PBGen: Partial Binarization of Deconvolution-Based Generators for Edge Intelligence

Abstract

This work explores the binarization of the deconvolution-based generator in a GAN for memory saving and speedup of image construction. Our study suggests that different from convolutional neural networks (including the discriminator) where all layers can be binarized, only some of the layers in the generator can be binarized without significant performance loss. Supported by theoretical analysis and verified by experiments, a direct metric based on the dimension of deconvolution operations is established, which can be used to quickly decide which layers in the generator can be binarized. Our results also indicate that both the generator and the discriminator should be binarized simultaneously for balanced competition and better performance. Experimental results based on CelebA suggest that directly applying state-of-the-art binarization techniques to all the layers of the generator will lead to 2.83$\times$ performance loss measured by sliced Wasserstein distance compared with the original generator, while applying them to selected layers only can yield up to 25.81$\times$ saving in memory consumption, and 1.96$\times$ and 1.32$\times$ speedup in inference and training respectively with little performance loss.

arXiv comment17 pages, paper re-organized

References in corpus (25)

• TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
• Generative Adversarial Text to Image Synthesis
• Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1
• Conditional Image Synthesis With Auxiliary Classifier GANs
• DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients
• Progressive Growing of GANs for Improved Quality, Stability, and Variation
• Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations
• NIPS 2016 Tutorial: Generative Adversarial Networks
• Ternary Weight Networks
• Trained Ternary Quantization
• Compressing Neural Networks with the Hashing Trick
• Speeding up Convolutional Neural Networks with Low Rank Expansions
• Incremental Network Quantization: Towards Lossless CNNs with Low-Precision Weights
• Convolutional Neural Networks using Logarithmic Data Representation
• Least Squares Generative Adversarial Networks
• XNOR Neural Engine: a Hardware Accelerator IP for 21.6 fJ/op Binary Neural Network Inference
• Connecting Generative Adversarial Networks and Actor-Critic Methods
• Deep Learning with Low Precision by Half-wave Gaussian Quantization
• Constructing Energy-efficient Mixed-precision Neural Networks through Principal Component Analysis for Edge Intelligence
• Quantization of Fully Convolutional Networks for Accurate Biomedical Image Segmentation
• On the Universal Approximability and Complexity Bounds of Quantized ReLU Neural Networks
• Efficient Hybrid Network Architectures for Extremely Quantized Neural Networks Enabling Intelligence at the Edge
• Flexible Network Binarization with Layer-wise Priority

Cited by in corpus (1)

• Real-Time Boiler Control Optimization with Machine Learning