Computer and Robot Vision Vol. 1, by R.M. Haralick and Linda G. Shapiro, Addison-Wesley, 1992, ISBN 0-201-10887-1.

Computer and Robot Vision

Suggested by the structure of the visual nervous system, a new algorithm is proposed for pattern recognition. This algorithm can be realized with a multilayered network consisting of neuron-like cells. The network, “neocognitron”, is self-organized by unsupervised learning, and acquires the ability to recognize stimulus patterns according to the differences in their shapes: Any patterns which we human beings judge to be alike are also judged to be of the same category by the neocognitron. The neocognitron recognizes stimulus patterns correctly without being affected by shifts in position or even by considerable distortions in shape of the stimulus patterns.

Neocognitron--A New Algorithm for Pattern Recognition Tolerant of Deformations and Shifts in Position

This paper presents a novel adaptive resonance theory (ART)-based modular architecture for unsupervised learning, namely the distributed dual vigilance fuzzy ART (DDVFA). DDVFA consists of a global ART system whose nodes are local fuzzy ART modules. It is equipped with the distinctive features of distributed higher-order activation and match functions, using dual vigilance parameters responsible for cluster similarity and data quantization. Together, these allow DDVFA to perform unsupervised modularization, create multi-prototype clustering representations, retrieve arbitrarily-shaped clusters, and control its compactness. Another important contribution is the reduction of order-dependence, an issue that affects any agglomerative clustering method. This paper demonstrates two approaches for mitigating order-dependence: preprocessing using visual assessment of cluster tendency (VAT) or postprocessing using a novel Merge ART module. The former is suitable for batch processing, whereas the latter can be used in online learning. Experimental results in the online learning mode carried out on 30 benchmark data sets show that DDVFA cascaded with Merge ART statistically outperformed the best other ART-based systems when samples were randomly presented. Conversely, they were found to be statistically equivalent in the offline mode when samples were pre-processed using VAT. Remarkably, performance comparisons to non-ART-based clustering algorithms show that DDVFA (which learns incrementally) was also statistically equivalent to the non-incremental (offline) methods of DBSCAN, single linkage hierarchical agglomerative clustering (HAC), and k-means, while retaining the appealing properties of ART. Links to the source code and data are provided. Considering the algorithm's simplicity, online learning capability, and performance, it is an ideal choice for many agglomerative clustering applications.

Distributed dual vigilance fuzzy adaptive resonance theory learns online, retrieves arbitrarily-shaped clusters, and mitigates order dependence

Distributed dual vigilance fuzzy adaptive resonance theory learns online, retrieves arbitrarily-shaped clusters, and mitigates order dependence.

The term “big data” is now becoming more and more important in many fields. This data should not only be gathered, but also analyzed and, in some cases, classified. The categorization of each sample is becoming increasingly multifaceted, since it often means to assign not only one category from one ontology but multiple labels from multiple ontologies. This study investigates the improvement of classification performance in such multi-label problems with the help of association rule mining. The innovative character of the study lies in the use of an extended neural classifier based on adaptive resonance theory-networks, as well as of rare association rule mining to extract useful knowledge from classification data. The central hypothesis of the study is that the discovery of deep connections between multi-labels of different taxonomies improves the prediction of the classifier system and allows the extraction of interesting knowledge. This is based on the fact that classifiers can learn well labels that are high in the ontology, since these labels have many examples. The deeper the labels, the fewer samples the classifier will have for training, and other methods need to be developed to cope with this difficulty. On the other hand, some taxonomies are easier to learn and their predictions, together with association rules, can help to increase the prediction quality of the system. Further, since the classification system in the big data setup will become increasingly complex, the interaction of experts with the system should allow to identify conflicts in the classification rules and to correct them. To formalize the problem, we seek to create a classification model that maps objects from a description space into a set of classes given by multiple taxonomies (taxonomy space). Here, we prefer models that provide sensible and understandable rules, as they allow examining and verifying the knowledge that they extract. We assume that the taxonomies’ spaces are so complex that for the practical use, each taxonomy is organized as a tree. We expect the most interesting connections between taxonomies to be found in the cases when the taxonomies are very different, both conceptually and in nature (including those mapped by different structures). In such scenario, we expect new, surprising knowledge to arise from the connections between the taxonomies. The main contribution of the thesis is the extensive examination of using rare association rules for the improvement of multi-label predictions in the setup of cross-ontology classification, especially the proposed approach called Multi-label Improvement with Rare Association Rules (MIRAR). A further contribution is the ML-HARAM, a hierarchical multi-label classifier based on the Adaptive Resonance Theory (ART). The last contribution is the Rule Explorer, a graphical user interface to analyze each step of the classification process in depth: from the creation of the classification rule over its examination up to the use with interestingness measures to improve the prediction quality of the classifier and its predicted labels. Extensive experiments indicate, under statistical significance, that the results support well the hypothesis and, with common multi-label classification measures, the results achieved were over state-of-the-art performance.

Multi-label Classification with Multiple Class Ontologies

In this paper, a cascade fuzzy ART (CFART) network is developed and applied to 3D object recognition. The proposed CFART network contains multiple layers which can express a hierarchical representation of an input pattern. The learning processes of the proposed network are unsupervised and self-organizing, which include a top-down search process and a bottom-up learning process. The proposed CFART can accept both binary and analog inputs. With fast learning and categorization capabilities, the proposed network is capable of acting as an extensible database, providing a multi-resolutional representation of 3D objects. In the experiments, we use superquadrics as a demonstration example.

CFART: a multi-resolutional adaptive resonance system

In this report, we propose a cascade fuzzy ART (CFART) neural network which can function as an extensible database in a model-based object recognition system. The proposed CFART network contains multiple layers. It preserves the prominent characteristics of a fuzzy ART network and extends fuzzy ART's capability toward hierarchical representation of input patterns. The learning process of the proposed network is unsupervised and self-organizing, and includes a top-down searching process and a bottom-up learning process. The top-down and bottom-up learning processes interact with each other in a closely coupled manner. Basically, the top-down searching guides the bottom-up learning whereas the bottom-up learning in uences the top-down searching by changing its searching fuzzy boundary. In addition, a global searching tree is built to speed up the learning and recognition processes. The proposed CFART can accept both binary and analog inputs. With fast learning and categorization capabilities, the proposed network is able to function like an extensible database and to provide an e cient multi-resolutional representation capability for 3D objects. Experimental results, using both synthetic and real 3D data, prove that the proposed method is indeed an e cient and powerful representation scheme. 2

Shing-Jong Lin

Papers

CFART: a multi-resolutional adaptive resonance system

CFART: a New Multi-Resolutional Adaptive Resonance System for Object Recognition