Classification with Rejection Based on Cost-sensitive Classification
Authors: Nontawat Charoenphakdee, Zhenghang Cui, Yivan Zhang, Masashi Sugiyama
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the usefulness of our proposed approach in clean-labeled, noisy-labeled, and positive-unlabeled classification. In this section, we provide experimental results of classification with rejection. |
| Researcher Affiliation | Academia | Nontawat Charoenphakdee 1 2 Zhenghang Cui 1 2 Yivan Zhang 1 2 Masashi Sugiyama 2 1 1The University of Tokyo, Tokyo, Japan 2RIKEN AIP, Tokyo, Japan. |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code for the methodology or a direct link to a code repository. |
| Open Datasets | Yes | Datasets and models: For binary classification, we used the subjective-versus-objective classification (Subj), which is a text dataset (Pang & Lee, 2004). Moreover, we used Phishing and Spambase, which are tabular datasets, and Twonorm, which is a synthetic dataset drawn from different multivariate Gaussian distributions (Lichman et al., 2013). We also used the Gisette dataset, which is the problem of separating the highly confusible digits 4 and 9 with noisy features (Guyon et al., 2005). ... We also used the image datasets, which are MNIST (Le Cun, 1998), Kuzushiji-MNIST (KMNIST) (Clanuwat et al., 2018), and Fashion-MNIST (Xiao et al., 2017). |
| Dataset Splits | No | The paper mentions using 'additional training data' for hyperparameter tuning for some methods, but does not provide specific percentages, sample counts, or detailed splitting methodology for a validation set. |
| Hardware Specification | Yes | We would like to thank ... the Supercomputing Division, Information Technology Center, The University of Tokyo, for providing us the Reedbush supercomputer system to conduct the experiments. |
| Software Dependencies | No | The paper states 'The implementation was done using Py Torch (Paszke et al., 2019).' but does not provide a specific version number for PyTorch or other software dependencies. |
| Experiment Setup | Yes | The varying rejection costs ranged from {0.1, 0.15, 0.20, 0.25, 0.30, 0.35, 0.40} for all settings. Both rejection threshold of ANGLE and the temperature parameter for SCE are chosen from the following candidate set of twenty numbers spaced evenly in a log scale from 0 to 1 (inclusively) and nine integers from 2 to 10. |