CVSL Logo
FrancaisHome
AboutPeopleResearchPublicationsEventsProfile
About
Publications

 

 

 

CERVIM

REPARTI

MIVIM

A New Loss Function for Temperature Scaling to have Better Calibrated Deep Networks


Azadeh Mozafari, Hugo Siqueira Gomes, Steeven Janny and Christian Gagné

More on this project...

Abstract - However Deep neural networks recently have achieved impressive results for different tasks, they suffer from poor uncertainty prediction. Temperature Scaling(TS) is an efficient post-processing method for calibrating DNNs toward to have more accurate uncertainty prediction. TS relies on a single parameter T which softens the logit layer of a DNN and the optimal value of it is found by minimizing on Negative Log Likelihood (NLL) loss function. In this paper, we discuss about weakness of NLL loss function, especially for DNNs with high accuracy and propose a new loss function called Attended-NLL which can improve TS calibration ability significantly.

download documentdownload document

Bibtex:

@article{Mozafari1206,
    author    = { Azadeh Mozafari and Hugo Siqueira Gomes and Steeven Janny and Christian Gagné },
    title     = { A New Loss Function for Temperature Scaling to have Better Calibrated Deep Networks },
    volume    = { 1810.11586 },
    year      = { 2018 },
    month     = { 10 },
    journal   = { ArXiv e-prints },
    web       = { https://arxiv.org/abs/1810.11586 }
}

Last modification: 2018/12/03 by cgagne

     
   
   

©2002-. Computer Vision and Systems Laboratory. All rights reserved