| Cognitive science is the interdisciplinary study of mind and intelligence, embracing artificial intelligence (AI), psychology, philosophy, linguistics, anthropology, neuroscience, etc, which is concerned with the representation, processing and transmission of information and knowledge by nervous system (such as human brain) and machine system (such as computer) in the process of perception, language, storage, reasoning, emotion, and so on. Researchers from different disciplines have studied the cognitive behavior of human brain and machine from different perspectives. In cognition, concept could be considered as a basic unit. The cognition of uncertain concepts has become an important foundation of cognitive research since uncertainty is inherent in the objective world and human cognition. Although there are many theoretical models, such as fuzzy sets, rough sets, quotient spaces, etc, to research this problem, most of them have some limitations due to the accurate logical and numerical computation. In fact, human deals with uncertain concepts based on words (concept intension), while computer based on sample set (concept extension). Therefore, the bidirectional cognitive computing between uncertain concept’s intension and extension is a key issue to achieve the unity of human brain thinking and computer intelligent computing. However, traditional machine learning and data mining have only studied the unidirectional cognitive computing, i.e., from concept’s extension to its intension.Turing award was awarded to Professor Judea Pearl in 2011, who introduced the Bayesian networks and probabilistic methods into artificial intelligence. This momentous event illustrates the importance of uncertainty theories and methods based on probability statistics. Cloud model is a probability statistics-based bidirectional cognitive model. Thus, this dissertation aims at constructing bidirectional cognitive computing method based on cloud model, and studying the bidirectional cognitive transformation between the intension and the extension of concept on the basis of probability statistics. The main contributions of this dissertation are listed as follows.(1) The recursive definition of pth-order (p≥2) normal cloud based on normal cloud transformation algorithm is presented, and some mathematical properties of the normal cloud are studied, which further develops the mathematical basis of cloud model.Firstly, this dissertation presents the probability distributions of the second-order normal cloud certainty degree in different scenarios based on the conditional probability, and then corrects the existing joint probability distribution function of the second-order normal cloud drops and their certainty degree. Secondly, this dissertation gives the recursive definition of pth-order normal cloud based on normal cloud transformation algorithm, and presents the probability density function (pdf) of the certainty degree of pth-order normal cloud. After the above mentioned contributions, an important conclusion is reached that the pdf of the certainty degree of pth-order normal cloud is irrelevant to numerical characteristics or order p. (See Chapter 2 for more details.)(2) Two stable multi-step backward cloud transform algorithms are proposed to address existing methods’defects, which establish the foundation for realizing the stable bidirectional cognitive computing.A detailed discussion of limitations in existing second-order backward cloud transformation algorithms is done, and to address these defects, two stable multi-step backward cloud transformation algorithms are proposed by means of parameter estimation and random sampling which are implemented in a grouping and reducing order way. The unbiasedness, the consistency and the convergence of the estimators in the proposed methods are evaluated by some evaluation criteria. The stability of the proposed methods is illustrated by experiments. Whereafter, the proposed methods are extended to Pth-order (p≥2) normal cloud, and the corresponding algorithms, which can be used to calculate the parameters of pth-order cloud model, are given. (See Chapter 3 for more details.)(3) The generic normal cloud model is proposed, and the relationship between normal cloud model and normal distribution is established, which further perfects the cloud model theory.Based on the traits of normal cloud drops, a second-order general forward normal cloud transformation algorithm is presented, which makes the second-order normal cloud and normal distribution become two special cases of the generic normal cloud. Then according to the mutually inverse characteristics between forward cloud and backward cloud, the second-order general backward cloud transformations based on ideal grouping and random grouping are given, and the relationship between them is also analyzed. The proposed methods are subsequently extended to pth-order (p≥2) generic normal cloud model, and pth-order generic forward cloud transformation and pth-order backward cloud transformation are obtained, which further improve the cloud model theory. (See Chapter 4 for more details.)(4) Bidirectional cognitive computing model is put forward, and combined with the characteristics of human cognition and the cognitive transformation algorithm, simulation of the bidirectional cognitive computing process is implemented by the way of calculation.Cloud model is an uncertainty transformation model between a qualitative concept (concept’s intension) expressed by linguistic values and its quantitative representation (concept’s extension), and the transformation is conducted by cloud transformation algorithms. Forward cloud transformation and backward cloud transformation are employed to study the universal laws contained in concepts, which make it possible to obtain the range and the distribution of quantitative data (concept’s extension) from qualitative information (concept’s intension), and to transform exact numerical values into appropriate qualitative concepts. For this reason, the bidirectional cognitive computing model is proposed. The model applies the forward and the backward cloud transformation algorithms to simulate the cognitive computing process for the uncertain concepts in a calculation way. In addition, the symmetrical Kullback-Leibler divergence is used to measure the excursion during the process of concepts’cognition. Whereafter, the bidirectional cognitive computation process is simulated by some numerical experiments. (See Chapter 5 for more details.)(5) Based on the characteristics of human cognition on image, applications of the bidirectional cognitive transformation algorithm are made to image segmentation.Image segmentation is to partition an image into non-overlapping regions according to the features of image so as to extract the object regions of interest. The whole image segmentation process consists of three steps in this method. Cloud concepts are first obtained from the image information (such as image grayscale, color, etc) availed by the cognitive transformation algorithms. Then a merged criterion for cloud concepts based on the characteristics of human cognition on image is built, and subsequently, a new cloud integrated approach is put forward to conduct the concept jumping-up, and finally the image is segmented into several parts (the foreground, the background) assisted by the "3En" rule of normal cloud. The proposed method is applied to segment gray/color images and images with noises, and its validity and effectiveness are shown by comparing with some classical methods, such as Otsu, K-means, GMM, S-PCNN, Type-2 fuzzy set, Atanassov’s intuitionistic fuzzy set(A-IFS) and cloud model method. (See Chapter 6 for more details.)... |
PDF Full Text Request