Continuous Rotation Group Equivariant Network Inspired by Neural Population Coding

Authors: Zhiqiang Chen, Yang Chen, Xiaolong Zou, Shan Yu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that 1) our approach achieves very competitive performances on MNIST-rot with at least 75% fewer parameters compared with previous SOTA methods, which is efficient in parameter; 2) Especially with small sample sizes, our approach exhibits more pronounced performance improvements (up to 24%); 3) It also has excellent rotation generalization ability on various datasets such as MNIST, CIFAR, and Image Net with both plain and Res Net architectures.
Researcher Affiliation Academia Zhiqiang Chen1,2, Yang Chen2, Xiaolong Zou3, Shan Yu2,4* 1Beijing Academy of Artificial Intelligence, Beijing, China 2Laboratory of Brain Atlas and Brain-inspired Intelligence, Institute of Automation(CASIA), Beijing, China 3Qiyuan Lab, Beijing, China 4School of Artificial Intelligence, University of Chinese Academy of Sciences (UCAS), Beijing, China chenzhiqiang@mails.ucas.ac.cn,yang.chen@ia.ac.cn,xiaolz@mail.pku.edu.cn,shan.yu@nlpr.ia.ac.cn
Pseudocode No The paper includes mathematical formulations and descriptions of its methods but does not provide a formally labeled pseudocode block or algorithm.
Open Source Code Yes https://gitee.com/chenzq/aaai 2024 appendix.git
Open Datasets Yes Firstly, we test the rotation equivariance in common used benchmark MNIST-rot (Larochelle et al. 2007). Secondly, we test the small sample learning ability on MNIST-rot. Finally, we test the rotation generalization ability on MNIST, CIFAR, and Image Net. See more details about datasets on appendix D.
Dataset Splits No The paper describes the usage of training and test sets and their augmentation, but does not explicitly mention or detail a separate 'validation' dataset split within the provided text.
Hardware Specification No The paper does not specify any particular hardware components such as GPU models, CPU types, or cloud computing instances used for running the experiments.
Software Dependencies No The paper does not list any specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes CRGEN-T is a tiny network containing 6 layers with 10 kernels each layer, which is similar to G-CNN (Cohen and Welling 2016). Each kernel has 16 orientation with 3 sparse points (i.e., ml = 3 in Eq. 4), which has trainable locations and orientations. The generated kernels have a spatial size of 5 5. ... For CRGENLN, we set the weight of the equivariant loss λ = 0.1.