Geometric Autoencoders - What You See is What You Decode

Authors: Philipp Nazari, Sebastian Damrich, Fred A Hamprecht

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

Reproducibility Variable Result LLM Response
Research Type Experimental 5. Experiments 5.1. Experimental Setup 5.2. Evaluation
Researcher Affiliation Academia 1HCI/IWR at University of Heidelberg, 69120 Heidelberg, Germany 2University of T ubingen, 72074 T ubingen, Germany.
Pseudocode Yes Algorithm 1 Calculating the Generalized Jacobian Determinant
Open Source Code Yes We provide the code as an open-source package for Py Torch. It can be found at https://github.com/hci-unihd/ Geometric Autoencoder.
Open Datasets Yes Datasets Besides the classical image datasets MNIST (Le Cun et al., 1998) and Fashion MNIST (Xiao et al., 2017), we use the three single-cell datasets Zilionis (Zilionis et al., 2019), CElegans (Packer et al., 2019) and PBMC (Zheng et al., 2017).
Dataset Splits No The paper does not explicitly mention a validation dataset split or a methodology for using one during training.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory) used for running experiments.
Software Dependencies No The paper mentions 'Py Torch', 'Geomstats package', and 'functorch library' but does not specify their version numbers.
Experiment Setup Yes All of the autoencoders except for the UMAP autoencoder are optimized using ADAM (Kingma & Ba, 2015), and trained using a batch size of 125, learning rate 10 3 and a weight decay of 10 5. ... The vanilla, topological and geometric autoencoders are trained for 100 epochs. For the proposed geometric autoencoder, we found α = 0.1 to be a good weight for the geometric loss term.