General Pattern Recognition

An immense number of books and papers dealing with pattern recognition has been published by mathematicians, statisticians and other professional pattern recognizers . Nevertheless, it is hard to find introductory text-books which make pattern recognition clear to a chemist who has no experience in this field. A very subjective selection of references is given below. 

---

The general pattern recognition problem can be described as follows. The input data, or pattern X, is defined by a number of specific data measurements, x, defined at a particular point in time ... 

Campbell, J.L. and Johnson, K.E. (1993). Abductive Networks Generalization, Pattern Recognition, and Prediction of Chemical Behavior. Can.J.Chem., 71,1800-1804. 

Numerous reviews have been written about chemical applications of pattern recognition. Many of them include rather easy to read introductions to general pattern recognition methods. At least some of these papers and books should be read by beginners of pattern recognition in chemistry. 

The problem of invariant pattern recognition is recognized as being a highly complex and difficult one. It is not surprising, therefore, that a wide variety of techniques have been invented to deal with specific or general instances of this problem. 

The general invariant pattern recognition problem is to construct a system which takes as input an element/of V and computes a value s(f), with the intention that s(f) = c(f) for all f V. 

In 1972, the concept of pattern recognition as a general problem solving tool for a broad scope of chemical appHcations was introduced (9,10). 

Yeh and Spiegelman [24], Very good results were also obtained by using simple neural networks of the type described in Section 33.2.9 to derive a decision rule at each branching of the tree [25]. Classification trees have been used relatively rarely in chemometrics, but it seems that in general [26] their performance is comparable to that of the best pattern recognition methods. 

The similarity in approach to LDA (Section 33.2.2) and PLS (Section 33.2.8) should be pointed out. Neural classification networks are related to neural regression networks in the same way that PLS can be applied both for regression and classification and that LDA can be described as a regression application. This can be generalized all regression methods can be applied in pattern recognition. One must expect, for instance, that methods such as ACE and MARS (see Chapter 11) will be used for this purpose in chemometrics. 

The most serious problem with input analysis methods such as PCA that are designed for dimension reduction is the fact that they focus only on pattern representation rather than on discrimination. Good generalization from a pattern recognition standpoint requires the ability to identify characteristics that both define and discriminate between pattern classes. Methods that do one or the other are insufficient. Consequently, methods such as PLS that simultaneously attempt to reduce the input and output dimensionality while finding the best input-output model may perform better than methods such as PCA that ignore the input-output relationship, or OLS that does not emphasize input dimensionality reduction. 

From both a theoretical and practical view, it is ideal to use Bayesian Decision Theory because it represents an optimal classifier. From a theoretical perspective, Bayesian Decision Theory offers a general definition of the pattern recognition problem and, with appropriate assumptions, it can be shown to be the basis of many of the so-called non-PDF approaches. In practice, however, it is typically treated as a separate method because it places strong data availability requirements for direct use compared to other approaches. 

When describing mathematical modeling in general (not just for classification of bacteria), it is important to point out the mathematical meaning of pattern recognition the mapping of an n-dimensional function to describe a set of... 

Because of their ability to classify complex data types that have no explicit mathematical model, neural networks have become a powerful and widely used approach to pattern recognition problems in general. A neural network is a series of mathematical operations performed on input data that ultimately... 

Seelig, A., A general pattern for substrate recognition by P-glyco-protein, Eur. J. Biochem. 1998, 252, 252-261. 

Kowalski and Bender presented chemometrics (at this time called pattern recognition and roughly considered as a branch of artificial intelligence) in a broader scope as a general approach to interpret chemical data, especially by mapping multivariate data with the purposes of cluster analysis and classification (Kowalski and Bender 1972). 

In general, it appears that expert systems which combine symbolic/numeric processing capabilities are necessary to effectively automate decision-making in applications involving analytical and process instrumentation/sensors. Furthermore, these integrated decision structures will likely be embedded (67-69) within the analytical or process units to provide fully automated pattern recognition/correlation systems for future intelligent instrumentation. 

Without going into the details of the numerous techniques that are being used in pattern recognition, a general outline of the method of problan handling by means of the ARIHUR package may be clearly illustrated fran an approach to the air pollution problem. (See Table I)... 

---