Learning Group Actions on Latent Representations

Authors: Yinzhu Jin, Aman Shrivastava, Tom Fletcher

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our approach on five image datasets with diverse groups acting on them and demonstrate superior performance to recently proposed methods for modeling group actions. 6 Experiments We conduct experiments on five different image datasets.
Researcher Affiliation Academia Yinzhu Jin Department of Computer Science University of Virginia yj3cz@virginia.edu Aman Shrivastava Department of Computer Science University of Virginia as3ek@virginia.edu P. Thomas Fletcher Department of Electrical and Computer Engineering Department of Computer Science University of Virginia ptf8v@virginia.edu
Pseudocode No The paper provides architectural descriptions and figures but no explicit pseudocode or algorithm blocks.
Open Source Code No The architecture and the training was implemented in Py Torch and the code will be made available upon publication.
Open Datasets Yes Rotated MNIST dataset is obtained by simply rotating images from MNIST dataset [6]... Brain MRI dataset is derived from the Human Connectome Project [11]. Neural 3D mesh renderer (NMR) [14] dataset has been used in multiple previous works in the field of novel view synthesis. This dataset is derived from Shape Net [3]... Plane in the sky dataset is our own rendering of Shape Net Core [3] airplane objects.
Dataset Splits Yes We use the original train-test split and image size of 28 28. (Rotated MNIST) among which 880 images are used for training and 233 for testing. (Brain MRI) We randomly split out 20% as the testing set. (Plane in the sky) We randomly split 1/8 of the training set as the validation set.
Hardware Specification Yes Our architecture was trained on 1 A100 GPU with a batch-size of 256 using the Adam optimizer.
Software Dependencies No All our experiments are implemented with the Py Torch [20] package.
Experiment Setup Yes Our architecture was trained on 1 A100 GPU with a batch-size of 256 using the Adam optimizer. The learning rate is 0.0001.