Hyperspherical Prototype Networks

Authors: Pascal Mettes, Elise van der Pol, Cees Snoek

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimentally, we show the benefit of hyperspherical prototype networks for classification, regression, and their combination over other prototype methods, softmax cross-entropy, and mean squared error approaches. (Abstract)
Researcher Affiliation Academia Pascal Mettes ISIS Lab University of Amsterdam Elise van der Pol Uv A-Bosch Delta Lab University of Amsterdam Cees G. M. Snoek ISIS Lab University of Amsterdam
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes The code and prototypes are available at: https://github.com/psmmettes/hpn.
Open Datasets Yes CIFAR-100 consists of 60,000 images of size 32x32 from 100 classes. Image Net-200 is a subset of Image Net, consisting of 110,000 images of size 64x64 from 200 classes [18]. ... Omni Art [38]. ... CUB Birds 200-2011 [42].
Dataset Splits No The paper specifies test set sizes but does not explicitly mention a separate validation set split or how it was used for training purposes.
Hardware Specification No The paper does not provide specific hardware details such as GPU or CPU models used for the experiments.
Software Dependencies No The paper mentions using SGD and standard deep learning components, but does not list specific software dependencies with version numbers (e.g., PyTorch 1.x, Python 3.x).
Experiment Setup Yes For all our experiments, we use SGD, with a learning rate of 0.01, momentum of 0.9, weight decay of 1e-4, batch size of 128, no gradient clipping, and no pre-training. All networks are trained for 250 epochs, where after 100 and 200 epochs, the learning rate is reduced by one order of magnitude. For data augmentation, we perform random cropping and random horizontal flips.