R-divergence for Estimating Model-oriented Distribution Discrepancy

Authors: Zhilin Zhao, Longbing Cao

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the test power across various unsupervised and supervised tasks and find that R-divergence achieves state-of-the-art performance.
Researcher Affiliation Academia Zhilin Zhao Longbing Cao Data Science Lab, School of Computing & Data X Research Centre Macquarie University, Sydney, NSW 2109, Australia
Pseudocode Yes The procedure of estimating the model-oriented discrepancy between two probability distributions is summarized in Algorithm 1 (see Appendix B.1).
Open Source Code Yes The source code is publicly available at: https://github.com/Lawliet-zzl/R-div.
Open Datasets Yes Benchmark Datasets: Following the experimental setups as [41] and [62], we adopt four benchmark datasets, namely Blob, HDGM, HIGGS, and MNIST.
Dataset Splits Yes For fair comparisons, we follow the same setup as [62] for all methods. Because all the compared methods have hyper-parameters, we evenly split each dataset into a training set and a validation set to tune hyper-parameters and compute the final test power, respectively.
Hardware Specification No The paper mentions using specific network architectures (e.g., Res Net18, Alex Net) but does not provide any details about the hardware (e.g., GPU/CPU models, memory) used for running experiments.
Software Dependencies No The paper mentions using Adam optimizer but does not provide specific version numbers for any programming languages, libraries, or frameworks used.
Experiment Setup Yes In the pretraining and retraining processes, the learning rate [60] starts at 0.1 and is divided by 10 after 100 and 150 epochs. The batch size is 128, and the number of epochs is 200.