Achieving Rotational Invariance with Bessel-Convolutional Neural Networks

Authors: Valentin Delchevalerie, Adrien Bibal, Benoît Frénay, Alexandre Mayer

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments are described and discussed in Section 5. Finally, a more general discussion about some aspects of B-CNNs, as well as some future works, are presented in Section 6 before concluding the paper in Section 7.
Researcher Affiliation Academia Valentin Delchevalerie,1 Adrien Bibal,2 Benoît Frénay,2 Alexandre Mayer3 1 PRe CISE, NADI & na Xys institutes, University of Namur, Belgium 2 PRe CISE, NADI institute, University of Namur, Belgium 3 Department of Physics, na Xys institute, University of Namur, Belgium {valentin.delchevalerie, adrien.bibal, benoit.frenay, alexandre.mayer}@unamur.be
Pseudocode No The paper describes the mathematical operations and shows a schematic illustration of the process (Figure 2), but it does not include any formal pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about releasing source code or a link to a code repository for the described methodology.
Open Datasets Yes The first dataset is MNIST [27]... The second dataset is Outex-TC-00010-r dataset [28]... The third dataset is made of 128 128 3 brain MRI images [29].
Dataset Splits No For MNIST: "For all runs, the training set contains 60, 000 images and the test set 10, 000 images." For Outex: "The training set contains 480 images with 20 orientations, and the test set is composed of 3840 images with 160 orientations." For Brain MRI: "The training and test sets contain 190 and 63 images, respectively." While a validation set is mentioned for model selection ("by picking the one that performs the best on a validation set"), its specific size or proportion is not provided for any dataset, making the full split details incomplete for reproducibility.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments (e.g., GPU models, CPU types, memory specifications).
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., programming languages, libraries, frameworks).
Experiment Setup Yes Adam optimization with a learning rate of 0.0001 and an exponential decay rate of 0.9 is used for the three models. ... Activation functions are relu, except for the B-CNN where tanh leads to better performances on the validation set.