Gaussian Process Surrogate Models for Neural Networks

被引:0
|
作者
Li, Michael Y. [1 ]
Grant, Erin [2 ]
Griffiths, Thomas L. [3 ]
机构
[1] Stanford Univ, Dept Comp Sci, Stanford, CA 94305 USA
[2] UCL, Gatsby Computat Neurosci Unit, London, England
[3] Dept Psychol & Comp Sci, Princeton, NJ USA
来源
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Not being able to understand and predict the behavior of deep learning systems makes it hard to decide what architecture and algorithm to use for a given problem. In science and engineering, modeling is a methodology used to understand complex systems whose internal processes are opaque. Modeling replaces a complex system with a simpler, more interpretable surrogate. Drawing inspiration from this, we construct a class of surrogate models for neural networks using Gaussian processes. Rather than deriving kernels for infinite neural networks, we learn kernels empirically from the naturalistic behavior of finite neural networks. We demonstrate our approach captures existing phenomena related to the spectral bias of neural networks, and then show that our surrogate models can be used to solve practical problems such as identifying which points most influence the behavior of specific neural networks and predicting which architectures and algorithms will generalize well for specific datasets.
引用
收藏
页码:1241 / 1252
页数:12
相关论文
共 50 条
  • [31] A Comparison of Acoustic Models Based on Neural Networks and Gaussian Mixtures
    Pavelka, Tomas
    Ekstein, Kamil
    TEXT, SPEECH AND DIALOGUE, PROCEEDINGS, 2009, 5729 : 291 - 298
  • [32] The Gaussian equivalence of generative models for learning with shallow neural networks
    Goldt, Sebastian
    Loureiro, Bruno
    Reeves, Galen
    Krzakala, Florent
    Mezard, Marc
    Zdeborova, Lenka
    MATHEMATICAL AND SCIENTIFIC MACHINE LEARNING, VOL 145, 2021, 145 : 426 - 471
  • [33] Detecting Misclassification Errors in Neural Networks with a Gaussian Process Model
    Qiu, Xin
    Miikkulainen, Risto
    THIRTY-SIXTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE / THIRTY-FOURTH CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE / TWELVETH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2022, : 8017 - 8027
  • [34] Scalable Gaussian Process Regression Using Deep Neural Networks
    Huang, Wenbing
    Zhao, Deli
    Sun, Fuchun
    Liu, Huaping
    Chang, Edward
    PROCEEDINGS OF THE TWENTY-FOURTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE (IJCAI), 2015, : 3576 - 3582
  • [35] Accelerated Bayesian Inference for Molecular Simulations using Local Gaussian Process Surrogate Models
    Shanks, Brennon L.
    Sullivan, Harry W.
    Shazed, Abdur R.
    Hoepfner, Michael P.
    JOURNAL OF CHEMICAL THEORY AND COMPUTATION, 2024, 20 (09) : 3798 - 3808
  • [36] Comparison of Ordinal and Metric Gaussian Process Regression as Surrogate Models for CMA Evolution Strategy
    Pitra, Zbynek
    Bajer, Lukas
    Repicky, Jakub
    Holena, Martin
    PROCEEDINGS OF THE 2017 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE COMPANION (GECCO'17 COMPANION), 2017, : 1764 - 1771
  • [37] Scalable Training of Inference Networks for Gaussian-Process Models
    Shi, Jiaxin
    Khan, Mohammad Emtiyaz
    Zhu, Jun
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 97, 2019, 97
  • [38] Local approximate Gaussian process regression for data-driven constitutive models: development and comparison with neural networks
    Fuhg, Jan N.
    Marino, Michele
    Bouklas, Nikolaos
    COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING, 2022, 388
  • [39] Training a Neural-Network-Based Surrogate Model for Aerodynamic Optimisation Using a Gaussian Process
    Ghazi, Yousef
    Alhazmi, Nahla
    Tezaur, Radek
    Farhat, Charbel
    INTERNATIONAL JOURNAL OF COMPUTATIONAL FLUID DYNAMICS, 2022, 36 (07) : 538 - 554
  • [40] Interpretable surrogate models to approximate the predictions of convolutional neural networks in glaucoma diagnosis
    Sigut, Jose
    Fumero, Francisco
    Arnay, Rafael
    Estevez, Jose
    Diaz-Aleman, Tinguaro
    MACHINE LEARNING-SCIENCE AND TECHNOLOGY, 2023, 4 (04):