Variational closed-Form deep neural net inference

Name: Variational closed-Form deep neural net inference
Author: Kandemir, Melih

İsim	Variational closed-Form deep neural net inference
Yazar	Kandemir, Melih
Basım Tarihi:	2018-09
Basım Yeri	- Elsevier
Konu	Bayesian Neural Networks, Variational Bayes, Online learning, Active learning
Tür	Süreli Yayın
Dil	İngilizce
Dijital	Evet
Yazma	Hayır
Kütüphane:	Özyeğin Üniversitesi
Demirbaş Numarası	0167-8655
Kayıt Numarası	bafad6f4-fd58-46ac-8b4c-98cf9e4ee511
Lokasyon	Computer Science
Tarih	2018-09
Örnek Metin	We introduce a Bayesian construction for deep neural networks that is amenable to mean field variational inference that operates solely by closed-form update rules. Hence, it does not require any learning rate to be manually tuned. We show that by this virtue it becomes possible with our model to perform effective deep learning on three setups where conventional neural nets are known to perform suboptimally: i) online learning, ii) learning from small data, and iii) active learning. We compare our approach to earlier Bayesian neural network inference techniques spanning from expectation propagation to gradient-based variational Bayes, as well as deterministic neural nets with various activations functions. We observe our approach to improve on all these alternatives in two mainstream vision benchmarks and two medical data sets: diabetic retinopathy screening and exudate detection from eye fundus images.
DOI	10.1016/j.patrec.2018.07.001
Cilt	112

Kaynağa git Özyeğin Üniversitesi

Aramaya Dön

Özyeğin Üniversitesi

Kaynağa git

Variational closed-Form deep neural net inference

Yazar Kandemir, Melih

Basım Tarihi 2018-09

Basım Yeri - Elsevier

Konu Bayesian Neural Networks, Variational Bayes, Online learning, Active learning

Tür Süreli Yayın

Dil İngilizce

Dijital Evet

Yazma Hayır

Kütüphane Özyeğin Üniversitesi

Demirbaş Numarası 0167-8655

Kayıt Numarası bafad6f4-fd58-46ac-8b4c-98cf9e4ee511

Lokasyon Computer Science

Tarih 2018-09

Örnek Metin We introduce a Bayesian construction for deep neural networks that is amenable to mean field variational inference that operates solely by closed-form update rules. Hence, it does not require any learning rate to be manually tuned. We show that by this virtue it becomes possible with our model to perform effective deep learning on three setups where conventional neural nets are known to perform suboptimally: i) online learning, ii) learning from small data, and iii) active learning. We compare our approach to earlier Bayesian neural network inference techniques spanning from expectation propagation to gradient-based variational Bayes, as well as deterministic neural nets with various activations functions. We observe our approach to improve on all these alternatives in two mainstream vision benchmarks and two medical data sets: diabetic retinopathy screening and exudate detection from eye fundus images.

DOI 10.1016/j.patrec.2018.07.001

Cilt 112