Boosted Dynamic Neural Networks

Authors: Haichao Yu, Haoxiang Li, Gang Hua, Gao Huang, Humphrey Shi

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments show it achieves the state-of-the-art performance on CIFAR100 and Image Net datasets in both anytime and budgeted-batch prediction modes.
Researcher Affiliation Collaboration 1University of Illinois Urbana-Champaign 2Wormpex AI Research 3Tsinghua University 4University of Oregon
Pseudocode Yes Algorithm 1: Training pipeline for a Boost Net model.
Open Source Code Yes Our code is released at https://github.com/SHI-Labs/Boosted-Dynamic-Networks.
Open Datasets Yes We do experiments on two datasets, CIFAR100 (Krizhevsky 2009) and Image Net (Deng et al. 2009)
Dataset Splits Yes To calculate the confidence threshold determining whether the model exits or not, we randomly reserve 5000 samples in CIFAR100 and 50000 samples in Image Net as a validation set.
Hardware Specification No The paper states, "We use single GPU with training batch size 64. For Image Net...We use 4 GPUs with training batch size 64 on each GPU." However, it does not provide specific model numbers for the GPUs or other hardware components, which falls under the criteria for
Software Dependencies Yes Our training framework is implemented in Py Torch (Paszke et al. 2019).
Experiment Setup Yes For CIFAR100, the models are trained for 300 epochs with SGD optimizer with momentum 0.9 and initial learning rate 0.1. The learning rate is decayed at epochs 150 and 225. We use single GPU with training batch size 64.