Calibrating the Rigged Lottery: Making All Tickets Reliable

Authors: Bowen Lei, Ruqi Zhang, Dongkuan Xu, Bani Mallick

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on multiple datasets, model architectures, and sparsities show that our method reduces ECE values by up to 47.8% and simultaneously maintains or even improves accuracy with only a slight increase in computation and storage burden.
Researcher Affiliation Academia Bowen Lei Texas A&M University bowenlei@stat.tamu.edu Dongkuan Xu North Carolina State University dxu27@ncsu.edu Ruqi Zhang Purdue University ruqiz@purdue.edu Bani Mallick Texas A&M University bmallick@stat.tamu.edu
Pseudocode Yes Algorithm 1 Cig L
Open Source Code Yes The implementation code can be found in https://github.com/Steven Boys/Cig L.
Open Datasets Yes Our experiments are based on three benchmark datasets: CIFAR-10 and CIFAR100 (Krizhevsky et al., 2009) and Image Net-2012 (Russakovsky et al., 2015).
Dataset Splits No The paper mentions training models and using a validation set for temperature scaling but does not specify the train/validation/test dataset splits (e.g., percentages or sample counts) for its experiments.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions "Py Torch" as the code platform but does not specify a version number or other software dependencies with their versions.
Experiment Setup Yes The parameters are optimized by SGD with momentum. For the learning rate, we use piecewise constant decay scheduler. For CIFAR-10 and CIFAR-100, we train all the models for 250 epochs with a batch size of 128. For Image Net, we train all the models for 100 epochs with a batch size of 64.