1. |
- Abiri, Najmeh, et al.
(author)
-
Establishing strong imputation performance of a denoising autoencoder in a wide range of missing data problems
- 2019
-
In: Neurocomputing. - Amsterdam : Elsevier BV. - 0925-2312 .- 1872-8286. ; 365, s. 137-146
-
Journal article (peer-reviewed)abstract
- Dealing with missing data in data analysis is inevitable. Although powerful imputation methods that address this problem exist, there is still much room for improvement. In this study, we examined single imputation based on deep autoencoders, motivated by the apparent success of deep learning to efficiently extract useful dataset features. We have developed a consistent framework for both training and imputation. Moreover, we benchmarked the results against state-of-the-art imputation methods on different data sizes and characteristics. The work was not limited to the one-type variable dataset; we also imputed missing data with multi-type variables, e.g., a combination of binary, categorical, and continuous attributes. To evaluate the imputation methods, we randomly corrupted the complete data, with varying degrees of corruption, and then compared the imputed and original values. In all experiments, the developed autoencoder obtained the smallest error for all ranges of initial data corruption.
|
|
2. |
- Abiri, Najmeh
(author)
-
Statistical inference with deep latent variable models
- 2019
-
Doctoral thesis (other academic/artistic)abstract
- Finding a suitable way to represent information in a dataset is one of the fundamental problems in Artificial Intelligence. With limited labeled information, unsupervised learning algorithms help to discover useful representations. One of the applications of such models is imputation, where missing values are estimated by learning the underlying correlations in a dataset. This thesis explores two of unsupervised techniques: stacked denoising autoencoders and variational autoencoders (VAEs). Using stacked denoising autoencoders, we developed a consistent framework to handle incomplete data with multi-type variables. This deterministic method improved missing data estimation compared to several state-of-the-art imputation methods. Further, we explored variational autoencoders, a probabilistic form of autoencoders that jointly optimize the neural network-based inference and generative models. Despite the promise of these techniques, the main difficulty is an uninformative latent space. We propose a flexible family, Student's t-distributions, as priors for VAEs to learn a more informative latent representation. By comparing different forms of the covariance matrix for both Gaussian and Student's t-distributions, we conclude that using a weakly informative prior such as the Student's t with a low number of parameters improves the ability of VAEs to approximate the true posterior.Finally, we used VAEs both with the Gaussian and Student's t-priors as multiple imputation methods on two datasets with missing values. Moreover, with the provided labels on these datasets, we used a supervised network and evaluated the estimation of missing variables. In both cases, VAEs show improvements compared to other methods.
|
|
3. |
- Abiri, Najmeh, et al.
(author)
-
Variational auto-encoders with Student’s t-prior
- 2019
-
In: ESANN 2019 - Proceedings : The 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning - The 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. - Bruges : ESANN. - 9782875870650
-
Conference paper (peer-reviewed)abstract
- We propose a new structure for the variational auto-encoders (VAEs) prior, with the weakly informative multivariate Student’s t-distribution. In the proposed model all distribution parameters are trained, thereby allowing for a more robust approximation of the underlying data distribution. We used Fashion-MNIST data in two experiments to compare the proposed VAEs with the standard Gaussian priors. Both experiments showed a better reconstruction of the images with VAEs using Student’s t-prior distribution.
|
|
4. |
- Farooq, Zia, et al.
(author)
-
Artificial intelligence to predict West Nile virus outbreaks with eco-climatic drivers
- 2022
-
In: The Lancet Regional Health - Europe. - : Elsevier BV. - 2666-7762. ; 17
-
Journal article (peer-reviewed)abstract
- Background: In Europe, the frequency, intensity, and geographic range of West Nile virus (WNV)-outbreaks have increased over the past decade, with a 7.2-fold increase in 2018 compared to 2017, and a markedly expanded geographic area compared to 2010. The reasons for this increase and range expansion remain largely unknown due to the complexity of the transmission pathways and underlying disease drivers. In a first, we use advanced artificial intelligence to disentangle the contribution of eco-climatic drivers to WNV-outbreaks across Europe using decade-long (2010-2019) data at high spatial resolution. Methods: We use a high-performance machine learning classifier, XGBoost (eXtreme gradient boosting) combined with state-of-the-art XAI (eXplainable artificial intelligence) methodology to describe the predictive ability and contribution of different drivers of the emergence and transmission of WNV-outbreaks in Europe, respectively. Findings: Our model, trained on 2010-2017 data achieved an AUC (area under the receiver operating characteristic curve) score of 0.97 and 0.93 when tested with 2018 and 2019 data, respectively, showing a high discriminatory power to classify a WNV-endemic area. Overall, positive summer/spring temperatures anomalies, lower water availability index (NDWI), and drier winter conditions were found to be the main determinants of WNV-outbreaks across Europe. The climate trends of the preceding year in combination with eco-climatic predictors of the first half of the year provided a robust predictive ability of the entire transmission season ahead of time. For the extraordinary 2018 outbreak year, relatively higher spring temperatures and the abundance of Culex mosquitoes were the strongest predictors, in addition to past climatic trends. Interpretation: Our AI-based framework can be deployed to trigger rapid and timely alerts for active surveillance and vector control measures in order to intercept an imminent WNV-outbreak in Europe. Funding: The work was partially funded by the Swedish Research Council FORMAS for the project ARBOPREVENT (grant agreement 2018-05973).
|
|
5. |
|
|
6. |
- Lantz, Victor, et al.
(author)
-
Deep learning for inverse problems in quantum mechanics
- 2021
-
In: International Journal of Quantum Chemistry. - : Wiley. - 0020-7608 .- 1097-461X. ; 121:9
-
Journal article (peer-reviewed)abstract
- Inverse problems are important in quantum mechanics and involve such questions as finding which potential give a certain spectrum or which arrangement of atoms give certain properties to a molecule or solid. Inverse problems are typically very hard to solve and tend to be very compute intense. We here show that neural networks can easily solve inverse problems in quantum mechanics. It is known that a neural network can compute the spectrum of a given potential, a result which we reproduce. We find that the (much harder) inverse problem of computing the correct potential that gives a prescribed spectrum is equally easy for a neural network. We extend previous work where neural networks were used to find the electronic many-particle density given a potential by considering the inverse problem. That is, we show that neural networks can compute the potential that gives a prescribed many-electron density.
|
|