Out-of-Variable Generalisation for Discriminative Models

Authors: Siyuan Guo, Jonas Bernhard Wildberger, Bernhard Schölkopf

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform both synthetic and real-world (Appendix D.4) experiments to evaluate our algorithm s OOV learning performance. ... Table 1 records the mean and standarad deviation of the MSE loss between the predicted and observed target values for different methods.
Researcher Affiliation Academia Correspondence to: siyuan.guo@tuebingen.mpg.de University of Cambridge, United Kingdom Max Planck Institute for Intelligent Systems, Tübingen, Germany
Pseudocode Yes See Algorithm 1 in Appendix D.3 for a detailed procedure.
Open Source Code Yes Code: https://github.com/syguo96/Out-of-Variable-Generalization
Open Datasets No The paper mentions generating 'synthetic data' and conducting 'real-world (Appendix D.4) experiments'. For synthetic data, no public dataset is used. For real-world data, the provided text does not offer concrete access information (link, DOI, citation) to a publicly available dataset in Appendix D.4 or elsewhere.
Dataset Splits No The paper mentions '10k data in the source environment' and '100k points from the source environment' but does not specify training, validation, and test splits (e.g., percentages or counts) for any dataset.
Hardware Specification No The paper does not provide any specific details regarding the hardware (e.g., CPU, GPU models, memory) used to run the experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9) required to replicate the experiments.
Experiment Setup Yes We generate synthetic data according to 3.1 for a range of function classes. The inputs X 2 R3 are independently generated from a Gamma distribution. Variable Y is a function of the inputs, and the observed values are generated with noise, Yobs = Y + , where N(0, σ2), σ = 0.1. ... We use 10k data in the source environment and randomly sample 5 functions in each function class and average the results after a hyperparameter sweep.