Adaptive Resonance Theory, Self-Organizing Maps, and Survival Analysis
Machine learning allows computers to mimic human behavior by training them with historical and predicted data. This section will examine some of the most exciting machine learning algorithmssuch as adaptive resonance theory, self-organizing maps and survival analysis.
Adaptive resonance theory
Stephen Grossberg and Gail Carpenter created the neural network technique known as Adaptive Resonance Theory in 1987. The technique of unsupervised learning is at the heart of ART. Stephen Grossberg and Gail Carpenter developed the adaptive resonance theory (ART) to explain aspects of how the brain processes information. It describes various neural network models that address problems such as pattern recognition and prediction using supervised and unsupervised learning techniques.
The underlying assumption of the ART model is that, in general, object identification and recognition results from the interaction of “top-down” observer expectation and “bottom-up” sensory input. Depending on the model, “top-down” expectations are represented by a memory model or a prototype. So it is with the fundamental characteristics of an object as they are perceived by the senses. The captured object will be considered as belonging to the expected class if this difference between sensation and expectation does not exceed a predetermined limit called “vigilance parameter”. The system thus provides a solution to the problem of “plasticity/stability”, or the problem of learning new information without altering previously learned information, also called incremental learning.
Self-Organized Maps (SOM), also known as Kohonen map or Kohonen network, was created in the 1980s. Kohonen map or network is a computationally efficient simplification. It is Alan Turing’s morphogenesis of the 1950s and the biological models of neural systems of the 1970s. It is a type of artificial neural network inspired by the biological models of neural systems of the 1970s.
SOM is an unsupervised machine learning technique. It produces a low-dimensional (usually two-dimensional) representation of a higher-dimensional dataset while preserving its topological structure. Moreover, it uses an unsupervised learning strategy and a competitive learning algorithm to train its network. SOM is for clustering and mapping (or dimensionality reduction) techniques. It maps multidimensional data onto lower dimensional data, simplifying the interpretation of complex problems.
Survival analysis is a branch of statistics that examines how long it is likely that something like a person will die or a machine will break down. In engineering, this is called reliability theory or reliability analysis. In economics, this is called duration analysis or modeling; in sociology, this is called the analysis of the history of events.
In a broader sense, survival analysis involves modeling time-to-event data; in this context, failure or death are “events” in the literature on survival analysis; historically, only one event occurs for each subject, as a result of which the organism or mechanism is dead or broken. The assumption is relaxed in repeated or recurring event models. Nevertheless, in many areas of social science and medical research, as well as system reliability, the study of regular events is essential.