6. Bottom-up and Top-down Approaches to Modelling Context

6.2. Bottom-up

In other domains it is much more difficult to establish general principles for learning and inference. Here a more bottom-up approach needs to be taken. A small sub-domain is typically chosen and then relevant examples considered to establish the likely heuristics involved. In this approach the specific data and facts come first and the more abstract principles and theories come second. The models are formulated to capture or explain observed processes and will be judged in this light. Later more abstract models (or laws) might be posited from testing against these models and the data. Thus this approach could be dubbed the `scientific' approach.

The bottom-up approach is perhaps taken most seriously by those who advocate a constructivist approach to AI. Here care is taken to assume as little as possible in advance so that as much as possible of the behaviour is available for capture in the models induced [17].

Figure 2. Top-down and bottom-up approaches to the investigation of context

The community interested in context is unevenly split into these two approaches. The `foundationalists' are searching for a sort of mathematics of context, their approach is principle-based and has a potentially general applicability. However they are dogged by problems of scalability from the toy-problems they are tested on and will only be as strong as their a priori principles turn out to be. The `scientists' are typically working in a specific domain with more realistic processes and problems, they face the same difficulties of generalisation as other scientists - it is a slow and difficult process to discover successful theories. On the other hand any progress they make will be strongly grounded in real processes and have clear conditions of applicability. In other sciences both kinds of approach have turned out to be useful, the foundationalists have typically had a role in producing a palate of formalisms, a few of which turn out to be useful to the scientists who use them for describing or modelling the actual phenomena*1. The high odds against a particular type of formalism turning out to be useful means that it is vital that the maximum possible variety of approaches be developed. The scientists have the job of finding the mappings from the phenomena concerned to models expressed in these formalisms. This job is harder, which possibly explains why they are in the minority.

The Pragmatic Roots of Context - Bruce Edmonds - 31 MAR 99

[Next] [Previous] [Up] [Top] [Contents]

Generated with CERN WebMaker