/1/1
Y1 - 2018/1/1
N2 - Predictive models can fail to generalize from training to deployment environments because of dataset shift, posing a threat to model reliability in practice. As opposed to previous methods which use samples from the target distribution to reactively correct dataset shift, we propose using graphical knowledge of the causal mechanisms relating variables in a prediction problem to proactively remove variables that participate in spurious associations with the prediction target, allowing models to generalize across datasets. To accomplish this, we augment the causal graph with latent counter-factual variables that account for the underlying causal mechanisms, and show how we can estimate these variables. In our experiments we demonstrate that models using good estimates of the latent variables instead of the observed variables transfer better from training to target domains with minimal accuracy loss in the training domain.
AB - Predictive models can fail to generalize from training to deployment environments because of dataset shift, posing a threat to model reliability in practice. As opposed to previous methods which use samples from the target distribution to reactively correct dataset shift, we propose using graphical knowledge of the causal mechanisms relating variables in a prediction problem to proactively remove variables that participate in spurious associations with the prediction target, allowing models to generalize across datasets. To accomplish this, we augment the causal graph with latent counter-factual variables that account for the underlying causal mechanisms, and show how we can estimate these variables. In our experiments we demonstrate that models using good estimates of the latent variables instead of the observed variables transfer better from training to target domains with minimal accuracy loss in the training domain.
UR - http://www.scopus.com/inward/record.url?scp=85058014446&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85058014446&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85058014446
VL - 2
SP - 947
EP - 957
BT - 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018
A2 - Silva, Ricardo
A2 - Globerson, Amir
A2 - Globerson, Amir
PB - Association For Uncertainty in Artificial Intelligence (AUAI)
ER -