Topological Obstructions and How to Avoid Them

Authors: Babak Esmaeili, Robin Walters, Heiko Zimmermann, Jan-Willem van de Meent

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this paper, we theoretically and empirically characterize obstructions to training encoders with geometric latent spaces... We empirically evaluate our model on 2 domains. We observe improved stability during training and a higher chance of converging to a homeomorphic encoder.
Researcher Affiliation Academia Babak Esmaeili Generative AI Group Eindhoven University of Technology b.esmaeili@tue.nl Robin Walters Khoury College of Computer Sciences Northeastern University r.walters@northeastern.edu Heiko Zimmermann Amsterdam Machine Learning Lab University of Amsterdam h.zimmermann@uva.nl Jan-Willem van de Meent Amsterdam Machine Learning Lab University of Amsterdam j.w.vandemeent@uva.nl
Pseudocode No The paper includes architectural diagrams (Figure 1) but does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described, nor does it include a specific repository link or an explicit code release statement.
Open Datasets Yes We train on images of an L-shaped tetromino [Bozkurt et al., 2021], a teapot, and an airplane [Shilane et al., 2004]. The SO(2) manifold corresponding to each object is made by rotating the image of the object around the center.
Dataset Splits No The paper does not provide specific dataset split information, such as exact percentages, sample counts, or citations to predefined splits, for training, validation, or testing.
Hardware Specification No The paper does not provide specific hardware details, such as exact GPU/CPU models, processor types, or memory amounts, used for running its experiments.
Software Dependencies No The paper mentions using the 'RAdam optimizer [Liu et al., 2019]' but does not provide specific version numbers for any software dependencies or libraries used to replicate the experiment.
Experiment Setup Yes We train all our models for 150 epochs with a batch-size of 600. For optimization, we use the RAdam optimizer [Liu et al., 2019] with a learning rate of 5e-4. All models were initialized and trained with 153 different random seeds. For all image datasets, we use a 4-layer CNN with kernel, stride, and padding of size 4, 2 and 1 respectively followed by a leaky Re LU activation (Table 2) for the encoder and a Sigmoid activation for the decoder.