Differentiable Joint Pruning and Quantization for Hardware Efficiency

Wang, Ying; Lu, Yao; Blankevoort, Tijmen

doi:10.1007/978-3-030-58526-6_16

Cited by 46 publications

(41 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Experimental results show that, our method achieves a substantially higher compression rate than recent works, while maintains comparable or better performance. For instance, our method achieves higher classification accuracy and compactness than the recent DJPQ [38]. It gets similar performance with the recent DHP [22] on image super-resolution, meanwhile saves about 50% computation.…”

mentioning

confidence: 60%

“…However, it is hard to implement unstructured pruning with typical hardware. A recent work DJPQ [38] is proposed combining pruning method VIB-net [4,34] and quantization method DQ [35]. Our work shares certain similarity with DJPQ [38] in that, it also leverages DQ [35] for end-to-end training.…”

Section: Related Workmentioning

confidence: 93%

“…Also, our end-to-end training is implemented with a meta-learning framework. DJPQ [38] leverages pruning loss and regularization loss to guide the pruning process and quantization process. Our method manages to unify these two compression processes into the meta-learning framework.…”

Section: Related Workmentioning

confidence: 99%

“…To seek more effective compression strategies, our method focuses on structured pruning and quantization. A simple way to combine pruning and quantization is to directly combine the two methods in a unified framework like [38]. It simply applies two independent pruning and quantization methods trained together for a network, and empirically sets the strength of quantization and pruning.…”

mentioning

confidence: 99%

See 3 more Smart Citations

Hybrid Network Compression via Meta-Learning

Zhang

Wang

2021

Proceedings of the 29th ACM International Conference on Multimedia

View full text Add to dashboard Cite

Neural network pruning and quantization are two major lines of network compression. This raises a natural question that whether we can find the optimal compression by considering multiple network compression criteria in a unified framework. This paper incorporates two criteria and seeks layer-wise compression by leveraging the meta-learning framework. A regularization loss is applied to unify the constraint of input and output channel numbers, bit-width of network activations and weights, so that the compressed network can satisfy a given Bit-OPerations counts (BOPs) constraint. We further propose an iterative compression constraint for optimizing the compression procedure, which effectively achieves a high compression rate and maintains the original network performance. Extensive experiments on various networks and vision tasks show that the proposed method yields better performance and compression rates than recent methods. For instance, our method achieves better image classification accuracy and compactness than the recent DJPQ. It achieves similar performance with the recent DHP in image super-resolution, meanwhile saves about 50% computation. CCS CONCEPTS• Computing methodologies → Computer vision.

show abstract

mentioning

confidence: 60%

Section: Related Workmentioning

confidence: 93%

Section: Related Workmentioning

confidence: 99%

mentioning

confidence: 99%

See 2 more Smart Citations

Hybrid Network Compression via Meta-Learning

Zhang

Wang

2021

Proceedings of the 29th ACM International Conference on Multimedia

View full text Add to dashboard Cite

show abstract

“…Following [9,19,28], OPs is the sum of low-bit operations and floating-point operations, i.e., for 𝑀-bit networks, OPs = BOPs/64…”

Section: Comparison With Existing Sr Quantization Modelsmentioning

confidence: 99%

Fully Quantized Image Super-Resolution Networks

Chen

Zhuang

et al. 2021

Proceedings of the 29th ACM International Conference on Multimedia

View full text Add to dashboard Cite

With the rising popularity of intelligent mobile devices, it is of great practical significance to develop accurate, real-time and energyefficient image Super-Resolution (SR) methods. A prevailing method for improving the inference efficiency is model quantization, which allows for replacing the expensive floating-point operations with efficient bitwise arithmetic operations. To date, it is still challenging for quantized SR frameworks to deliver a feasible accuracyefficiency trade-off. Here, we propose a Fully Quantized image Super-Resolution framework (FQSR) to jointly optimize efficiency and accuracy. In particular, we target obtaining end-to-end quantized models for all layers, especially including skip connections, which was rarely addressed in the literature of SR quantization. We further identify obstacles faced by low-bit SR networks and propose a novel method to counteract them accordingly. The difficulties are caused by 1) for SR task, due to the existence of skip connections, high-resolution feature maps would occupy a huge amount of memory spaces; 2) activation and weight distributions being vastly distinctive in different layers; 3) the inaccurate approximation of the quantization. We apply our quantization scheme on multiple mainstream super-resolution architectures, including SRResNet [16], SRGAN [16] and EDSR [17]. Experimental results show that our FQSR with low-bits quantization is able to achieve on par performance compared with the full-precision counterparts on five benchmark datasets and surpass the state-of-the-art quantized SR methods with significantly reduced computational cost and memory consumption. Code is available at: https://git.io/JWxPp CCS CONCEPTS• Computing methodologies → Computer Vision; • Image processing; • Reconstruction;

show abstract

Model Compression Techniques in Deep Neural Networks

Yesuf,

Assefa

2023

Communications in Computer and Information Science

View full text Add to dashboard Cite

Differentiable Joint Pruning and Quantization for Hardware Efficiency

Cited by 46 publications

References 12 publications

Hybrid Network Compression via Meta-Learning

Hybrid Network Compression via Meta-Learning

Fully Quantized Image Super-Resolution Networks

Model Compression Techniques in Deep Neural Networks

Contact Info

Product

Resources

About