Classical methods for interpreting objective function minimization as intelligent inference
- Univ. of Texas, Dallas, TX (United States)
Most recognition algorithms and neural networks can be formally viewed as seeking a minimum value of an appropriate objective function during either classification or learning phases. The goal of this paper is to argue that in order to show a recognition algorithm is making intelligent inferences, it is not sufficient to show that the recognition algorithm is computing (or trying to compute) the global minimum of some objective function. One must explicitly define a {open_quotes}relational system{close_quotes} for the recognition algorithm or neural network which identifies the: (i) sample space, (ii) the relevant sigmafield of events generated by the sample space, and (iii) the {open_quotes}relation{close_quotes} for that relational system. Only when such a {open_quotes}relational system{close_quotes} is properly defined, is it possible to formally establish the sense in which computing the global minimum of an objective function is an intelligent, inference.
- OSTI ID:
- 466459
- Report Number(s):
- CONF-9610138--
- Country of Publication:
- United States
- Language:
- English
Similar Records
Multidimensional neural growing networks and computer intelligence
Intelligent robots and computer vision IX: Algorithms and techniques; Proceedings of the Meeting, Boston, MA, Nov. 5-7, 1990