Differentiating through the Fréchet Mean

Authors: Aaron Lou, Isay Katsman, Qingxuan Jiang, Serge Belongie, Ser-Nam Lim, Christopher De Sa

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

Reproducibility Variable Result LLM Response
Research Type Experimental To demonstrate this integration, we present two case studies. First, we apply our Fr echet mean to the existing Hyperbolic Graph Convolutional Network, replacing its projected aggregation to obtain state-of-the-art results on datasets with high hyperbolicity.
Researcher Affiliation Collaboration 1Department of Computer Science, Cornell University, NY, Ithaca, USA 2Facebook AI, NY, New York, USA.
Pseudocode Yes Algorithm 1 Poincar e model Fr echet mean algorithm
Open Source Code Yes Our Py Torch implementation of the differentiable Fr echet mean can be found at https://github.com/CUVL/Differentiable Frechet-Mean.
Open Datasets Yes Our new aggregation allows us to achieve new state-of-the-art results on the Disease and Disease-M graph datasets (Chami et al., 2019). ... On the rather non-hyperbolic Co RA (Sen et al., 2008) dataset...
Dataset Splits Yes we test with precisely the same hyperparameters (learning rate, test/val split, and the like) as Chami et al. (2019) for a fair comparison.
Hardware Specification Yes Experiments were run with an Intel Skylake Core i7-6700HQ 2.6 GHz Quad core CPU.
Software Dependencies No The paper mentions a 'Py Torch implementation' but does not provide specific version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes We use precisely the same architecture as in Chami et al. (2019), except we substitute all hyperbolic aggregation layers with our differentiable Fr echet mean layer. Furthermore, we test with precisely the same hyperparameters (learning rate, test/val split, and the like) as Chami et al. (2019) for a fair comparison.