Posterior Re-calibration for Imbalanced Datasets

Authors: Junjiao Tian, Yen-Cheng Liu, Nathaniel Glaser, Yen-Chang Hsu, Zsolt Kira

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our results on six different datasets and five different architectures show state of art accuracy, including on large-scale imbalanced datasets such as i Naturalist for classification and Synthia for semantic segmentation.
Researcher Affiliation Academia Junjiao Tian Georgia Institute of Technology jtian73@gatech.edu Yen-Cheng Liu Georgia Institute of Technology ycliu@gatech.edu Nathaniel Glaser Georgia Institute of Technology nglaser@gatech.edu Yen-Chang Hsu Georgia Institute of Technology yenchang.hsu@gatech.edu Zsolt Kira Georgia Institute of Technology zkira@gatech.edu
Pseudocode Yes Algorithm 1: UNO-IC: multi-modal fusion algorithm for prior and likelihood shift
Open Source Code Yes Please see https://github.com/GT-RIPL/UNO-IC.git for implementation.
Open Datasets Yes Cifar10, Cifar100, i Naturalist18 [21], Synthia [23] are listed in Table 1, and [21] and [23] are formal citations to the dataset papers.
Dataset Splits Yes The hyperparameter λ is searched on the validation set. We randomly split the validation set into three subsets and use 3-fold cross validation for tuning and testing our IC method. The hyperparameter λ is tuned on a validation split and the algorithm is tested on a test set.
Hardware Specification No The paper does not provide specific details on hardware specifications (e.g., GPU/CPU models, memory) used for running experiments.
Software Dependencies No The paper does not specify software dependencies (e.g., library names with version numbers like PyTorch 1.9, TensorFlow 2.x) needed to replicate the experiment.
Experiment Setup No The paper mentions tuning the hyperparameter λ and provides some values (e.g., "IC λ = 0.4", "UNO-IC λ = 0.4"). However, it does not provide specific values for common experimental setup details like learning rate, batch size, number of epochs, or optimizer settings for the neural network training.