If you have any problems related to the accessibility of any content (or if you want to request that a specific publication be accessible), please contact (email@example.com). We will work to respond to each request in as timely a manner as possible.
Recursive Hyperspheric Classification
AuthorReed, Salyer Byberg
Computer Science and Engineering
StatisticsView Usage Statistics
Data guide decisions and processes. Moreover, while the practice of obtaining data may be mundane, the analysis of data provides vision and understanding. Within data -- ideally -- are hidden patterns and metadata that must be identified and culled. Without insight gleaned and garnered from data, decisions and judgments would be futile and lack accountability. Additionally, data are ubiquitous and can be extracted from just about any observable -- either tangible or intangible -- entity. As the data are obtained from quantifiable observations and recordings, they are aggregated and stored for analysis and consumption. This work is concerned with the classification of labeled data as well as the recognition of unlabeled data, endeavoring to create intelligent systems that are able to generalize and illicit satisfactory responses to various inputs, or stimuli. In particular, the algorithm employed to create these intelligent systems is termed Recursive Hyperspheric Classification.Recursive Hyperspheric Classification (RHC) is a novel supervised learning algorithm that uniquely classifies noisy data sets by spawning a hierarchy of hyperspheres, or balls. Utilizing standard tree traversal algorithms, hyperspheres are generated, partitioning and classifying the rich feature space. Once mapped, the resulting taxonomic data structure may be used to recognize unlabeled vectors in the search space as well as provide guidance and direction.This work details and discusses the RHC algorithm, its extensions, and its applications. First, the RHC algorithm is discussed, identifying the components and the algorithmic process for classification and recognition. Furthermore, this work describes additional addenda to the RHC algorithm, improving its classification prowess. These include reducing the memory footprint of the algorithm by coupling the RHC algorithm with another popular linear classifier: Linear Discriminant Analysis (LDA). Other addenda are explored that expedite the time required to train the system as well as increase the recognition rates of unlabeled data.Finally, this work discusses two applications and extensions of the RHC algorithm. First, a system, which utilizes the RHC algorithm, is developed by observing an actor's responses to various stimuli. This systems then imprints the learned behaviors onto a robot, and the robot is able to exhibit the learned, autonomous behavior in the controlled environments. In the second application, the RHC algorithm is extended, coupling it with a temporal queue that allows the RHC algorithm to classify and recognize dynamic gestures, which are spatiotemporal sequences. This second extension has been termed Spatiotemporal Recursive Hyperspheric Classification.