Few Shot Network Compression via Cross Distillation

Authors: Haoli Bai, Jiaxiang Wu, Irwin King, Michael Lyu3203-3210

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments n benchmark datasets demonstrate that cross distillation can significantly improve the student network s accuracy when only a few training instances are available.
Researcher Affiliation Collaboration 1The Chinese University of Hong Kong, 2Tencent AI Lab
Pseudocode Yes Algorithm 1 Cross distillation
Open Source Code Yes Our implementation in Py Torch is available at https://github.com/haolibai/ Cross-Distillation.git.
Open Datasets Yes evaluations are performed on CIFAR-10 and Image Net-ILSVRC12.
Dataset Splits Yes As we consider the setting of few shot image classification, we randomly select K-shot instances per class from the training set. All experiments are averaged over five runs with different random seeds, and results of means and standard deviations are reported.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU models, CPU types) used for running its experiments.
Software Dependencies No The paper mentions 'Py Torch' but does not specify a version number or other software dependencies with versions.
Experiment Setup Yes For Ours, we choose μ = 0.6 for VGG networks and μ = 0.9 for Res Nets. For Ours-S, we set (α, β) = (0.9, 0.3) on VGG networks and (0.9, 0.5) on Res Nets. Sensitivity analysis on these hyper-parameters are presented later.