Shaken, and Stirred: Long-Range Dependencies Enable Robust Outlier Detection with PixelCNN++

Authors: Barath Mohan Umapathi, Kushal Chauhan, Pradeep Shenoy, Devarajan Sridharan

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

Reproducibility Variable Result LLM Response
Research Type Experimental We test our approaches extensively with five grayscale and six natural image datasets and show that they achieve or exceed stateof-the-art outlier detection, particularly on datasets with complex, natural images.
Researcher Affiliation Collaboration Barath Mohan Umapathi1 , Kushal Chauhan2 , Pradeep Shenoy2 and Devarajan Sridharan3,4, 1Department of Physics, Indian Institute of Science 2Google Research 3Center for Neuroscience, Indian Institute of Science 4Computer Science and Automation, Indian Institute of Science
Pseudocode No The paper describes methods in prose and figures but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes Code associated with this paper is available at: https://github. com/coglabiisc/googleresearch/tree/main/pixelcnn ood
Open Datasets Yes We trained Pixel CNN++ models on each of five grayscale image datasets: MNIST, Fashion MNIST, EMNIST Letters, Sign Language MNIST, and CLEVR [Deng, 2012; Xiao et al., 2017; Cohen et al., 2017; Johnson et al., 2017; Sign Language MNIST, 2017]. Similarly, we trained Pixel CNN++ models on each of six natural image datasets SVHN, Celeb A, Comp Cars, GTSRB, CIFAR10, and LSUN (classroom) [Netzer et al., 2011; Liu et al., 2015; Yang et al., 2015; Stallkamp et al., 2011; Krizhevsky, 2009; Yu et al., 2015].
Dataset Splits No The paper mentions training and test sets and sometimes refers to training log-likelihoods, but does not explicitly provide details on how the datasets were split into training, validation, and test sets (e.g., percentages or counts).
Hardware Specification No The paper does not provide specific hardware details such as CPU/GPU models, memory, or cloud instance types used for running experiments. It mentions time and space complexity in Table 1 but no specific hardware.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used in the implementation.
Experiment Setup No The paper does not explicitly detail hyperparameter values, optimization settings, or other specific experimental setup configurations in the main text.