SDE-Net: Equipping Deep Neural Networks with Uncertainty Estimates

Authors: Lingkai Kong, Jimeng Sun, Chao Zhang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical results are presented in Section 4. We evaluate four tasks where uncertainty plays a fundamental role: out-of-distribution detection, misclassification detection, adversarial samples detection and active learning. We find that SDE-Net can outperform state-of-the-art uncertainty estimation methods or achieve competitive results across these tasks on various datasets.
Researcher Affiliation Academia Lingkai Kong 1 Jimeng Sun 2 Chao Zhang 1 1School of Computational Science and Engineering, Georgia Institute of Technology, Atlanta, GA 2Department of Computer Science, University of Illinois at Urbana-Champaign, Urbana, IL. Correspondence to: Chao Zhang <chaozhang@gateatech.edu>.
Pseudocode Yes Algorithm 1 Training of SDE-Net. h1 is the downsampling layer; h2 is the fully connected layer; f and g are the drift net and diffusion net; L is the loss function.
Open Source Code No The paper does not provide an explicit statement or a specific link indicating the availability of its source code for the described methodology.
Open Datasets Yes Table 1 shows the OOD detection performance as well as the classification accuracy on two image classification datasets: MNIST and SVHN." and "We use the Year Prediction MSD dataset (Dua & Graff, 2017) as training data and the Boston Housing dataset (Bos) as test OOD data." and "Boston dataset. https://www.cs.toronto.edu/~delve/data/boston/bostonDetail.html.
Dataset Splits No The paper references training and test sets (e.g., 'training data and the Boston Housing dataset (Bos) as test OOD data', 'evaluate performance on the whole test set'), but does not explicitly describe a separate validation set or provide specific percentages or counts for train/validation/test splits.
Hardware Specification No The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running its experiments.
Software Dependencies No The paper mentions 'Tensorflow and Pytorch' for implementation but does not specify version numbers for these or any other software dependencies.
Experiment Setup No The paper describes some aspects of the experimental setup, such as using the Euler-Maruyama scheme and performing 10 stochastic forward passes at test time. However, it lacks specific numerical values for common hyperparameters like learning rate, batch size, number of epochs, or the name of the optimizer used.