Bayesian Computation Statistics
Morteza Amini; Moein Monemi; Mahmoud Taheri; Mohammad Arashi
Abstract
We investigate the problem of weight uncertainty originally proposed by [Blundell et al. (2015). Weight uncertainty in neural networks. In International conference on machine learning, 1613-1622, PMLR.] in the context of neural networks designed for regression tasks, and we extend their framework by ...
Read More
We investigate the problem of weight uncertainty originally proposed by [Blundell et al. (2015). Weight uncertainty in neural networks. In International conference on machine learning, 1613-1622, PMLR.] in the context of neural networks designed for regression tasks, and we extend their framework by incorporating variance uncertainty into the model. Our analysis demonstrates that explicitly modeling uncertainty in the variance parameter can significantly enhance the predictive performance of Bayesian neural networks. By considering a full posterior distribution over the variance, the model achieves improved generalization compared to approaches that treat variance as fixed or deterministic. We evaluate the generalization capability of our proposed approach through a function approximation example and further validate it on the riboflavin genetic dataset. Our exploration encompasses both fully connected dense networks and dropout neural networks, employing Gaussian and spike-and-slab priors respectively for the network weights, providing a comprehensive assessment of how variance uncertainty affects model performance across different architectural choices.
Machine Learning
Morteza Amini; Kiana Ghasemifard
Abstract
The diabetes data set gathered by Michael Kahn, at Washington University, St. Louis, MO, which is available online at UCI machine learning repository is one of the rarely used data sets, specially for glucose prediction purposes in diabetic patients. In this paper, we study the problem of blood glucose ...
Read More
The diabetes data set gathered by Michael Kahn, at Washington University, St. Louis, MO, which is available online at UCI machine learning repository is one of the rarely used data sets, specially for glucose prediction purposes in diabetic patients. In this paper, we study the problem of blood glucose range prediction, rather than raw glucose prediction, along with two other important tasks, which are the detection of increment or decrement of glucose as well as abnormal value prediction, based on regular and NPH insulin doses, based on this data set. Two commonly used machine learning approaches for time series data, namely LSTM and CNN are used along with a promising statistical regression approach, that is non-parametric multivariate Gaussian additive mixed model, for the prediction task. It is observed that, although LSTM and CNN models are preferable concerning the prediction error, the statistical method performs significantly better in the sense of abnormal value detection, which is a critical task for diabetic patients.
