Research on Scalable Learning System and Its Evaluation

Learning System and Programs - Literature Review

In order to be able to examine the literature that is available regarding scalable learning systems and programs, it is necessary to begin with a discussion of some of the literature about what is meant by the concept of a scalable learning system. Porter, Chakrabarti, Harvey & Kenyon that a scalable learning system has the ability to change its size based on the amount of data or the number of layers that are part of the program. For example, in a learning system that involves decision trees, larger amounts of data for a given problem will be collected over time. However, as the larger amounts of data are collected, it is possible for a program to simply run out of room. A point can occur at which no more information can be added and no more information can be processed. With a scalable learning system, the algorithms that are used to process the information and to perform the tasks can be scaled based on the data that are added. In essence, there is virtually no limit to the amount of data over which algorithms and processes can occur (Li).

Another important aspect of a scalable learning system is that it must be flexible with regards to the connections that are made between nodes and the ways in which algorithms are performed (Li). This means that rather than an algorithm reading all of the data and becoming slowed down or even stopped, a learning process occurs in which algorithms only read data in batches and connect nodes that are necessary for a specific task or situation. The benefit in this is not just about speed and efficiency, but also about the ability to operate with limited memory. As the algorithms in a program examines specific batches of data, they are placed within a computer's memory. However, once those batches of data are no longer needed, they are removed frm the memory so that the program can examine other data and perform other tasks within the memory that memory (Li).

Porter, Chakrabarti, Harvey & Kenyon explained that the scale of the processing layers in scalable learning systems are essentially implied in relation to the changing size of the processes as they are performed. In this regard, the processes and algorithms that operate in such programs are indeed changing over time and are not necessarily the same size at any given time. This is indeed what makes scalable learning systems scalable in how they operate and carry out tasks and procedures. As the work that is performs changes, the programs and systems change and adapt.

The idea of a system changing and adapting based on the conditions and situations that exist is what Ma'ndziuk & Shastri noted as a system learning in increments based on smaller problems as opposed to having specific instructions to follow regardless of the situation that exists. The systems learn how to deal with smaller problems and provide information and results to users based on the problems that have been encountered. Based on the learning that occurs for one problem, then system or program will apply those results to other problems or situations that are encountered. In this way, the systems are not just scalable, but they are truly learning systems. Learning takes place by applying the information obtained from the problems faced in one situation to addressing a problem in another situation. On an overall basis, there is a learning process that occurs over time that allows the systems or programs to make connections and operate as efficiently as possible.

References

Li, X-B. A scalable decision tree system and its application in pattern recognition and intrusion detection. Decision Support Systems, 41(1), 112-130.

Ma'ndziuk, J. & Shastri, L. Incremental class learning approach and its applications to handwritten digit recognition. Proceedings of the Fifth International Conference on Neuro-Information processing, 21-23.

Porter, R., Chakrabarti, C., Harvey, N. & Kenyon, g. A scalable learning system for video recognition IEEE, March, 1-8.