Pusat Pengajian Sains Matematik - Tesis

Browse

Now showing 1 - 5 of 436

A Resolution Based Automated Theorem Proving System Using Concurrent Processing Approach
(1994-05)
Natarajan, Surash
The research reported in this thesis is devoted to the use of concurrent processing for developing a resolution based automated theorem proving system, an application in the area of artificial intelligence. Our purpose in doing this is to study the usefulness of concurrent processing in enhancing the problem solving process in a resolution based automated theorem proving system. During our research here we investigated which component of the theorem prover can be decomposed into introducing concurrent processing and how this should be done. Our main aim in building this theorem prover was not mainly in producing a high performance theorem prover but to build a system that can be considered to be a prototype that would illustrate the idea of introducing concurrent processing in resolution based theorem provers. We believe that concurrent processing is the intermediate step in moving from sequential processing towards parallel processing. Concurrent processing provides the simplicity of sequential system design with efficient processing capabilities of parallel system. In our discussion here we present a novel design of the system and how we propose to implement it.
Manpower Planning Model For Pilot Accessions In The Royal Malaysian Air Force
(1994-04)
M. Jamil, Khairron Anuar
This research aims to evaluate recruitment policies via mathematical model to project the strength of RMAF pilots for more than ten years. Previous researches have adopted the weighted Linear Goal Programming (LGP) technique to evaluate recruitment policies, however a preemptive LGP technique is used in this research. This technique has some advantages over the weighted LGP technique.
Y-type Random 2-satisfiability In Discrete Hopfield Neural Network
(2024-09)
Guo, Yueling
In the current development of Artificial Intelligence, Satisfiability plays a crucial role as a symbolic language of Artificial Intelligence for the transparency of black box models. However, the main problem of existing Satisfiability is the lack of combined logical rule, so the benefits of combined logical rule have not yet been investigated. The rule namely Y-Type Random 2-Satisfiability is proposed by combining the systematic and non-systematic logical rule. Next, the newly proposed logical rule as the symbolic instruction was implemented into the Discrete Hopfield Neural Network to govern the neurons of the network. Experimental results demonstrated the compatibility of the proposed logical rule and the Discrete Hopfield Neural Network. Additionally, the proposed Hybrid Differential Evolution Algorithm was implemented into the training phase to ensure that the cost function of the Discrete Hopfield Neural Network is minimized. During the retrieval phase, a new activation function and Swarm Mutation were proposed to ensure the diversity of the neuron states. The proposed algorithm and mutation mechanism showed optimal performances as compared to the existing algorithms. Finally, a new logic mining model namely Y-Type Random 2-Satisfiability Reverse Analysis was proposed, which showed optimal performances in terms of several metrics as compared to the existing classification models. The developed logic mining will be used to analyze the Alzheimer's Disease Neuroimaging Initiative dataset.
Visualization Of Curve And Surface Data Using Rational Cubic Ball Functions
(2018-02)
Wan Jaafar, Wan Nurhadani
This study considered the problem of shape preserving interpolation through regular data using rational cubic Ball which is an alternative scheme for rational Bezier functions. A rational Ball function with shape parameters is easy to implement because of its less degree terms at the end polynomial compared to rational Bezier functions. In order to understand the behavior of shape parameters (weights), we need to discuss shape control analysis which can be used to modify the shape of a curve, locally and globally. This issue has been discovered and brought to the study of conversion between Ball and Bezier curve. A conversion formula was obtained after a Bezier curve converted to the generalized form of Ball curve. It proved that this formulae not only valuable for geometric properties studies but also improves on the computational speed of the Ball curves.
Optimization Methods In Training Neural Networks
(2003-07)
Sathasivam, Saratha
In steepest descent methods, we construct a functional which when extremized will deliver a solution. The function will have convexity properties, so that the vector that extremizes the function is the solution of the algebraic problem in question. This means the search for the vector for which the gradient of the function is zero can be done in an iterative fashion. A special steepest descent method which is appropriate for the solution of the linear algebraic problem is the 'Conjugate Gradient Method'.

Browse

Recent Submissions