Browsing by Subject "Neural networks (Computer science)."

Now showing 1 - 7 of 7

Open Access
Design and stability of Hopfield associative memory
(1991) Savran, M. Erkan
This thesis is concerned with the selection of connection weights of Hopfield neural network model so that the network functions as a content addressable memory (CAM). We deal with both the discrete and the continuous-time versions of the model using hard-limiter and sigmoid type nonlinearities in the neuron outputs. The analysis can be employed if any other invertible nonlinearity is used. The general characterization of connection weights for fixed-point programming and a condition for asymptotic stability of these fixed points are presented. The general form of connection weights is then inserted in the condition to obtain a design rule. The characterization procedure is also employed for discrete-time cellular neural networks.
Open Access
Distributed scheduling
(1999) Toptal, Ayşegül
Distributed Scheduling (DS) is a new paradigm that enables the local decisionmakers make their own schedules by considering local objectives and constraints within the boundaries and the overall objective of the whole system. Local schedules from different parts of the system are then combined together to form a final schedule. Since each local decision-maker acts independently from each other, the communication system in a distributed architecture should be carefully designed to achieve better overall system performance. These systems are preferred over the traditional systems due to the ability to update the schedule, flexibility, reactivity and shorter lead times. In this thesis, we review the existing work on DS and propose a new classification framework. We also develop a number of bidding based DS algorithms. These algorithms are tested under various manufacturing environments.
Open Access
A general purpose rotation, scaling, and translation invariant pattern classification system
(1992) Yüceer, Cem
Artificial neural networks have recently been used for pattern classification purposes. In this work, a general purpose pattern classification system which is rotation, scaling, and, translation invariant is introduced. The system has three main blocks; a Karhunen-Loeve transformation based preprocessor, an artificial neural network based classifier, and an interpreter. Through experimentation on the English alphabet, the Japanese Katakana alphabet, and some geometric symbols the power of the system in maintaining invariancies and performing pattern classification has been shown.
Open Access
Implementation of the backpropagation algorithm on iPSC/2 hypercube multicomputer system
(1990) Ercoşkun, Deniz
Backpropagation is a supervised learning procedure for a class of artificial neural networks. It has recently been widely used in training such neural networks to perform relatively nontrivial tasks like text-to-speech conversion or autonomous land vehicle control. However, the slow rate of convergence of the basic backpropagation algorithm has limited its application to rather small networks since the computational requirements grow significantly as the network size grows. This thesis work presents a parallel implementation of the backpropagation learning algorithm on a hypercube multicomputer system. The main motivation for this implementation is the construction of a parallel training and simulation utility for such networks, so that larger neural network applications can be experimented with.
Open Access
Motion planning of a mechanical snake using neural networks
(1998) Fidan, Barış
In this thesis, an optimal strategy is developed to get a mechanical snake (a robot composed of a sequence of articulated links), which is located arbitrarily in an enclosed region, out of the region through a specified exit without violating certain constraints. This task is done in two stages: Finding an optimal path that can be tracked, and tracking the optimal path found. Each stage is implemented by a neural network. Neural network of the second stage is constructed by direct evaluation of the weights after designing an efficient structure. Two independent neural networks are designed to implement the first stage, one trained to implement an algorithm we have derived to generate minimal paths and the other trained using multi-stage neural network approach. For the second design, the intuitive multi-stage neural network back propagation approach in the literature is formalized.
Open Access
Simulation metamodeling with neural networks
(1997) Touhami, Souheyl
Modern manufacturing environments increasingly call for more sophisticated cind fast decision aiding systems for their management. Artificial neural networks have been proposed as an alternative cipproach for formalizing various quantitative and qualitative aspects of manufacturing systems. This research attempts to lay down the motivation behind using neural networks as a simulation metamodeling approach. This research can be classified under the major headings of simulation metamodeling for the purpose of estimating system performance. Steiidy state perfornuince of non-terminating type systems and transient state performance of terminating tyj^e systems are examined under job shop environments by applying Back Propagation neural networks. We attempt to study the peribrrnance of neural metamodels with respect to estimating two performance measures (mean machine utilization and mean job tardiness), with respect to system complexity, with different types of system configurations (deterministic cuid stochastic), with respect to multiple metamodel accuracy assessment criteria and various metamodel design settings. The objective of this analysis is to investigate the potential application of neural metamodeling.
Open Access
Variations in associative memory design
(1996) Akar, Mehmet
This thesis is concerned with the anaiysis and synthesis of neurai networks to be used as associative memories. First considering a discrete-time neurai network modei which uses a quantizer-type muitiievei activation function, a way of seiecting the connection weights is proposed. In addition to this, the idea of overiapping decompositions, which is extensiveiy used in the soiution of iarge-scaie probiems, is appiied to discrete-time neurai networks with binary neurons. 'I’lie necesscuy toois for expansions and contractions are derived, and algorithms for decomposition of a set equiiibria into smaiier dimensionai equiiibria sets and for designing neurai networks for these smaiier ciimensionai equiiibria sets are given. The concept is iiiustrated with various exarnpies.