Deep learning methods for prediction and classification

Időpont: 
2019. 04. 18. 14:15
Hely: 
BME H. épület 406 terem
Előadó: 
Marijana Zekić-Sušac Josip Juraj Strossmayer University, Faculty of Economics, Osijek, Croatia

                     

                                MEGHÍVÓ

      Szeretettel meghívjuk Marijana Zekić-Sušac előadására

            az Optimalizálási Szeminárium keretében

              2019. április 18., csütörtök 14:15 - 15:45

                Helyszín: BME H. épület 406 terem

 

 

 

Előadó: Marijana Zekić-Sušac Josip Juraj Strossmayer University, Faculty of Economics, Osijek, Croatia

 

Deep learning methods for prediction and classification

Abstract:

Artificial neural networks (ANNs) are machine learning methods aimed to find relationships among input and output variables on historical data, i.e. approximate functions that produce minimum error on the observed dataset. ANNs have shown success in solving prediction, classification and association problems.The most commonly used multi-layer perceptron (MLP) network consists of an input layer which loads values from an input vector in each iteration, a hidden layer which uses the summed weighted inputs from the input layer and applies an activation function, and an output layer where the output is produced and the error computed. The most commonly used algorithm for optimizing the ANN error is the gradient descent, while the second-order methods can be also applied. In case of multiple hidden layers, we talk about deep learning neural networks (DNNs) which are able to process huge datasets. They gained its popularity with Big Data platforms and new generations of fast computer processors such as GPU and TPU. By DNNs, different activation functions can be used in each hidden layer, and various methods of feature extraction can be added. In this lecture, the procedure of creating DNN models for prediction and classification in the area of energy management will be presented using R software tool.

 

Curriculum vitae:

Marijana Zekić-Sušac is a full professor with tenure at Josip Juraj Strossmayer University, Faculty of Economics in Osijek, Croatia. She earned her doctoral degree at the University of Zagreb, Faculty of Organization and Informatics, Varaždin, Croatia. Her research interests include artificial intelligence, machine learning, and data mining in business, education, and medicine. She currently teaches seven ICT courses at undergraduate, graduate, and doctoral levels. She has published 3 books, more than 60 scientific papers in journals and conference proceedings. She was the president of the Croatian Operational Research Society from 2010. to 2016, and the editor-in-chief of the Croatian Operational Research Review journal. Currently, she serves as the associate editor of the Central European Journal of Operational Research and a member of the editorial board of the journal EconViews - Review of Contemporary Business, Entrepreneurship and Economic Issues and of several conferences. She is also a leader of the research project “Methodological framework of intelligent data analytics in energy management – MERIDA” funded by the Croatian Science Foundation.