Network Deconvolution

Authors: Chengxi Ye, Matthew Evanusa, Hua He, Anton Mitrokhin, Tom Goldstein, James A. Yorke, Cornelia Fermuller, Yiannis Aloimonos

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that the network deconvolution operation is able to deliver performance improvement in all cases on the CIFAR-10, CIFAR-100, MNIST, Fashion-MNIST, Cityscapes, and Image Net datasets.
Researcher Affiliation Academia Department of Computer Science, University of Maryland, College Park {cxy, mevanusa, huah, amitrokh}@umd.edu {tomg@cs,yorke@,fer@umiacs,yiannis@cs}.umd.edu
Pseudocode Yes Algorithm 1 Computing the Deconvolution Matrix
Open Source Code Yes Source code can be found at: https://github.com/yechengxi/deconvolution
Open Datasets Yes Extensive experiments show that the network deconvolution operation is able to deliver performance improvement in all cases on the CIFAR-10, CIFAR-100, MNIST, Fashion-MNIST, Cityscapes, and Image Net datasets.
Dataset Splits No The paper mentions training on various datasets and evaluating performance, but it does not explicitly specify exact training, validation, and test splits (e.g., percentages or sample counts for each partition).
Hardware Specification No The paper mentions 'CPU timing' in Table 3 but does not specify any particular CPU models, GPU models, or other hardware specifications used for the experiments.
Software Dependencies No The paper mentions adapting models from 'pytorch-cifar' and using models from the 'Py Torch model zoo', implying the use of PyTorch. However, no specific version numbers for PyTorch or any other software dependencies are provided.
Experiment Setup Yes For the convolutional layers, we set B = 64 before calculating the covariance matrix. For the fully-connected layers, we set B equal to the input channel number, which is usually 512. We set the batch size to 128 and the weight decay to 0.001. All models are trained with SGD and a learning rate of 0.1.