Monomial Matrix Group Equivariant Neural Functional Networks

Authors: Hoang Tran, Thieu Vo, Tho Huu, An Nguyen The, Tan Nguyen

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

Reproducibility Variable Result LLM Response
Research Type Experimental We provide empirical evidences to demonstrate the advantages of our model over existing baselines, achieving competitive performance and efficiency. [...] We evaluate Monomial-NFNs on three tasks: predicting CNN generalization from weights using Small CNN Zoo [64], weight space style editing, and classifying INRs using INRs data [71]. Experimental results show that our model achieves competitive performance and efficiency compared to existing baselines.
Researcher Affiliation Collaboration Viet-Hoang Tran Department of Mathematics National University of Singapore hoang.tranviet@u.nus.edu Thieu N. Vo Department of Mathematics National University of Singapore thieuvo@nus.edu.sg Tho Tran-Huu Department of Mathematics National University of Singapore thotranhuu@u.nus.edu.vn An T. Nguyen FPT Software AI Center annt68@fpt.com Tan M. Nguyen Department of Mathematics National University of Singapore tanmn@nus.edu.sg
Pseudocode No The paper does not contain any sections or figures explicitly labeled 'Pseudocode' or 'Algorithm'.
Open Source Code Yes The code is publicly available at https://github.com/Mathematical AI-NUS/Monomial NFN.
Open Datasets Yes We employ the Small CNN Zoo [64]... We utilize the dataset from [71], which comprises pretrained INR networks [58] that encode images from the CIFAR-10 [36], MNIST [39], and Fashion MNIST [69] datasets.
Dataset Splits Yes The original Re LU subset of the CNN Zoo dataset includes 6050 instances for training and 1513 instances for testing. For the Tanh dataset, it includes 5949 training and 1488 testing instances. [...] We use the Binary Cross Entropy (BCE) loss function and train the model for 50 epochs, with early stopping based on τ on the validation set
Hardware Specification Yes which takes 35 minutes to train on an A100 GPU.
Software Dependencies No The paper mentions using 'Adam' as an optimizer but does not specify version numbers for programming languages, machine learning frameworks (e.g., PyTorch, TensorFlow), or other key software dependencies.
Experiment Setup Yes Hyperparameter settings and the number of parameters can be found in Appendix D. [...] The hyperparameters for our model are presented in Table 18.