Multiscale Score Matching for Out-of-Distribution Detection

Authors: Ahsan Mahmood, Junier Oliva, Martin Andreas Styner

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Despite its simplicity, our experiments show that this methodology significantly outperforms the state-of-the-art in detecting out-of-distribution images.
Researcher Affiliation Academia Department of Computer Science University of North Carolina at Chapel Hill {amahmood, joliva, styner}@cs.unc.edu
Pseudocode No The paper describes its methods in narrative text and mathematical equations but does not include formal pseudocode or algorithm blocks.
Open Source Code Yes We make our code and results publicly available on Github 1. 1https://github.com/ahsan Mah/msma
Open Datasets Yes We consider CIFAR-10 (Krizhevsky et al. (2009)) and SVHN (Netzer et al. (2011)) as our inlier datasets. For out-of-distribution datasets, we choose the same as Liang et al. (2017): Tiny Image Net3, LSUN (Yu et al. (2015)), i SUN (Xu et al. (2015)). ...We employ 3500 high resolution T1-weighted MR images obtained through the NIH large scale ABCD study (Casey et al. (2018))... For our outlier data, we employ MRI datasets of children aged 1, 2, 4 and 6 years (500 each) from the UNC EBDS database Stephens et al. (2020); Gilmore et al. (2020).
Dataset Splits Yes CIFAR-10: ...There are 50,000 training images and 10,000 test images. SVHN: ...We use the official splits: 73,257 digits for training, 26,032 digits for testing. For our Gaussian Mixture Models, we mean normalize the data and perform a grid search over the number of components (ranging from 2 to 20), using 10-fold cross validation.
Hardware Specification No The paper mentions 'GPU memory usage' but does not provide specific hardware details such as GPU models, CPU types, or memory amounts.
Software Dependencies No The paper mentions 'TensorFlow implementation' and 'Keras' but does not specify version numbers for these software components or any other libraries.
Experiment Setup Yes The model architecture used is a Refine Net with 128 filters. The batch size is also fixed to 128. We train for 200k iterations using the Adam optimizer. Following Song & Ermon (2019), we use L = 10 standard deviations for our Gaussian noise perturbation such that {σi}L i=1 is a geometric sequence with σ1 = 1 and σ10 = 0.01. ...For our Gaussian Mixture Models, we mean normalize the data and perform a grid search over the number of components (ranging from 2 to 20), using 10-fold cross validation. Our normalizing flow model is constructed with a MAF using two hidden layers with 128 units each, and a Standard Normal as the base distribution. It is trained for a 1000 epochs with a batch size of 128.