# Conditional Deep Gaussian Processes: Empirical Bayes Hyperdata Learning

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Related Work

## 3. Background

#### 3.1. Gaussian Process

#### 3.2. Deep Gaussian Process

#### 3.3. Marginal Prior, Covariance and Marginal Likelihood

## 4. Model

#### 4.1. Conditional Deep Gaussian Process

**Lemma**

**1.**

#### 4.2. When Conditional DGP Is Almost a GP

**Lemma**

**2.**

**Proof.**

**Remark**

**1.**

#### 4.3. Non-Gaussian Aspect

## 5. Results

#### 5.1. Mauna Loa Data

#### 5.2. Airline Data

## 6. Discussion

## 7. Conclusions

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## Abbreviations

GP | Gaussian Process |

DGP | Deep Gaussian Process |

DKL | Deep Kernel Learning |

SE | Squared Exponential |

## Appendix A

**Lemma**

**A1.**

**Proof.**

## References

- Damianou, A.; Lawrence, N. Deep gaussian processes. In Proceedings of the Artificial Intelligence and Statistics, Scottsdale, AZ, USA, 29 April–1 May 2013; pp. 207–215. [Google Scholar]
- Goodfellow, I.; Bengio, Y.; Courville, A. Deep Learning; MIT Press: Cambridge, MA, USA, 2016. [Google Scholar]
- Rasmussen, C.E.; Williams, C.K.I. Gaussian Process for Machine Learning; MIT Press: Cambridge, MA, USA, 2006. [Google Scholar]
- Minka, T.P. Expectation propagation for approximate Bayesian inference. In Proceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence, Seattle, WA, USA, 2–5 August 2001; Morgan Kaufmann Publishers Inc.: Burlington, MA, USA, 2001; pp. 362–369. [Google Scholar]
- Bui, T.; Hernández-Lobato, D.; Hernandez-Lobato, J.; Li, Y.; Turner, R. Deep Gaussian processes for regression using approximate expectation propagation. In Proceedings of the International Conference on Machine Learning, New York, NY, USA, 20–22 June 2016; pp. 1472–1481. [Google Scholar]
- Salimbeni, H.; Deisenroth, M. Doubly stochastic variational inference for deep gaussian processes. In Proceedings of the Advances in Neural Information Processing Systems, Long Beach, CA, USA, 4–9 December 2017. [Google Scholar]
- Salimbeni, H.; Dutordoir, V.; Hensman, J.; Deisenroth, M.P. Deep Gaussian Processes with Importance-Weighted Variational Inference. arXiv
**2019**, arXiv:1905.05435. [Google Scholar] - Yu, H.; Chen, Y.; Low, B.K.H.; Jaillet, P.; Dai, Z. Implicit Posterior Variational Inference for Deep Gaussian Processes. In Proceedings of the Advances in Neural Information Processing Systems, Vancouver, BC, Canada, 8–14 December 2019; pp. 14502–14513. [Google Scholar]
- Ustyuzhaninov, I.; Kazlauskaite, I.; Kaiser, M.; Bodin, E.; Campbell, N.; Ek, C.H. Compositional uncertainty in deep Gaussian processes. In Proceedings of the Conference on Uncertainty in Artificial Intelligence, Virtual, 3–6 August 2020; pp. 480–489. [Google Scholar]
- Havasi, M.; Hernández-Lobato, J.M.; Murillo-Fuentes, J.J. Inference in deep Gaussian processes using stochastic gradient Hamiltonian Monte Carlo. In Proceedings of the Advances in Neural Information Processing Systems, Montreal, QC, Canada, 3–8 December 2018; pp. 7506–7516. [Google Scholar]
- Duvenaud, D.; Rippel, O.; Adams, R.; Ghahramani, Z. Avoiding pathologies in very deep networks. In Proceedings of the Artificial Intelligence and Statistics, Reykjavik, Iceland, 22–25 April 2014; pp. 202–210. [Google Scholar]
- Dunlop, M.M.; Girolami, M.A.; Stuart, A.M.; Teckentrup, A.L. How deep are deep Gaussian processes? J. Mach. Learn. Res.
**2018**, 19, 2100–2145. [Google Scholar] - Tong, A.; Choi, J. Characterizing Deep Gaussian Processes via Nonlinear Recurrence Systems. In Proceedings of the AAAI Conference on Artificial Intelligence, Vancouver, BC, Canada, 2–9 February 2021; Volume 35, pp. 9915–9922. [Google Scholar]
- Agrawal, D.; Papamarkou, T.; Hinkle, J.D. Wide Neural Networks with Bottlenecks are Deep Gaussian Processes. J. Mach. Learn. Res.
**2020**, 21, 1–66. [Google Scholar] - Pleiss, G.; Cunningham, J.P. The Limitations of Large Width in Neural Networks: A Deep Gaussian Process Perspective. arXiv
**2021**, arXiv:2106.06529. [Google Scholar] - Wilson, A.G.; Hu, Z.; Salakhutdinov, R.; Xing, E.P. Deep kernel learning. In Proceedings of the Artificial Intelligence and Statistics, Cadiz, Spain, 9–11 May 2016; pp. 370–378. [Google Scholar]
- Salakhutdinov, R.; Hinton, G.E. Using Deep Belief Nets to Learn Covariance Kernels for Gaussian Processes. Citeseer
**2007**, 7, 1249–1256. [Google Scholar] - Calandra, R.; Peters, J.; Rasmussen, C.E.; Deisenroth, M.P. Manifold Gaussian processes for regression. In Proceedings of the 2016 International Joint Conference on Neural Networks (IJCNN), Vancouver, BC, Canada, 24–29 July 2016; pp. 3338–3345. [Google Scholar]
- Ober, S.W.; Rasmussen, C.E.; van der Wilk, M. The promises and pitfalls of deep kernel learning. arXiv
**2021**, arXiv:2102.12108. [Google Scholar] - Titsias, M. Variational learning of inducing variables in sparse Gaussian processes. In Proceedings of the Artificial Intelligence and Statistics, Clearwater Beach, FL, USA, 16–18 April 2009; pp. 567–574. [Google Scholar]
- Titsias, M.; Lawrence, N. Bayesian Gaussian process latent variable model. In Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, Sardinia, Italy, 13–15 May 2010; pp. 844–851. [Google Scholar]
- Matthews, A.G.d.G.; Hensman, J.; Turner, R.; Ghahramani, Z. On sparse variational methods and the Kullback–Leibler divergence between stochastic processes. In Proceedings of the Artificial Intelligence and Statistics, Cadiz, Spain, 9–11 May 2016; pp. 231–239. [Google Scholar]
- Lu, C.K.; Yang, S.C.H.; Hao, X.; Shafto, P. Interpretable deep Gaussian processes with moments. In Proceedings of the International Conference on Artificial Intelligence and Statistics, Virtual, 26–28 August 2020; pp. 613–623. [Google Scholar]
- Murphy, K.P. Machine Learning: A Probabilistic Perspective; MIT Press: Cambridge, MA, USA, 2012. [Google Scholar]
- Kingma, D.P.; Welling, M. Auto-encoding variational bayes. arXiv
**2013**, arXiv:1312.6114. [Google Scholar] - Garnelo, M.; Rosenbaum, D.; Maddison, C.; Ramalho, T.; Saxton, D.; Shanahan, M.; Teh, Y.W.; Rezende, D.; Eslami, S.A. Conditional Neural Processes. In Proceedings of the International Conference on Machine Learning, Stockholm, Sweden, 10–15 July 2018; pp. 1704–1713. [Google Scholar]
- Snelson, E.; Ghahramani, Z.; Rasmussen, C.E. Warped gaussian processes. In Proceedings of the Advances in Neural Information Processing Systems, Vancouver, BC, Canada, 13–18 December 2004; pp. 337–344. [Google Scholar]
- Lázaro-Gredilla, M. Bayesian warped Gaussian processes. In Proceedings of the Advances in Neural Information Processing Systems, Lake Tahoe, NV, USA, 3–6 December 2012; pp. 1619–1627. [Google Scholar]
- Ma, C.; Li, Y.; Hernández-Lobato, J.M. Variational implicit processes. In Proceedings of the International Conference on Machine Learning, Long Beach, CA, USA, 9–15 June 2019; pp. 4222–4233. [Google Scholar]
- Ustyuzhaninov, I.; Kazlauskaite, I.; Ek, C.H.; Campbell, N. Monotonic gaussian process flows. In Proceedings of the International Conference on Artificial Intelligence and Statistics, Sicily, Italy, 26–28 August 2020; pp. 3057–3067. [Google Scholar]
- Telgarsky, M. Benefits of depth in neural networks. In Proceedings of the Conference on Learning Theory, New York, NY, USA, 23–26 June 2016; pp. 1517–1539. [Google Scholar]
- Pearce, T.; Tsuchida, R.; Zaki, M.; Brintrup, A.; Neely, A. Expressive priors in bayesian neural networks: Kernel combinations and periodic functions. In Proceedings of the Uncertainty in Artificial Intelligence, Virtual, 3–6 August 2020; pp. 134–144. [Google Scholar]
- Gal, Y.; Ghahramani, Z. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In Proceedings of the International Conference on Machine Learning, New York, NY, USA, 20–22 June 2016; pp. 1050–1059. [Google Scholar]
- Cutajar, K.; Bonilla, E.V.; Michiardi, P.; Filippone, M. andom feature expansions for deep Gaussian processes. In Proceedings of the 34th International Conference on Machine Learning-Volume 70. JMLR. org, Sydney, Australia, 6–11 August 2017; pp. 884–893. [Google Scholar]
- Rahimi, A.; Recht, B. Random features for large-scale kernel machines. In Proceedings of the Advances in Neural Information Processing Systems, Vancouver, BC, Canada, 8–11 December 2008; pp. 1177–1184. [Google Scholar]
- Cho, Y.; Saul, L.K. Kernel methods for deep learning. In Proceedings of the Advances in Neural Information Processing Systems, Vancouver, BC, Canada, 7–10 December 2009; pp. 342–350. [Google Scholar]
- Schölkopf, B.; Smola, A.; Müller, K.R. Nonlinear component analysis as a kernel eigenvalue problem. Neural Comput.
**1998**, 10, 1299–1319. [Google Scholar] [CrossRef][Green Version] - Girard, A.; Rasmussen, C.E.; Candela, J.Q.; Murray-Smith, R. Gaussian process priors with uncertain inputs application to multiple-step ahead time series forecasting. In Proceedings of the Advances in Neural Information Processing Systems, Vancouver, BC, Canada, 8–13 December 2003; pp. 545–552. [Google Scholar]
- Muandet, K.; Fukumizu, K.; Dinuzzo, F.; Schölkopf, B. Learning from distributions via support measure machines. In Proceedings of the Advances in Neural Information Processing Systems, Lake Tahoe, NV, USA, 3–6 December 2012; pp. 10–18. [Google Scholar]
- Li, Y.; Swersky, K.; Zemel, R. Generative moment matching networks. In Proceedings of the International Conference on Machine Learning, Lille, France, 7–9 July 2015; pp. 1718–1727. [Google Scholar]
- Quiñonero-Candela, J.; Rasmussen, C.E. A unifying view of sparse approximate Gaussian process regression. J. Mach. Learn. Res.
**2005**, 6, 1939–1959. [Google Scholar] - Shi, J.; Titsias, M.K.; Mnih, A. Sparse Orthogonal Variational Inference for Gaussian Processes. arXiv
**2019**, arXiv:1910.10596. [Google Scholar] - Lázaro-Gredilla, M.; Figueiras-Vidal, A.R. Inter-domain Gaussian Processes for Sparse Inference using Inducing Features. Citeseer
**2009**, 22, 1087–1095. [Google Scholar] - Dutordoir, V.; Durrande, N.; Hensman, J. Sparse Gaussian processes with spherical harmonic features. In Proceedings of the International Conference on Machine Learning, Virtual, 13–18 July 2020; pp. 2793–2802. [Google Scholar]
- Dutordoir, V.; Hensman, J.; van der Wilk, M.; Ek, C.H.; Ghahramani, Z.; Durrande, N. Deep Neural Networks as Point Estimates for Deep Gaussian Processes. arXiv
**2021**, arXiv:2105.04504. [Google Scholar] - Rudner, T.G.; Sejdinovic, D.; Gal, Y. Inter-domain deep Gaussian processes. In Proceedings of the International Conference on Machine Learning, Virtual, 13–18 July 2020; pp. 8286–8294. [Google Scholar]
- Ober, S.W.; Aitchison, L. Global inducing point variational posteriors for bayesian neural networks and deep gaussian processes. In Proceedings of the International Conference on Machine Learning, Virtual, 18–24 July 2021; pp. 8248–8259. [Google Scholar]
- GPy. GPy: A Gaussian Process Framework in Python. 2012. Available online: http://github.com/SheffieldML/GPy (accessed on 15 October 2021).
- Lu, C.K.; Shafto, P. Conditional Deep Gaussian Process: Multi-fidelity kernel learning. arXiv
**2021**, arXiv:2002.02826. [Google Scholar] - Isserlis, L. On a formula for the product-moment coefficient of any order of a normal frequency distribution in any number of variables. Biometrika
**1918**, 12, 134–139. [Google Scholar] [CrossRef][Green Version] - Vladimirova, M.; Verbeek, J.; Mesejo, P.; Arbel, J. Understanding priors in bayesian neural networks at the unit level. In Proceedings of the International Conference on Machine Learning, Long Beach, CA, USA, 9–15 June 2019; pp. 6458–6467. [Google Scholar]
- Yaida, S. Non-Gaussian processes and neural networks at finite widths. In Proceedings of the First Mathematical and Scientific Machine Learning Conference, Virtual, 20–24 July 2020; pp. 165–192. [Google Scholar]
- Zavatone-Veth, J.; Pehlevan, C. Exact marginal prior distributions of finite Bayesian neural networks. arXiv
**2021**, arXiv:2104.11734. [Google Scholar] - Duvenaud, D.; Lloyd, J.; Grosse, R.; Tenenbaum, J.; Zoubin, G. Structure discovery in nonparametric regression through compositional kernel search. In Proceedings of the International Conference on Machine Learning, Atlanta, GA, USA, 17–19 June 2013; pp. 1166–1174. [Google Scholar]
- Sun, S.; Zhang, G.; Wang, C.; Zeng, W.; Li, J.; Grosse, R. Differentiable compositional kernel learning for Gaussian processes. In Proceedings of the International Conference on Machine Learning, Stockholm, Sweden, 10–15 July 2018; pp. 4828–4837. [Google Scholar]
- Mhaskar, H.; Liao, Q.; Poggio, T. Learning functions: When is deep better than shallow. arXiv
**2016**, arXiv:1603.00988. [Google Scholar] - Dutordoir, V.; Salimbeni, H.; Hambro, E.; McLeod, J.; Leibfried, F.; Artemev, A.; van der Wilk, M.; Deisenroth, M.P.; Hensman, J.; John, S. Pflux: A library for Deep Gaussian Processes. arXiv
**2021**, arXiv:2104.05674. [Google Scholar]

**Figure 1.**Extrapolation of standardized CO

_{2}time series data (yellow dots for training and red dots for test) using GP with three kernels. The dark solid line represents the predictive mean, and the shaded area is the the model’s confidence. Panel (

**a**) displays the result using a single GP with an SE kernel. Panel (

**b**) was obtained following the kernel composition in [3]. Panel (

**c**) came from using the effective kernel of 2-layer zero-mean DGP with SE used in both layers [23]. (

**a**) SE kernel; (

**b**) SE+periodic SE+RQ kernel; (

**c**) SE[SE] kernel.

**Figure 2.**Extrapolation of standardized CO

_{2}using DKL and variational inference [6] for the DGP implemented in GPFlux [57]. Panel (

**a**) was obtained using the DKL with three-layer RELU network. Panel (

**b**) shows the results from the two-layer zero-mean DGP model. Panel (

**c**) shows the results of the three-layer zero-mean DGP. (

**a**) DKL; (

**b**) Two-layer DGP; (

**c**) Three-layer DGP.

**Figure 3.**Extrapolation of the standardized CO

_{2}using conditional DGP. Panel (

**a**) is for the two-layer model, and (b) for the three-layer model. Top and middle panels shows the mean and confidence in the posterior over the latent functions. See text for details. (

**a**) Two-layer conditional DGP; (

**b**) Three-layer conditional DGP.

**Figure 4.**Extrapolation of the standardized airline data with three different GPs. (

**a**) SE kernel; (

**b**) SE+periodic SE+RQ kernel; (

**c**) SE[SE] kernel.

**Figure 5.**Extrapolation of the standardized airline data using DKL (

**a**), 2-layer DGP (

**b**) and 3-layer DGP (c). (

**a**) DKL; (

**b**) Two-layer DGP; (c) Three-layer DGP.

**Figure 6.**Extrapolation of airline data using conditional DGP. The upper panel shows the learned latent function and uncertainty from hyperdata learning, and the bottom panel shows the extrapolation from the past data. (

**a**) The first model had 23 hyperdata supporting the latent GP. (

**b**) The other model had 13. (

**a**) 2-layer cDGP with 23 hyperdata; (

**b**) 2-layer cDGP with 13 hyperdata.

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Lu, C.-K.; Shafto, P.
Conditional Deep Gaussian Processes: Empirical Bayes Hyperdata Learning. *Entropy* **2021**, *23*, 1387.
https://doi.org/10.3390/e23111387

**AMA Style**

Lu C-K, Shafto P.
Conditional Deep Gaussian Processes: Empirical Bayes Hyperdata Learning. *Entropy*. 2021; 23(11):1387.
https://doi.org/10.3390/e23111387

**Chicago/Turabian Style**

Lu, Chi-Ken, and Patrick Shafto.
2021. "Conditional Deep Gaussian Processes: Empirical Bayes Hyperdata Learning" *Entropy* 23, no. 11: 1387.
https://doi.org/10.3390/e23111387