Towards Adaptive Residual Network Training: A Neural-ODE Perspective

Authors: Chengyu Dong, Liyuan Liu, Zichao Li, Jingbo Shang

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

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments, it achieves comparable performance while reducing 50% of training time.
Researcher Affiliation Academia 1University of California, San Diego 2University of Illinois at Urbana-Champaign.
Pseudocode Yes Algorithm 1 presents a generic adaptive training setup. Algorithm 2 Our Lip Grow Algorithm
Open Source Code Yes 1https://github.com/shwinshaker/LipGrow
Open Datasets Yes We conduct experiments on three benchmark datasets, CIFAR-10, CIFAR-100 (Krizhevsky et al., 2009) 5, and Tiny-Image Net6. 5www.cs.toronto.edu/~kriz/cifar.html 6www.kaggle.com/c/tiny-imagenet
Dataset Splits Yes We follow their standard train, validation, and test splits in our experiments.
Hardware Specification Yes evaluated on a single Nvidia Ge Force GTX 1080 Ti GPU. All models on the Tiny-Image Net dataset are trained for 90 epochs, and evaluated on a single Nvidia Quadro RTX 8000 GPU.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup Yes All models on the CIFAR datasets are trained for 164 epochs... We use a batch size of 128 for training, and 100 for validation. Weight decay and momentum are set to be 2 10 4 and 0.9, respectively.