ABSTRACT
This chapter demonstrates the need for model validation with some illustrative examples. It proposes a stochastic similarity measure – based on Hidden Markov Model analysis – for comparing stochastic, dynamic, multi-dimensional trajectories. The chapter suggests that post-training model validation is not only desirable, but essential to establish the degree to which the human and model-generated trajectories are similar. In general, similarity between model-generated trajectories and the human control data will vary continuously for different models, from completely dissimilar to nearly identical. In speech recognition, where Hidden Markov Model (HMM) has found their widest application, human auditory signals are analyzed as speech patterns. Transient sonar signals are classified with HMMs for ocean surveillance in. G. Radons, et.al. analyze 30-electrode neuronal spike activity in a monkey’s visual cortex with HMMs.
