Adaptive Quantization of Neural Networks

Authors: Soroosh Khoram, Jing Li

ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on MNIST, CIFAR, and SVHN datasets showed that the proposed method can achieve near or better than state-of-the-art reduction in model size with similar error rates.
Researcher Affiliation Academia Soroosh Khoram Department of Electrical and Computer Engineering University of Wisconsin Madison khoram@wisc.edu Jing Li Department of Electrical and Computer Engineering University of Wisconsin Madison jli@ece.wisc.edu
Pseudocode Yes Algorithm 1 Quantization of a parameter; Algorithm 2 Adaptive Quantization; Algorithm 3 Choosing hyper-parameters
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository.
Open Datasets Yes We use MNIST (Le Cun et al., 1998), CIFAR-10 (Krizhevsky & Hinton, 2009), and SVHN (Netzer et al., 2011) benchmarks in our experiments.
Dataset Splits No The paper refers to a 'training set' for loss calculation and uses standard benchmarks, but does not explicitly provide specific percentages or counts for training, validation, and test splits, nor does it detail how these splits are managed for reproducibility.
Hardware Specification Yes We implement the proposed quantization on Intel Core i7 CPU (3.5 GHz) with Titan X GPU performing training and quantization.
Software Dependencies No The paper does not list specific software dependencies (e.g., libraries, frameworks, or programming languages) with version numbers that would be necessary for reproduction.
Experiment Setup Yes In our experiments in the next section Scale and Steps are set to 1.1 and 20, respectively.