Bayesian Inference Classification Deep Learning Distributed Computing Federated Learning Inverse Problem Model Identification Optimization Robust Learning Sampling Software Surrogate Modeling Unsupervised Learning 

                   All years
                   20242023202220212020201820172016201520142013
                
                   All types
                   Journal ArticlesConferencesProceedings ArticlesPhD Theses
                
                   All authors
                    Aerts S Ait-Mlouk A Akesson M Alawadi S Braet Q Cheng L Chu J Claeys T Costa M F Couckuyt I Coulier A Deschrijver D Dhaene T Drawert B Ekmefjord M Elamin M M Elsayed K Enberg R Engblom S Eriksson R Ferranti F Fronk C Geiger M Gong X Hast A Hellander Avan der Herten J Hoebeke J Jacob B Javed O Jiang R Jiang R M Joseph W Ju L Karakulev A Koay Y S Martens L Matthew S Moretti S others  Petzold L Petzold L R Petzold. L Pissoort D Plets D Reger G Reger G Rogier H Rossi M Rumsey B Singh P Spjuth O Sturrock M Tanghe E Toor S Trogh J Vandenbosch G A Vats E Vermeeren G Waltari H Wrede F Yi T Yun J Zachariah D
                
Show all
 Karakulev A;  Zachariah D;  Singh P
Adaptive Parameter-Free Robust Learning using Latent Bernoulli Variables Conference 
arXiv preprint, vol. arXiv:2312.00585, 2023.
Abstract | Links | BibTeX | Tags: Bayesian Inference, Deep Learning, Optimization, Robust Learning
@conference{nokey,

title = {Adaptive Parameter-Free Robust Learning using Latent Bernoulli Variables},

author = {Aleksandr Karakulev and Dave Zachariah and Prashant Singh},

url = {https://arxiv.org/abs/2312.00585},

doi = {https://doi.org/10.48550/arXiv.2312.00585},

year  = {2023},

date = {2023-12-01},

urldate = {2023-12-01},

booktitle = {arXiv preprint},

volume = {arXiv:2312.00585},

abstract = {We present an efficient parameter-free approach for statistical learning from corrupted training sets. We identify corrupted and non-corrupted samples using latent Bernoulli variables, and therefore formulate the robust learning problem as maximization of the likelihood where latent variables are marginalized out. The resulting optimization problem is solved via variational inference using an efficient Expectation-Maximization based method. The proposed approach improves over the state-of-the-art by automatically inferring the corruption level and identifying outliers, while adding minimal computational overhead. We demonstrate our robust learning method on a wide variety of machine learning tasks including online learning and deep learning where it exhibits ability to adapt to different levels of noise and attain high prediction accuracy.},

keywords = {Bayesian Inference, Deep Learning, Optimization, Robust Learning},

pubstate = {published},

tppubtype = {conference}

}

Close
We present an efficient parameter-free approach for statistical learning from corrupted training sets. We identify corrupted and non-corrupted samples using latent Bernoulli variables, and therefore formulate the robust learning problem as maximization of the likelihood where latent variables are marginalized out. The resulting optimization problem is solved via variational inference using an efficient Expectation-Maximization based method. The proposed approach improves over the state-of-the-art by automatically inferring the corruption level and identifying outliers, while adding minimal computational overhead. We demonstrate our robust learning method on a wide variety of machine learning tasks including online learning and deep learning where it exhibits ability to adapt to different levels of noise and attain high prediction accuracy.
Close
https://arxiv.org/abs/2312.00585
doi:https://doi.org/10.48550/arXiv.2312.00585
Close