University of Patras
Ανάπτυξη και θεμελίωση μεθόδων για αξιόπιστους νευρωνικούς υπολογισμούς
Abstract
dc:descriptionThis dissertation deals with the development of methods and algorithms offering reliable neural computation. An important part of the research has focused on providing the necessary theoretical basis for these developments. More specifically, research in this thesis focuses on handling uncertainty issues inherent in the data processing, that is, the algorithms used in neural computing. The questions that highlight and qualify the uncertainty in the data processing level are related to matters such as, the initial and boundary values of the algorithms, the robustness of the algorithms against computation or data errors, the multitude of heuristics which substitute mathematical modeling of unknown parameters etc. For modeling and resolving issues such as the above, we, mainly, adopted concepts and methods of Interval Analysis. We also used mathematical approaches such as global optimization and the zero norm of a vector. For specific analysis purposes we used clustering and data mining techniques. In this thesis we study neural networks such as multi-layer perceptrons. Formulation of reliable algorithms implies dealing with uncertainty on matters such as the following: a) initialization of synaptic weights, b) optimization of the network architecture, c) determining the search space where a global minimizer is located for the global optimization procedure when training a network, and d) reliable estimation of a neural network’s domain of validity.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Adam, Stavros
- Αδάμ, Σταύρος
Subjects
dc:subject × 36- Νευρωνικοί υπολογισμοί
- Υπολογιστική νοημοσύνη
- Αλγόριθμοι
- Αξιόπιστοι αλγόριθμοι και υπολογισμοί
- Ανάλυση διαστημάτων
- Νοημοσύνη σμήνους
- Νευρωνικά δίκτυα
- Ομαδοποίηση δεδομένων
- Δίκτυα Kohonen
- Πολυστρωματικά δίκτυα perceptron
- Αρχικοποίηση συναπτικών βαρών
- Διευθέτηση πολυπλοκότητας
- Μηδενική στάθμη
- Καθολική βελτιστοποίηση
- ΕΚΠΑΙΔΕΥΣΗ ΝΕΥΡΩΝΙΚΩΝ ΔΙΚΤΥΩΝ
- Αντιστροφή νευρωνικών δικτύων
- Στατιστική ανάλυση δεδομένων
- Reliable algorithms and computations
- Interval analysis
- Swarm intelligence
- Neural networks
- Clustering
- Kohonen networks
- Multilayer perceptrons
- Complexity regularization
- Zero norm
- Global optimization
- NEURAL NETWORK TRAINING
- Neural network inversion
- Statistical data analysis
- Φυσικές Επιστήμες
- Μαθηματικά
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Mathematics
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/38490