Ensemble Pruning for Out-of-distribution Generalization

Authors: Fengchun Qiao, Xi Peng

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on common benchmarks demonstrate the superiority of our approach in both multiand single-source Oo D generalization.
Researcher Affiliation Academia 1Deep REAL Lab, Department of Computer and Information Sciences, University of Delaware, DE, USA. Correspondence to: Xi Peng <xipeng@udel.edu>.
Pseudocode No The paper describes mathematical formulations and procedural steps but does not include structured pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes The source codes are publicly available at: https://github.com/joffery/TEP.
Open Datasets Yes We evaluate our method on the common Oo D generalization benchmark Domain Bed (Gulrajani & Lopez-Paz, 2020). VLCS contains photographic images from four domains: Caltech101, Label Me, SUN09, and VOC2007. Terra Incognita consists of photos of wild animals captured by camera traps at four different locations.
Dataset Splits Yes Following (Gulrajani & Lopez-Paz, 2020), we use a validation set selected from the training domains for model selection and all the experimental results are averaged over 3 trials.
Hardware Specification No The paper does not explicitly describe the specific hardware used for running its experiments, such as GPU models, CPU types, or memory.
Software Dependencies No The paper mentions using ResNet-50 and standard SDP solvers but does not provide specific software library names with their version numbers.
Experiment Setup Yes We empirically set λ = 1 for all experiments. Following (He et al., 2024), we set K = N/2 for all experiments.