PAC Generalization via Invariant Representations

Authors: Advait U Parulekar, Karthikeyan Shanmugam, Sanjay Shakkottai

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we empirically demonstrate that in the setting described above using a notion of generalization that we describe, most approximately invariant representations generalize to most new distributions.
Researcher Affiliation Collaboration 1Department of Electrical and Computer Engineering, University of Texas at Austin 2Google Research India.
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm block.
Open Source Code Yes The code is available at https://github.com/advaitparulekar/PAC IRM
Open Datasets No We consider the 7-node linear SEM in Figure 3. The target variable is taken to be Xt. Each edge weight is set to 1 for the observational distribution.
Dataset Splits No The paper discusses drawing training and test samples but does not specify a separate validation set or exact split percentages for reproduction.
Hardware Specification No The paper describes experiments but does not specify hardware details such as GPU/CPU models or memory.
Software Dependencies No The paper provides a link to its code but does not list specific software dependencies with version numbers.
Experiment Setup Yes We consider the 7-node linear SEM in Figure 3. The target variable is taken to be Xt. Each edge weight is set to 1 for the observational distribution. We consider an interventional distribution Dhard with support over the set of hard interventions on nodes {v3, v4, v5}. Recall that a hard intervention consists of assigning a value to a node. We draw m interventional distributions from Dhard as our training interventions, and draw a sample consisting of N = 200000 datapoints from each distribution. In our experiments, σ2 = 1.