近年來手機處理器的大幅進展讓在移動裝置上使用深度學習網路更加有機會，也因此產生了許多以深度學習生成對抗網路為名的修圖應用。但要運行大型的生成對抗網路需要運算量對多數移動裝置而言仍然過於巨大。本篇論文我們提出了一個利用微生物基因演算法的方法來達成可調整式漸進生長的生成對抗網路，使其能在符合硬體限制的運算成本下生成擬真影像如人臉。實際使用時，生成器先隨機編碼成數個二元字串為一個群體，再以基因演算法做優化到一個穩定的狀態，接著使用微生物基因演算法找出學習好的重要的卷積核，接著這個學習好的生成對抗網路的大小就能以改變那些卷積層結構的方式被調整。實驗顯示我們的方法在各種壓縮率下都能達到不錯的表現。

Recent advances in mobile processors made it feasible to employ deep learning networks to mobile devices. Because of this, many image editing applications based on “deep learning Generative Adversarial Networks (GANs)”techniques were born. However, the computational cost for a big GAN model is still too high for general mobile devices. In this paper, we developed an adjustable generative adversarial network via microbial genetic algorithm. The proposed adjustable GAN algorithm is able to generate realistic images, such as human faces, at various computational costs conditioning on different hardware limits. In this paper, the proposed generator is first encoded as a binary string representing a population and optimized iteratively via genetic algorithm to reach an initial stable state, then a microbial genetic algorithm is utilized to discover the importance level of learned convolution filters. Afterward, the learned GAN model size can be further adjusted by editing the structure of those convolution filters. Extensive experiments demonstrate that the proposed method is effective in various model sizes while achieving satisfactory results.

[1] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley,
Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In
NIPS, pages 2672–2680, 2014.
[2] Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, ´ Andrew Cunningham, Alejandro Acosta, Andrew P Aitken, Alykhan Tejani, Johannes Totz, Zehan
Wang, et al. Photorealistic single image super-resolution using a generative adversarial network. In CVPR, volume 2, page 4, 2017.
[3] Jun-Yan Zhu, Philipp Krahenb ¨ uhl, Eli Shechtman, and ¨ Alexei A Efros. Generative visual manipulation on the natural image manifold. In ECCV, pages 597–613, 2016.
[4] Radford, A., L. Metz, and S. Chintala, Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. CoRR, 2015.abs/1511.06434.
[5] T. Karras, T. Aila, S. Laine, and J. Lehtinen. Progressive growing of GANs for improved quality, stability, and variation. CoRR, abs/1710.10196, 2017
[6] Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and TAMT Meyarivan. A fast and
elitist multiobjective genetic algorithm: Nsga-ii. IEEE transactions on evolutionary computation, 6(2):182–197, 2002.
[7] Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired imageto-image translation using cycle-consistent adversarial networks. In ICCV, pages
2223–2232, 2017.
[8] Harvey, I. (2011). The microbial genetic algorithm. In Advances in artificial life. Darwin Meets von Neumann, pages 126–133. Springer.
[9] Yunhe Wang, Chang Xu, Jiayan Qiu, Chao Xu, and Dacheng Tao. Towards evolutional compression. In SIGKDD, 2018.
[10] Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and TAMT Meyarivan. A fast and
elitist multiobjective genetic algorithm: Nsga-ii. IEEE transactions on evolutionary computation, 6(2):182–197, 2002.
[11] H. Shu, Y. Wang, X. Jia, K. Han, H. Chen, C. Xu, Q. Tian, and C. Xu “Coevolutionary compression for unpaired image translation,”in ICCV, 2019.
[12] Rosenblatt, Frank. ”The Perceptron—a perceiving and recognizing automaton”. Report 85-460-1. Cornell Aeronautical Laboratory.
[13] Huang, Xun, and Serge Belongie. “Arbitrary Style Transfer in Real-Time with
Adaptive Instance Normalization.”In ICCV, 2017.
[14] Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE CVPR, pages 4401–
4410, 2019.