Notes on the Behavior of MC Dropout

Francesco Verdoja; Ville Kyrki

Notes on the Behavior of MC Dropout

Research output: Contribution to conference › Paper › Scientific › peer-review

Abstract

Among the various options to estimate uncertainty in deep neural networks, Monte-Carlo dropout is widely popular for its simplicity and effectiveness. However the quality of the uncertainty estimated through this method varies and choices in architecture design and in training procedures have to be carefully considered and tested to obtain satisfactory results. In this paper we present a study offering a different point of view on the behavior of Monte-Carlo dropout, which enables us to observe a few interesting properties of the technique to keep in mind when considering its use for uncertainty estimation.

Original language	English
Number of pages	6
Publication status	Published - 23 Jul 2021
MoE publication type	Not Eligible
Event	ICML Workshop on Uncertainty & Robustness in Deep Learning - Virtual, Online Duration: 23 Jul 2021 → 23 Jul 2021 Conference number: 2021 https://sites.google.com/view/udlworkshop2021

Workshop

Workshop	ICML Workshop on Uncertainty & Robustness in Deep Learning
Abbreviated title	ICML UDL
City	Virtual, Online
Period	23/07/2021 → 23/07/2021
Internet address	https://sites.google.com/view/udlworkshop2021

Access to Document

ROSE: Robots and the Future of Welfare Services
Kyrki, V. (Principal investigator), Brander, T. (Project Member), Racca, M. (Project Member), Lundell, J. (Project Member) & Verdoja, F. (Project Member)
01/01/2018 → 30/04/2021
Project: Academy of Finland: Strategic research funding

Cite this

@conference{4f2e753556e94addb686c9f787ac6393,

title = "Notes on the Behavior of MC Dropout",

abstract = "Among the various options to estimate uncertainty in deep neural networks, Monte-Carlo dropout is widely popular for its simplicity and effectiveness. However the quality of the uncertainty estimated through this method varies and choices in architecture design and in training procedures have to be carefully considered and tested to obtain satisfactory results. In this paper we present a study offering a different point of view on the behavior of Monte-Carlo dropout, which enables us to observe a few interesting properties of the technique to keep in mind when considering its use for uncertainty estimation. ",

author = "Francesco Verdoja and Ville Kyrki",

year = "2021",

month = jul,

day = "23",

language = "English",

note = "ICML Workshop on Uncertainty & Robustness in Deep Learning, ICML UDL ; Conference date: 23-07-2021 Through 23-07-2021",

url = "https://sites.google.com/view/udlworkshop2021",

}

TY - CONF

T1 - Notes on the Behavior of MC Dropout

AU - Verdoja, Francesco

AU - Kyrki, Ville

N1 - Conference code: 2021

PY - 2021/7/23

Y1 - 2021/7/23

N2 - Among the various options to estimate uncertainty in deep neural networks, Monte-Carlo dropout is widely popular for its simplicity and effectiveness. However the quality of the uncertainty estimated through this method varies and choices in architecture design and in training procedures have to be carefully considered and tested to obtain satisfactory results. In this paper we present a study offering a different point of view on the behavior of Monte-Carlo dropout, which enables us to observe a few interesting properties of the technique to keep in mind when considering its use for uncertainty estimation.

AB - Among the various options to estimate uncertainty in deep neural networks, Monte-Carlo dropout is widely popular for its simplicity and effectiveness. However the quality of the uncertainty estimated through this method varies and choices in architecture design and in training procedures have to be carefully considered and tested to obtain satisfactory results. In this paper we present a study offering a different point of view on the behavior of Monte-Carlo dropout, which enables us to observe a few interesting properties of the technique to keep in mind when considering its use for uncertainty estimation.

M3 - Paper

T2 - ICML Workshop on Uncertainty & Robustness in Deep Learning

Y2 - 23 July 2021 through 23 July 2021

ER -

Notes on the Behavior of MC Dropout

Abstract

Workshop

Access to Document

Fingerprint

Projects

ROSE: Robots and the Future of Welfare Services

Cite this