A prediction rigidity formalism for low-cost uncertainties in trained neural networks

被引:0
|
作者
Bigi, Filippo [1 ]
Chong, Sanggyu [1 ]
Ceriotti, Michele [1 ]
Grasselli, Federico [1 ,2 ]
机构
[1] Ecole Polytech Fed Lausanne, Inst Materiaux, Lab Computat Sci & Modeling, CH-1015 Lausanne, Switzerland
[2] Univ Modena & Reggio Emilia, Dept Phys Informat & Math, via Giuseppe Campi 213-a, I-41125 Modena, Italy
来源
MACHINE LEARNING-SCIENCE AND TECHNOLOGY | 2024年 / 5卷 / 04期
基金
瑞士国家科学基金会; 欧洲研究理事会; 欧盟地平线“2020”;
关键词
predictions; rigidity; low-cost uncertainties; neural network; regression; pre-trained; uncertainty quantification;
D O I
10.1088/2632-2153/ad805f
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Quantifying the uncertainty of regression models is essential to ensure their reliability, particularly since their application often extends beyond their training domain. Based on the solution of a constrained optimization problem, this work proposes 'prediction rigidities' as a formalism to obtain uncertainties of arbitrary pre-trained regressors. A clear connection between the suggested framework and Bayesian inference is established, and a last-layer approximation is developed and rigorously justified to enable the application of the method to neural networks. This extension affords cheap uncertainties without any modification to the neural network itself or its training procedure. The effectiveness of this approach is shown for a wide range of regression tasks, ranging from simple toy models to applications in chemistry and meteorology.
引用
收藏
页数:20
相关论文
共 50 条
  • [21] A Low-Cost Fault-Tolerant Approach for Hardware Implementation of Artificial Neural Networks
    Ahmadi, A.
    Sargolzaie, M. H.
    Fakhraie, S. M.
    Lucas, C.
    Vakili, Sh.
    2009 INTERNATIONAL CONFERENCE ON COMPUTER ENGINEERING AND TECHNOLOGY, VOL II, PROCEEDINGS, 2009, : 93 - 97
  • [22] Forecasting low-cost housing demand in Pahang, Malaysia using Artificial Neural Networks
    Zainun, Noor Yasmin
    Rahman, Ismail Abdul
    Eftekhari, Mahroo
    INTERNATIONAL JOURNAL OF SUSTAINABLE CONSTRUCTION ENGINEERING AND TECHNOLOGY, 2011, 2 (01): : 83 - 88
  • [23] Pavement distress instance segmentation using deep neural networks and low-cost sensors
    Mahdy, Kamel
    Zekry, Ahmed
    Moussa, Mohamed
    Mohamed, Ahmed
    Mahdy, Hassan
    Elhabiby, Mohamed
    INNOVATIVE INFRASTRUCTURE SOLUTIONS, 2024, 9 (01)
  • [24] A low-cost and portable spectrometer for measuring light spectrum using artificial neural networks
    Amirazar, Armin
    Azarbayjani, Mona
    Molavi, Maziyar
    Karami, Morteza
    ENERGY AND BUILDINGS, 2021, 252
  • [25] A Dynamic Threshold Quantization Method for Ternary Neural Networks for Low-cost Mobile Robots
    Son Ngoc Truong
    INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND NETWORK SECURITY, 2020, 20 (02): : 16 - 20
  • [26] A LOW-COST, 3 CHANNEL NEURAL STIMULATOR
    YOUNG, WG
    BRAIN RESEARCH BULLETIN, 1980, 5 (04) : 483 - 484
  • [27] DESIGN OF LOW-COST, REAL-TIME SIMULATION SYSTEMS FOR LARGE NEURAL NETWORKS
    JAMES, M
    HOANG, D
    JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING, 1992, 14 (03) : 221 - 235
  • [28] Pavement distress instance segmentation using deep neural networks and low-cost sensors
    Kamel Mahdy
    Ahmed Zekry
    Mohamed Moussa
    Ahmed Mohamed
    Hassan Mahdy
    Mohamed Elhabiby
    Innovative Infrastructure Solutions, 2024, 9
  • [29] Robust Low-Cost Drone Detection and Classification Using Convolutional Neural Networks in Low SNR Environments
    Gluge, Stefan
    Nyfeler, Matthias
    Aghaebrahimian, Ahmad
    Ramagnano, Nicola
    Schupbach, Christof
    IEEE JOURNAL OF RADIO FREQUENCY IDENTIFICATION, 2024, 8 : 821 - 830
  • [30] Better Web searches and prediction with instantaneously trained neural networks
    Kak, S
    IEEE INTELLIGENT SYSTEMS & THEIR APPLICATIONS, 1999, 14 (06): : 78 - 81