Regularizing towards Causal Invariance: Linear Models with Proxies

Authors: Michael Oberst, Nikolaj Thams, Jonas Peters, David Sontag

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our theoretical findings in synthetic experiments and using real data of hourly pollution levels across several cities in China.
Researcher Affiliation Academia 1EECS, MIT, Cambridge, USA 2Department of Mathematical Sciences, University of Copenhagen, Copenhagen, Denmark.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Code for experiments is available at https://github. com/clinicalml/proxy-anchor-regression.
Open Datasets Yes We test our approach on a real-world heterogeneous dataset of hourly pollution readings in five cities in China, taken over several years (Liang et al., 2016)
Dataset Splits Yes For Proxy and Cross-Proxy AR (PAR, x PAR, see Section 3), we choose λ [0, 40] by leave-one-group-out cross-validation on the three training seasons, using the first year (2013) of data.
Hardware Specification No The paper does not provide specific details about the hardware used for the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes We simulate training data (at different levels of signal-to-variance) from an SCM with the structure given in Figure 2, fix λ := 5 and fit PAR and x PAR. For Proxy and Cross-Proxy AR (PAR, x PAR, see Section 3), we choose λ [0, 40] by leave-one-group-out cross-validation on the three training seasons, using the first year (2013) of data.