Inductive learning

The secret to success has been to learn from data and from experiments. Chemists have done a series of experiments, have analyzed them, have looked for common features and for those that are different, have developed models that made it po.ssiblc to put these observation.s into a systematic ordering scheme, have made inferences and checked them with new experiments, have then confirmed, rejected, or relined their models, and so on. This process is called inductive learning (Figure 1 -1), a method chemists have employed from the veiy beginnings ol chcmistiy. 

However, this process of inductive learning is still not over we arc still far away from understanding and predicting all chemical phenomena. This is most vividly illustrated by our poor knowledge of the tindcsircd side effects of compounds, such as toxicity. Wc still have to strive to increase our knowledge of chemistry,... 

There is, however, another type of learning inductive learning. From a series of observations inferences are made to predict new observations. In order to be able to do this, the observations have to be put into a scheme that allows one to order them, and to recognize the features these observations have in common and the essential features that are different. On the basis of these observations a model of the principles that govern these observations must be built such a model then allows one to make predictions by analogy. 

Inductive learning has been the major process of acquiring chemical knowledge from the very beginnings of chemistry - or, to make the point, alchemy. Chemists have done experiments, have made measurements on the properties of their compounds, have treated them with other compounds to study their reactions, and have run reactions to make new compounds. Systematic variations in the structure of compounds, or in reaction conditions, provided results that were ordered by developing models. These models then allowed predictions to be made. 

In the endeavor to deepen understanding of chemistry, many an experiment has been performed, and many data have been accumulated. Chapter 6, on databases, gives a vivid picture of the enormous amount of data that have been determined and made accessible. The task is then to derive knowledge from these data by inductive learning. In this context we have to define the terms, data, information, and knowledge, and we do so in a generally accepted manner. 

In the case of chemoinformatics this process of abstraction will be performed mostly to gain knowledge about the properties of compounds. Physical, chemical, or biological data of compounds will be associated with each other or with data on the structure of a compound. These pieces of information wQl then be analyzed by inductive learning methods to obtain a model that allows one to make predictions. 

Inductive methods for establishing a correlation between chemical compounds and their properties are the theme of Chapter 9. In many cases, the structure of chemical compounds has to be pre-processed in order to make it amenable to inductive learning methods. This is usually achieved by means of structure descriptors, methods for the calculation of which are outlined in Chapter 8. 

Now, chemists have acquired much of their knowledge on chemical reactions by inductive learning from a large set of individual reaction instances. How has this been done And how can we build on these methods and knowledge and perform it in a more systematic manner by algorithmic techniques ... 

The first step in an inductive learning process is always to order the observations to group those objects together that have essential features in common and to separate objects that are distinctly different. Thus, in learning from individual reactions we have to classify reactions - we have to define reaction types that encompass a series of reactions with essential common characteristics. Clearly, the definition of what are essential common features is subjective and thus a variety of different classification schemes have been proposed. 

Thus, when a large set of chemical reactions has to be investigated, an inductive learning process, deriving knowledge on chemical reactions and reactivity from a series of reactions, still has many merits. Such chemical knowledge can be put into models that then allow one to predict the course of new reactions. 

On top of that, reaction databases can also be used to derive knowledge on chemical reactions which can then be used for reaction prediction, The huge amount of information in reaction databases can be processed by inductive learning methods in order to condense these individual pieces of information into essential features... 

The understanding and simulation of chemical reactions is one of the great challenges of chemoinformatics. Each day millions of reactions are performed, sometimes with rather poor results because of our limited understanding of chemical reactivity and the influence of solvents, catalysts, temperature, etc. This problem has to be tackled by both deductive and inductive learning methods. 

The search procedure, S, used to uncover promising hyperrectangles in the decision space, X, associated with a desired y value (e.g., y = good ), is based on symbolic inductive learning algorithms, and leads to the identification of a final number of promising solutions, X, such as the ones in Fig. 2b. It is described in the following subsection. 

Task 2. Inductive learning of the relationship between the features of... 

The inductive learning process determines the discriminant functions, using prior examples of (p,C,) associations. 

Once the several records of a process variable have been generalized into a pattern, as indicated in the previous paragraph, we need a mechanism to induce relationships among features of the generalized descriptions. In this section we will discuss the virtues of inductive learning through decision trees. 

Inductive learning by decision trees is a popular machine learning technique, particularly for solving classification problems, and was developed by Quinlan (1986). A decision tree depicting the input/output mapping learned from the data in Table I is shown in Fig. 22. The input information consists of pressure, temperature, and color measurements of... 

Once the distinguishing features for each input variable have been extracted through the generalization of descriptions of the available records, these features become the inputs of the inductive learning procedure through decision trees. 

For the detailed discussion on the inductive learning of diagnostic and control rules around the fed-batch fermentor system, the reader should refer to the work of Bakshi and Stephanopoulos (1994b). 

The four-volume Handbook of Chemoinformatics—From Data to Knowledge (Gasteiger 2003) contains a number of introductions and reviews that are relevant to chemometrics Partial Least Squares (PLS) in Cheminformatics (Eriksson et al. 2003), Inductive Learning Methods (Rose 1998), Evolutionary Algorithms and their Applications (von Homeyer 2003), Multivariate Data Analysis in Chemistry (Varmuza 2003), and Neural Networks (Zupan 2003). 

RuleMaster is a general-purpose software package for building and delivering expert systems. Its features include l) knowledge acquisition by inductive learning,... 

Inductive Learning (RuleMaker). Experts are best able to explain complex concepts to human apprentices implicitly by using examples of the expert s decision-making, rather than by explicitly stating fundamental theoretical principles. The apprentice quickly generalizes these example decisions to form working rules, which he applies when similar situations are encountered. 

GloveAID ( ) predicts the most effective glove materials to choose for protection against hazardous chemicals. There are no established experts in this field, because much of the protection effectiveness measurements are Just now being performed. The inductive learning aspect of RuleMaster is used to help organize the date which is available and to suggest which measurements should be performed next. 

---