Deep Neural Network Quantization via Layer-Wise Optimization Using Limited Training Data

Authors: Shangyu Chen, Wenya Wang, Sinno Jialin Pan3329-3336

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on benchmark deep models are conducted to demonstrate the effectiveness of our proposed method using 1% of CIFAR10 and Image Net datasets.
Researcher Affiliation Academia Shangyu Chen,1 Wenya Wang,1 Sinno Jialin Pan1 1Nanyang Technological University schen025@e.ntu.edu.sg, wangwy@ntu.edu.sg, sinnopan@ntu.edu.sg
Pseudocode Yes Algorithm 1 Layer-wise Unsupervised Network Quantization
Open Source Code Yes Codes are available in: https://github.com/csyhhu/L-DNQ
Open Datasets Yes Two benchmark datasets are used including Image Net ILSVRC-2012 and CIFAR-10.
Dataset Splits Yes 500 training instances in CIFAR-10 and 12,800 in Image Net are randomly sampled to simulate the scenario of limited instances. ... For fair comparison with training-based quantization, we reduce training data to 1% of the original training dataset.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU model, CPU type) used for running the experiments.
Software Dependencies No The paper mentions general tools and frameworks (e.g., 'deep learning framework'), but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes 500 training instances in CIFAR-10 and 12,800 in Image Net are randomly sampled to simulate the scenario of limited instances. All experiments are conducted 5 times and the average result is reported. ... For fair comparison with training-based quantization, we reduce training data to 1% of the original training dataset. ... L-DNQ adopts the following quantization intervals: Ωl = αl {0, 20, 21, 22... 2b} for each layer.