Graph Structure of Neural Networks

Authors: Jiaxuan You, Jure Leskovec, Kaiming He, Saining Xie

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Here we systematically investigate how does the graph structure of neural networks affect their predictive performance. Using standard image classification datasets CIFAR-10 and Image Net, we conduct a systematic study on how the architecture of neural networks affects their predictive performance. We make several important empirical observations: A sweet spot of relational graphs lead to neural networks with significantly improved performance;
Researcher Affiliation Collaboration Jiaxuan You 1 Jure Leskovec 1 Kaiming He 2 Saining Xie 2 1Department of Computer Science, Stanford University 2Facebook AI Research. Correspondence to: Jiaxuan You <jiaxuan@cs.stanford.edu>, Saining Xie <s9xie@fb.com>.
Pseudocode No The paper describes methods mathematically and textually but does not include any labeled "Pseudocode" or "Algorithm" blocks.
Open Source Code No The paper does not contain an explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets Yes Using standard image classification datasets CIFAR-10 and Image Net... CIFAR-10 dataset (Krizhevsky, 2009)... Image Net classification (Russakovsky et al., 2015)...
Dataset Splits Yes CIFAR-10 dataset (Krizhevsky, 2009) which has 50K training images and 10K validation images... For Image Net experiments... 1.28M training images and 50K validation images.
Hardware Specification Yes Training an MLP model roughly takes 5 minutes on a NVIDIA Tesla V100 GPU, and training a Res Net model on Image Net roughly takes a day on 8 Tesla V100 GPUs with data parallelism.
Software Dependencies No The paper mentions using a 'cosine learning rate schedule' and 'Batch Norm layer' but does not specify software dependencies with version numbers (e.g., PyTorch, TensorFlow versions or specific library versions).
Experiment Setup Yes We train the model for 200 epochs with batch size 128, using cosine learning rate schedule... with an initial learning rate of 0.1... We train all MLP models with 5 different random seeds... For Image Net experiments... 100 epochs using cosine learning rate schedule with initial learning rate of 0.1. Batch size is 256 for Res Net-family models and 512 for Efficient Net-B0.