| 解码器可以被视为条件概率 <math>Pr(\hat{\mathbf{x}}_{t+1} | \mathbf{y}(t+1))</math> 的生成模型<ref name=":3">Li, S.H.; Wang, L. Neural Network Renormalization Group. Phys. Rev. Lett. 2018, 121, 260601.</ref><ref name=":4">Hu,H.; Wu,D.; You, Y.Z.; Olshausen, B.; Chen, Y. RG-Flow: A hierarchical and explainable flow model based on renormalization group and sparse prior. Mach. Learn. Sci. Technol. 2022, 3, 035009.</ref>,而编码器执行重整化过程。 | | 解码器可以被视为条件概率 <math>Pr(\hat{\mathbf{x}}_{t+1} | \mathbf{y}(t+1))</math> 的生成模型<ref name=":3">Li, S.H.; Wang, L. Neural Network Renormalization Group. Phys. Rev. Lett. 2018, 121, 260601.</ref><ref name=":4">Hu,H.; Wu,D.; You, Y.Z.; Olshausen, B.; Chen, Y. RG-Flow: A hierarchical and explainable flow model based on renormalization group and sparse prior. Mach. Learn. Sci. Technol. 2022, 3, 035009.</ref>,而编码器执行重整化过程。 |
| 有多种方法可以实现可逆神经网络<ref>Teshima, T.; Ishikawa, I.; Tojo, K.; Oono, K.; Ikeda, M.; Sugiyama, M. Coupling-based invertible neural networks are universal diffeomorphism approximators. Adv. Neural Inf. Process. Syst. 2020, 33, 3362–3373.</ref><ref>Teshima, T.; Tojo, K.; Ikeda, M.; Ishikawa, I.; Oono, K. Universal approximation property of neural ordinary differential equations. arXiv 2017, arXiv:2012.02414.</ref>。这里选择如图2所示的RealNVP模块<ref name=":0">Dinh, L.; Sohl-Dickstein, J.; Bengio, S. Density estimation using real nvp. arXiv 2016, arXiv:1605.08803.</ref>来具体实现可逆计算。 | | 有多种方法可以实现可逆神经网络<ref>Teshima, T.; Ishikawa, I.; Tojo, K.; Oono, K.; Ikeda, M.; Sugiyama, M. Coupling-based invertible neural networks are universal diffeomorphism approximators. Adv. Neural Inf. Process. Syst. 2020, 33, 3362–3373.</ref><ref>Teshima, T.; Tojo, K.; Ikeda, M.; Ishikawa, I.; Oono, K. Universal approximation property of neural ordinary differential equations. arXiv 2017, arXiv:2012.02414.</ref>。这里选择如图2所示的RealNVP模块<ref name=":0">Dinh, L.; Sohl-Dickstein, J.; Bengio, S. Density estimation using real nvp. arXiv 2016, arXiv:1605.08803.</ref>来具体实现可逆计算。 |