Research into novel approaches like Machine learning (ML) promotes a new set of opportunities for sustainable development of applications through automation. However, there are certain ML tasks which are prone to spur...
详细信息
Research into novel approaches like Machine learning (ML) promotes a new set of opportunities for sustainable development of applications through automation. However, there are certain ML tasks which are prone to spurious classification, mainly due to the bias in legacy data. One well-known and highly actual misclassification case concerns gender. As the vast dataset for engineering rules, standards and experiments are based on men, a bias towards women is the subject of research. Accordingly, any bias should be contained before the algorithms are deployed to the service of the sustainable society. There is a substantial amount of data on ML gender-bias in the literature. In these, the majority of the investigated cases are for ML branches like image or sound processing and text recognition. However, utilizing ML for driving style investigations is not an extensively researched area. In this work, a novel application for gender-based classification with bias-attenuation using anonymized driving data will be presented. Using data devoid of biometric and geographic information, the proposed pipeline distinguishes manifested binary genders with 80% accuracy for the drivers in the holdout data set. In addition, a method for sustainable algorithm selection and its extension to embedded applications, is proposed. An investigation into the environmental burden of seven different types of ML algorithms was conducted and the popular neural network algorithm had the highest environmental burden.
Deep unfolding provides an approach to integrate the probabilistic generative models and the deterministic neural networks. Such an approach is benefited by deep representation, easy interpretation, flexible learning ...
详细信息
Deep unfolding provides an approach to integrate the probabilistic generative models and the deterministic neural networks. Such an approach is benefited by deep representation, easy interpretation, flexible learning and stochastic modeling. This study develops the unsupervised and supervisedlearning of deep unfolded topic models for document representation and classification. Conventionally, the unsupervised and supervised topic models are inferred via the variational inference algorithm where the model parameters are estimated by maximizing the lower bound of logarithm of marginal likelihood using input documents without and with class labels, respectively. The representation capability or classification accuracy is constrained by the variational lower bound and the tied model parameters across inference procedure. This paper aims to relax these constraints by directly maximizing the end performance criterion and continuously untying the parameters in learning process via deep unfolding inference (DUI). The inference procedure is treated as the layer-wise learning in a deep neural network. The end performance is iteratively improved by using the estimated topic parameters according to the exponentiated updates. Deep learning of topic models is therefore implemented through a back-propagation procedure. Experimental results show the merits of DUI with increasing number of layers compared with variational inference in unsupervised as well as supervised topic models.
暂无评论