Internet-Scale Pattern Recognition

Distributed Approach for Pattern Recognition 25

2.4 Pattern Distr ibution Techniques

Implementations of existing neural network/machine learning approaches

for pattern recognition have shown so me limitations. These include the gener-

alization problem and complex learning mechanisms. These limitations affect

the scalability of the approaches for real-time and large-s cale recognition de-

ployments. Furthermore, existing approaches are CPU-ce ntric, i.e., they have

been developed with the single-processing mechanism in mind. According to

Ikeda et al. [40], it is difficult for current neural network approaches to imple-

ment actual associative memo ry