Revisiting Discriminative vs. Generative Classifiers: Theory and Implications

Authors: Chenyu Zheng, Guoqiang Wu, Fan Bao, Yue Cao, Chongxuan Li, Jun Zhu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Simulation results on a mixture of Gaussian validate our theoretical findings. Experiments on various pre-trained deep vision models show that naïve Bayes consistently converges faster as the number of data increases.
Researcher Affiliation Collaboration 1Gaoling School of AI, Renmin University of China; Beijing Key Laboratory of Big Data Management and Analysis Methods, Beijing, China 2School of Software, Shandong University 3Dept. of Comp. Sci. & Tech., Institute for AI, Tsinghua-Huawei Joint Center for AI, BNRist Center, THBI Lab, Tsinghua University 4Beijing Academy of Artificial Intelligence. Correspondence to: Chongxuan Li <chongxuanli@ruc.edu.cn>.
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/ML-GSAI/Revisiting-Dis-vs Gen-Classifiers.
Open Datasets Yes Experiments on various pre-trained deep vision models... We systematically compare logistic regression and naïve Bayes on the CIFAR10 and CIFAR100 datasets (Krizhevsky et al., 2009).
Dataset Splits Yes We systematically compare logistic regression and naïve Bayes on the CIFAR10 and CIFAR100 datasets (Krizhevsky et al., 2009).
Hardware Specification Yes All experiments are done on a single Ge Force RTX 3090 GPU.
Software Dependencies No We train the logistic regression using scikit-learn s (Pedregosa et al., 2011) L-BFGS implementation... Res Net. We add the Res Net50 checkpoint released by Pytorch (Paszke et al., 2019). (Specific software versions are not provided).
Experiment Setup Yes For a fair comparison, we keep the linear evaluation setting in (Radford et al., 2021) throughout the experiments. Specially, we train the logistic regression using scikitlearn s (Pedregosa et al., 2011) L-BFGS implementation, with a maximum of 1000 iterations. We adjust the weight of ℓ2 regularization of logistic regression carefully... For each m, we obtain training samples randomly 5 times and record the mean test error of two models.