LieGG: Studying Learned Lie Group Generators

Authors: Artem Moskalev, Anna Sepliarskaia, Ivan Sosnovik, Arnold Smeulders

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments Firstly, we conduct the experiment on the synthetic problem to test the ability of our method to accurately retrieve a symmetry learned by a network. To do so, we adopt the invariant regression task from [14] with clean O(5) symmetry built-in.
Researcher Affiliation Collaboration Artem Moskalev Uv A-Bosch Delta Lab University of Amsterdam a.moskalev@uva.nl Anna Sepliarskaia Machine Learning Research Unit TU Wien seplanna@gmail.com Ivan Sosnovik Uv A-Bosch Delta Lab University of Amsterdam i.sosnovik@uva.nl Arnold Smeulders Uv A-Bosch Delta Lab University of Amsterdam a.w.m.smeulders@uva.nl
Pseudocode No The paper describes steps and calculations in text, but it does not contain structured pseudocode or algorithm blocks with formal labels.
Open Source Code Yes Source code: https://github.com/amoskalev/liegg
Open Datasets Yes In this experiment, we test the ability of our method to retrieve learned symmetries on the rotation MNIST dataset [23].
Dataset Splits Yes Once the network s performance plateaus on the validation split, we terminate the training, and apply our method to retrieve learned symmetries.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python version, library versions like PyTorch or TensorFlow).
Experiment Setup No The paper discusses varying network architectures (number of parameters from [40000, 200000] and depth from 1 to 5 hidden layers) and training until validation convergence. However, it does not explicitly state specific hyperparameter values like learning rate, batch size, or optimizer settings for reproducibility.