Auto-Encoding Transformations in Reparameterized Lie Groups for Unsupervised Learning

Authors: Feng Lin, Haohang Xu, Houqiang Li, Hongkai Xiong, Guo-Jun Qi8610-8617

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments demonstrate the proposed approach to Auto-Encoding Transformations exhibits superior performances on a variety of recognition problems. In this section, we present our experiment results by comparing the AETv2 with the AETv1 as well as the other unsupervised models.
Researcher Affiliation Collaboration Feng Lin,1 Haohang Xu,2 Houqiang Li,1,4 Hongkai Xiong,2 Guo-Jun Qi3,* 1 CAS Key Laboratory of GIPAS, EEIS Department, University of Science and Technology of China 2 Department of Electronic Engineering, Shanghai Jiao Tong University 3 Laboratory for MAPLE, Futurewei Technologies 4 Institute of Artificial Intelligence, Hefei Comprehensive National Science Center
Pseudocode No The paper does not include a structured pseudocode or algorithm block. The methodology is described in text and through a pipeline diagram.
Open Source Code No The paper does not provide an explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes Following the standard evaluation protocol in literature (Zhang et al. 2019; Qi et al. 2019; Oyallon and Mallat 2015; Dosovitskiy et al. 2014; Radford, Metz, and Chintala 2015; Oyallon, Belilovsky, and Zagoruyko 2017; Gidaris, Singh, and Komodakis 2018), we will adopt downstream classification tasks to evaluate the learned representations on CIFAR10, Image Net, and Places datasets.
Dataset Splits Yes Following the standard evaluation protocol in literature (Zhang et al. 2019; Qi et al. 2019; Oyallon and Mallat 2015; Dosovitskiy et al. 2014; Radford, Metz, and Chintala 2015; Oyallon, Belilovsky, and Zagoruyko 2017; Gidaris, Singh, and Komodakis 2018), we will adopt downstream classification tasks to evaluate the learned representations on CIFAR10, Image Net, and Places datasets. A classifier is then built on top of the second convolutional block to evaluate the quality of the learned representation following the standard protocol in literature (Zhang et al. 2019; Qi et al. 2019; Oyallon and Mallat 2015; Dosovit-skiy et al. 2014; Radford, Metz, and Chintala 2015; Oyallon, Belilovsky, and Zagoruyko 2017; Gidaris, Singh, and Komodakis 2018).
Hardware Specification No The paper does not provide specific hardware details such as GPU or CPU models, memory, or cloud computing instances used for running the experiments.
Software Dependencies No The paper mentions software components like the 'Adam solver' but does not provide specific version numbers for any software dependencies required to replicate the experiment.
Experiment Setup Yes The model is trained by the Adam solver with a learning rate of 10 5, a value of 0.9 and 0.999 for β1 and β2, and a weight decay rate of 5 10 4. ...train the network with a batch size of 768 original and transformed images.