No description available.

Researchers have always found difficulty in attaining funding from the National Research Foundation (NRF) for new research interests. The field of Artificial Intelligence (AI) holds the promise of improving the matching of research proposals to funding sources in the area of Intelligent Information Retrieval (IIR). IIR is a fairly new AI technique that has evolved from the traditional IR systems to solve real-world problems. Typically, an IIR system contains three main components, namely, a knowledge base, an inference engine and a user-interface. Due to its inferential capabilities. IIR has been found to be applicable to domains for which traditional techniques, such as the use of databases, have not been well suited. This applicability has led it to become a viable AI technique from both, a research and an application perspective. This dissertation concentrates on researching and implementing an IIR system in LPA Prolog, that we call FUND, to assist in the matching of research proposals of prospective researchers to funding sources within the National Research Foundation (NRF). FUND'S reasoning strategy for its inference engine is backward chaining that carries out a depth-first search over its knowledge representation structure, namely, a semantic network. The distance constraint of the Constrained Spreading Activation (CSA) technique is incorporated within the search strategy to help prune non-relevant returns by FUND. The evolution of IIR from IR was covered in detail. Various reasoning strategies and knowledge representation schemes were reviewed to find the combination that best suited the problem domain and programming language chosen. FUND accommodated a depth 4, depth 5 and an exhaustive search algorithm. FUND'S effectiveness was tested, in relation to the different searches with respect to their precision and recall ability and in comparison to other similar systems. FUND'S performance in providing researchers with better funding advice in the South African situation proved to be favourably comparable to other similar systems elsewhere. / Thesis (M.Sc.)- University of Natal, Pietermaritzburg, 2001.

Thesis (M. Tech.) -- Central University of Technology, Free State, 2010

Thesis (MSc) -- Stellenbosch University , 2003. / ENGLISH ABSTRACT: The integration of symbolic knowledge with artificial neural networks is becoming an increasingly popular paradigm for solving real-world problems. This paradigm named knowledge-based neurocomputing, provides means for using prior knowledge to determine the network architecture, to program a subset of weights to induce a learning bias which guides network training, and to extract refined knowledge from trained neural networks. The role of neural networks then becomes that of knowledge refinement. It thus provides a methodology for dealing with uncertainty in the initial domain theory. In this thesis, we address several advantages of this paradigm and propose a solution for the open question of determining the strength of this learning, or inductive, bias. We develop a heuristic for determining the strength of the inductive bias that takes the network architecture, the prior knowledge, the learning method, and the training data into consideration. We apply this heuristic to well-known synthetic problems as well as published difficult real-world problems in the domain of molecular biology and medical diagnoses. We found that, not only do the networks trained with this adaptive inductive bias show superior performance over networks trained with the standard method of determining the strength of the inductive bias, but that the extracted refined knowledge from these trained networks deliver more concise and accurate domain theories. / AFRIKAANSE OPSOMMING: Die integrasie van simboliese kennis met kunsmatige neurale netwerke word 'n toenemende gewilde paradigma om reelewereldse probleme op te los. Hierdie paradigma genoem, kennis-gebaseerde neurokomputasie, verskaf die vermoe om vooraf kennis te gebruik om die netwerkargitektuur te bepaal, om a subversameling van gewigte te programeer om 'n leersydigheid te induseer wat netwerkopleiding lei, en om verfynde kennis van geleerde netwerke te kan ontsluit. Die rol van neurale netwerke word dan die van kennisverfyning. Dit verskaf dus 'n metodologie vir die behandeling van onsekerheid in die aanvangsdomeinteorie. In hierdie tesis adresseer ons verskeie voordele wat bevat is in hierdie paradigma en stel ons 'n oplossing voor vir die oop vraag om die gewig van hierdie leer-, of induktiewe sydigheid te bepaal. Ons ontwikkel 'n heuristiek vir die bepaling van die induktiewe sydigheid wat die netwerkargitektuur, die aanvangskennis, die leermetode, en die data vir die leer proses in ag neem. Ons pas hierdie heuristiek toe op bekende sintetiese probleme so weI as op gepubliseerde moeilike reelewereldse probleme in die gebied van molekulere biologie en mediese diagnostiek. Ons bevind dat, nie alleenlik vertoon die netwerke wat geleer is met die adaptiewe induktiewe sydigheid superieure verrigting bo die netwerke wat geleer is met die standaardmetode om die gewig van die induktiewe sydigheid te bepaal nie, maar ook dat die verfynde kennis wat ontsluit is uit hierdie geleerde netwerke meer bondige en akkurate domeinteorie lewer.

Orientador: Fernando José Von Zuben / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-16T10:52:32Z (GMT). No. of bitstreams: 1 Violato_RicardoParanhosVelloso_M.pdf: 5520106 bytes, checksum: 47f6f741b5c013a3252e50dddb37923c (MD5) Previous issue date: 2010 / Resumo: Este trabalho apresenta um estudo comparativo de três algoritmos de quantização vetorial, aplicados para a compressão de sinais de fala: k-médias, NG (do inglês Neural-Gas) e ARIA. Na técnica de compressão utilizada, os sinais são primeiramente parametrizados e quantizados, para serem armazenados e/ou transmitidos. Para recompor o sinal, os vetores quantizados são mapeados em quadros de fala, que são, por sua vez, concatenados, através de uma técnica de síntese concatenativa. Esse sistema pressupõe a existência de um dicionário (codebook) de vetores-padrão (codevectors), os quais são utilizados na etapa de codificação, e de um dicionário de quadros, que é utilizado na etapa de decodificação. Tais dicionários são gerados aplicando-se um algoritmo de quantização vetorial juntoa uma base de treinamento. Em particular, deseja-se avaliar o algoritmo imuno-inspirado denominado ARIA e sua capacidade de preservação da densidade da distribuição dos dados. São testados também diferentes conjuntos de parâmetros para identificar aquele que produz os melhores resultados. Por fim, são propostas modificações no algoritmo ARIA visando ganho de desempenho tanto na preservação de densidade quanto na qualidade do sinal sintetizado / Abstract: This work presents a comparative study of three algorithms for vector quantization, applied for the compression of speech signals: k-means, NG (Neural-Gas) and ARIA. In the compression technique used, the signals are first parameterized and quantized to be stored and/or transmitted. To reconstruct the signal, the quantized vectors are mapped into speech frames, which are concatenated through a concatenative synthesis technique. This system assumes the existence of a dictionary (codebook) of reference vectors (codevectors), which is used in the coding step, and a dictionary of frames, which is used in the decoding step. These dictionaries are generated by applying a vector quantization algorithm within a training database. In particular, we want to evaluate the immune-inspired algorithm called ARIA and its ability to preserve the density of data distribution. Different sets of parameters are also tested in order to identify the one that produces the best results. Finally, modifications to the ARIA algorithm are proposed aiming at obtaining gain in performance in both the preservation of density and the quality of the synthesized signal / Mestrado / Engenharia de Computação / Mestre em Engenharia Elétrica

One of the most impressive qualities of the brain is its neuro-plasticity. The neocortex has roughly the same structure throughout its whole surface, yet it is involved in a variety of different tasks from vision to motor control, and regions which once performed one task can learn to perform another. Machine learning algorithms which aim to be plausible models of the neocortex should also display this plasticity. One such candidate is the stacked denoising autoencoder (SDA). SDA's have shown promising results in the field of machine perception where they have been used to learn abstract features from unlabeled data. In this thesis I develop a flexible distributed implementation of an SDA and train it on images and audio spectrograms to experimentally determine properties comparable to neuro-plasticity. Specifically, I compare the visual-auditory generalization between a multi-level denoising autoencoder trained with greedy, layer-wise pre-training (GLWPT), to one trained without. I test a hypothesis that multi-modal networks will perform better than uni-modal networks due to the greater generality of features that may be learned. Furthermore, I also test the hypothesis that the magnitude of improvement gained from this multi-modal training is greater when GLWPT is applied than when it is not. My findings indicate that these hypotheses were not confirmed, but that GLWPT still helps multi-modal networks adapt to their second sensory modality.

M. Tech. (Department of Information Communication Technology, Faculty of Applied and Computer Sciences), Vaal University of Technology. / Too many options on the internet are the reason for the information overload problem to obtain relevant information. A recommender system is a technique that filters information from large sets of data and recommends the most relevant ones based on people‟s preferences. Collaborative and content-based techniques are the core techniques used to implement a recommender system. A combined use of both collaborative and content-based techniques called hybrid techniques provide relatively good recommendations by avoiding common problems arising from each technique. In this research, a proactive University Library Book Recommender System has been proposed in which hybrid filtering is used for enhanced and more accurate recommendations. The prototype designed was able to recommend the highest ten books for each user. We evaluated the accuracy of the results using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). A measure value of 0.84904 MAE and 0.9579 RMSE found by our system shows that the combined use of both techniques gives an improved prediction accuracy for the University Library Book Recommender System.