پديد آورنده :
آدميت، سپيده
عنوان :
يادگيري عميق توالي با استفاده از چارچوب تغييراتي
مقطع تحصيلي :
كارشناسي ارشد
گرايش تحصيلي :
هوش مصنوعي
محل تحصيل :
اصفهان : دانشگاه صنعتي اصفهان
صفحه شمار :
سيزده، 75 ص. : مصور، جدول، نمودار
استاد راهنما :
مهران صفاياني
استاد مشاور :
عبدالرضا ميرزايي
توصيفگر ها :
خودكدگذارمتغير , داده‌هاي داراي توالي , شبكه‌هاي عصبي , يادگيري عميق , روش‌هاي متغير , يادگيري بدون نظارت , سيگنال مغزي , توانبخشي
استاد داور :
محمدرضا احمدزاده، سمانه حسيني
تاريخ ورود اطلاعات :
1399/08/06
رشته تحصيلي :
مهندسي كامپيوتر
دانشكده :
مهندسي برق و كامپيوتر
تاريخ ويرايش اطلاعات :
1399/08/07
چكيده انگليسي :
Deep sequence learning with variational framework Sepideh Adamiat s adamiat@ec iut ac ir June 2020 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisor Dr Mehran Safayani safayani@cc iut ac ir Advisor Dr Abdolreza Mirzaei mirzaei@cc iut ac ir Abstract One of the newest and most challenging areas of artificial intelligence is generating new data by machines with learningthe patterns in existing data Among the types of data sequential ones are more complex due to the dependencies that existbetween their subsequence and thus we require different methods for processing them The most well known methodsused to study and analyze these data are recurrent neural networks But simple recurrent neural networks are not generativemodels and for using them in task of generation they need to be implemented in the form of a generative model In this study we introduce and implemente recurrent neural networks in the form of a developed model of variational au toencoder This model is able to generate meaningful and new intervals of time steps in addition this is a multifunctionalmodel which can be used simultaneously for three important tasks in the field of machine learning 1 conditionally gen erating data This means that we can generate new sequential data by considering different classes of data This importantfeature does not exist in the simple variational autoencoder which is one of the most used generating models 2 Clusteringfor sequential data Today data clustering is one of the most challenging areas of machine learning due to the lower num ber of labeled data than unlabeled data The introduced model achieved an accuracy of up to 30 percentage better than itscompeting models in this task 3 Reconstructing missing parts for sequential data The learned model is able to predictmissing parts for unseen data In this task we have achieved a better result than the competing models by computing meansquare error of predicted parts and the original ones Key Words Variational Autoencoder Sequential Data Neural Networks Deep Learning VariationalMethods Unsupervised Learning EEG Signal Rehabilitation
استاد راهنما :
مهران صفاياني
استاد مشاور :
عبدالرضا ميرزايي
استاد داور :
محمدرضا احمدزاده، سمانه حسيني
