Dual Lottery Ticket Hypothesis

Authors: Yue Bai, Huan Wang, ZHIQIANG TAO, Kunpeng Li, Yun Fu

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on several public datasets and comparisons with competitive approaches validate our DLTH as well as the effectiveness of the proposed model RST.
Researcher Affiliation Collaboration 1Northeastern University, Boston, MA, USA 2Santa Clara University, Santa Clara, CA, USA 3Meta Research, Burlingame, CA, USA
Pseudocode No The paper describes the Random Sparse Network Transformation (RST) in text but does not provide any structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/yueb17/DLTH.
Open Datasets Yes Experiments are based on Res Net56/Res Net18 He et al. (2016) on CIFAR10/CIFAR100 Krizhevsky et al. (2009), and a Image Net subset Deng et al. (2009) to compare our method with Lottery Ticket Hypothesis (LTH) Frankle & Carbin (2018) and other strong baselines.
Dataset Splits Yes Experiments are based on Res Net56/Res Net18 He et al. (2016) on CIFAR10/CIFAR100 Krizhevsky et al. (2009), and a Image Net subset Deng et al. (2009). Total number of epochs is 200 with 0.1/0.01/0.001 learning rates starting at 0/100/150 epochs, respectively.
Hardware Specification Yes We use 4 NVIDIA Titan XP GPUs to perform our experimental evaluations.
Software Dependencies No The paper mentions optimization by SGD but does not specify software dependencies like Python, PyTorch, or CUDA with version numbers.
Experiment Setup Yes Experiments on CIFAR10/CIFAR100 are optimized by SGD with 0.9 momentum and 5e-4 weight decay using 128 batch size. Total number of epochs is 200 with 0.1/0.01/0.001 learning rates starting at 0/100/150 epochs, respectively.