Classifier approaches

Wang and Hirschberg [363] introduced the idea of using decision trees for phrase break prediction. Decision trees allow a wide variety of heterogeneous features, examples of whieh are given below  

These features (and others) were used as question in the tree, which was grown in the normal way. Wang and Hirschberg trained and tested Iheir system on 298 utterances from the ATIS corpus, giving the following results (the results aren t available in the formats we described above)  

It should be pointed out that this approach can t strictly be used for TTS purposes as acoustic features (e.g. time in seconds) measured from the corpus waveforms were used in addition to features that would be available at run time. Following this initial work, a number of studies have used decision trees [264] [418], and a wide variety of other machine learning algorithms have been applied to the problem including memory based learning [77] [402], Bayesian classifiers [516], support vector machines [87] and neural networks [157]. Similar results are reported in most cases, and it seems that the most important factors in the success of a system are the features used and the quality and quantity of data rather than the particular machine learning algorithm used. 

---

With this convention, we can now classify energy transfer processes either as resonant, if IA defined in equation (A3.13.81 is small, or non-resonant, if it is large. Quite generally the rate of resonant processes can approach or even exceed the Leimard-Jones collision frequency (the latter is possible if other long-range potentials are actually applicable, such as by pennanent dipole-dipole interaction). 

The fomialism outlined in the previous sections is very usefiil for small systems, but is, as explained, impractical for more than six to ten strongly interacting degrees of freedom. Thus, alternate approaches are required to represent dynamics for large systems. Currently, there are many new approaches developed and tested for this purpose, and these approaches are broadly classified as follows ... 

We employ the general scheme presented above as a starting point in our discussion of various approaches for handling the R-T effect in triatomic molecules. We And it reasonable to classify these approaches into three categories according to the level of sophistication at which various aspects of the problem are handled. We call them (1) minimal models (2) pragmatic models (3) benchmark treatments. The criterions for such a classification are given in Table I. 

Model-driven approaches classify reactions according to a preconceived model, a conceptual framework. 

Another approach to obtain an overview on chemical information or on information related to specified topies in chemistry, is to use websites that eontain link lists. These link lists arc usually provided by universities and private persons and are classified into subject areas, Table 5-7 gives an sample of the thousands of link lists in chemistry, and in addition some other valuable URLs that deal with chcmoinformatics. 

Woodruff and co-workers introduced the expert system PAIRS [67], a program that is able to analyze IR spectra in the same manner as a spectroscopist would. Chalmers and co-workers [68] used an approach for automated interpretation of Fourier Transform Raman spectra of complex polymers. Andreev and Argirov developed the expert system EXPIRS [69] for the interpretation of IR spectra. EXPIRS provides a hierarchical organization of the characteristic groups that are recognized by peak detection in discrete ames. Penchev et al. [70] recently introduced a computer system that performs searches in spectral libraries and systematic analysis of mixture spectra. It is able to classify IR spectra with the aid of linear discriminant analysis, artificial neural networks, and the method of fe-nearest neighbors. 

Note that a statistical study could be done on an electron micrograph like that shown in Fig. 1.1. The dimensions of the blobs could be converted to volumes and then to masses with a knowledge of the density of the deposited polymer. This approach could be organized into a table of classified data from which any of these averages could be calculated. 

Examination of the various classified listings of herbicides provides iasight iato the processes and approaches that lead to the discovery of new pesticides. The four principal development approaches are random screening, imitative chemistry, testing natural products, and biorational development. 

Classification of the many different encapsulation processes is usehil. Previous schemes employing the categories chemical or physical are unsatisfactory because many so-called chemical processes involve exclusively physical phenomena, whereas so-called physical processes can utilize chemical phenomena. An alternative approach is to classify all encapsulation processes as either Type A or Type B processes. Type A processes are defined as those in which capsule formation occurs entirely in a Hquid-filled stirred tank or tubular reactor. Emulsion and dispersion stabiUty play a key role in determining the success of such processes. Type B processes are processes in which capsule formation occurs because a coating is sprayed or deposited in some manner onto the surface of a Hquid or soHd core material dispersed in a gas phase or vacuum. This category also includes processes in which Hquid droplets containing core material are sprayed into a gas phase and subsequentiy solidified to produce microcapsules. Emulsion and dispersion stabilization can play a key role in the success of Type B processes also. 

Constraint control strategies can be classified as steady-state or dynamic. In the steady-state approach, the process dynamics are assumed to be much faster than the frequency with which the constraint control appHcation makes its control adjustments. The variables characterizing the proximity to the constraints, called the constraint variables, are usually monitored on a more frequent basis than actual control actions are made. A steady-state constraint appHcation increases (or decreases) a manipulated variable by a fixed amount, the value of which is determined to be safe based on an analysis of the proximity to relevant constraints. Once the appHcation has taken the control action toward or away from the constraint, it waits for the effect of the control action to work through the lower control levels and the process before taking another control step. Usually these steady-state constraint controls are implemented to move away from the active constraint at a faster rate than they do toward the constraint. The main advantage of the steady-state approach is that it is predictable and relatively straightforward to implement. Its major drawback is that, because it does not account for the dynamics of the constraint and manipulated variables, a conservative estimate must be taken in how close and how quickly the operation is moved toward the active constraints. 

Liquid-Phase Components. It is usual to classify organic Hquids by the nature of the polar or hydrophilic functional group, ie, alcohol, acid, ester, phosphate, etc. Because lowering of surface tension is a key defoamer property and since this effect is a function of the nonpolar portion of the Hquid-phase component, it is preferable to classify by the hydrophobic, nonpolar portion. This approach identifies four Hquid phase component classes hydrocarbons, polyethers, siHcones, and duorocarbons. 

Dyes may be classified according to chemical stmcture or by thek usage or appHcation method. The former approach is adopted by practicing dye chemists who use terms such as a2o dyes, anthraquinone dyes, and phthalocyanine dyes. The latter approach is used predominantiy by the dye user, the dye technologist, who speaks of reactive dyes for cotton and disperse dyes for polyester. Very often, both terminologies are used, for example, an a2o disperse dye for polyester and a phthalocyanine reactive dye for cotton. 

Extension of the approach discussed above suggests several structural types which may be classified under this category, the main variable being the nature and hybridization of the electrophilic centers and the nature of the atoms joining these two centers. They may be conveniently divided into two groups ... 

The preparation of esters can be classified into two main categories (1) carboxy-late activation with a good leaving group and (2) nucleophilic displacement of a caiboxylate on an alkyl halide or sulfonate. The latter approach is generally not suitable for the preparation of esters if the halide or tosylate is sterically hindered, but there has been some success with simple secondaiy halides and tosylates (ROTs, DMF, K2CO3, 69-93% yield). ... 

Another common approach is to use an information-processing model to classify human errors. The classification models the information processing which occurs when a person operates and controls complex systems such as processing plants. One such classification (Rouse and Rouse, 1983) identifies six steps in information processing. Exhibit 6.1 lists the six steps, and provides some examples of errors that can occur at each of these steps. 

The many methods used in kinetic studies can be classified in two major approaches. The classical study is based on clarification of the reaction mechanism and derivation of the kinetics from the mechanism. This method, if successful, can supply valuable information, by connecting experimental results to basic information about fundamental steps. During the study of reaction mechanisms many considerations are involved. The first of these is thermodynamics, not only for overall reactions, but also on so-called elementary steps. 

---