Αξιοποίηση νευρωνικών δικτύων γράφων για την πρόβλεψη του αλγορίθμου συσταδοποίησης με την καλύτερη επίδοση
View/
Keywords
Μηχανική Μάθηση ; Βαθιά Μάθηση ; Γράφοι ; GNN ; AutoML ; Κατηγοριοποίηση ; ΣυσταδοποίησηAbstract
In today's rapidly evolving world, artificial intelligence is becoming more and more integrated into our lives every day, revolutionizing the way we interact with technology. However, developing predictive models requires expertise and specialized knowledge, which makes it inaccessible to the majority of people. Automated machine learning (AutoML) attempts to solve this problem by reducing significantly human intervention, resulting in prediction models becoming more accessible to a larger portion of the world. This thesis deals with the implementation of an automated machine learning model, which takes as input a data set and tries to predict the best performing clustering algorithm utilizing graph neural network architectures. To implement the above model, the Dataset2Graph library was initially developed in the Python programming language, which provides the ability to convert a dataset into a graph and includes a variety of algorithms that process and simplify the generated graphs. At our disposal we had 50 datasets, which were converted into a graph. Different simplification techniques were used on the 50 graphs and over 150 graph datasets were generated. Finally, all graph neural network architectures developed for the prediction stage are evaluated on the different graph sets and the optimal model is compared with the main competitor MARCO-GE.
Postgraduate Studies Programme
Πληροφοριακά Συστήματα και ΥπηρεσίεςDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών ΣυστημάτωνNumber of pages
49Language
GreekCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα