Grace W. Rumantir

A Modified K-means Algorithm for Noise Reduction in Optical Motion Capture Data

This paper presents a modified K-means algorithm that can be used for removing noise in multicolor motion capture image sequences. These images have been produced using the illuminated line segment based marker system. The proposed algorithm takes into account the nature of the motion capture images in terms of the number of data pixels normally clustered together and the acceptable degree of compactness of a data cluster. The cleaned data can be used for accurate and effective tracking of the captured motion.

A Concept for Optimizing Behavioural Effectiveness & Efficiency

Both humans and machines exhibit strengths and weaknesses that can be enhanced by merging the two entities. This research aims to provide a broader understanding of how closer interactions between these two entities can facilitate more optimal goal-directed performance through the use of artificial extensions of the human body. Such extensions may assist us in adapting to and manipulating our environments in a more effective way than any system known today. To demonstrate this concept, we have developed a simulation where a semi interactive virtual spider can be navigated through an environment consisting of several obstacles and a virtual predator capable of killing the spider. The virtual spider can be navigated through the use of three different control systems that can be used to assist in optimising overall goal directed performance. The first two control systems use, an onscreen button interface and a touch sensor, respectively to facilitate human navigation of the spider. The third control system is an autonomous navigation system through the use of machine intelligence embedded in the spider. This system enables the spider to navigate and react to changes in its local environment. The results of this study indicate that machines should be allowed to override human control in order to maximise the benefits of collaboration between man and machine. This research further indicates that the development of strong machine intelligence, sensor systems that engage all human senses, extra sensory input systems, physical remote manipulators, multiple intelligent extensions of the human body, as well as a tighter symbiosis between man and machine, can support an upgrade of the human form.

Minimum Message Length Criterion for Second-Order Polynomial Model Selection Applied to Tropical Cyclone Intensity Forecasting

This paper outlines a body of work that tries to merge polynomial model selection research and tropical cyclone forecasting research. The contributions of the work are four-fold. First, a new criterion based on the Minimum Message Length principle specifically formulated for the task of polynomial model selection up to the second order is presented. Second, a programmed optimisation search algorithm for second-order polynomial models that can be used in conjunction with any model selection criterion is developed. Third, critical examinations of the differences in performance of the various criteria when applied to artificial vis-a-vis to real tropical cyclone data are conducted. Fourth, a novel strategy which uses a synergy between the new criterion built based on the Minimum Message Length principle and other model selection criteria namely, Minimum Description Length, Corrected Akaike’s Information Criterion, Structured Risk Minimization and Stochastic Complexity is proposed. The forecasting model developed using this new automated strategy has better performance than the benchmark models SHIFOR (Statistical HurrIcane FORcasting) [4] and SHIFOR94 [8] which are being used in operation in the Atlantic basin.

Sampling of Highly Correlated Data for Polynomial Regression and Model Discovery

The usual way of conducting empirical comparisons among competing polynomial model selection criteria is by generating artificial data from created true models with specified link weights. The robustness of each model selection criterion is then judged by its ability to recover the true model from its sample data sets with varying sizes and degrees of noise.If we have a set of multivariate real data and have empirically found a polynomial regression model that is so far seen as the right model represented by the data, we would like to be able to replicate the multivariate data artificially to enable us to run multiple experiments to achieve two objectives. First, to see if the model selection criteria can recover the model that is seen to be the right model. Second, to find out the minimum sample size required to recover the right model.This paper proposes a methodology to replicate real multivariate data using its covariance matrix and a polynomial regression model seen as the right model represented by the data. The sample data sets generated are then used for model discovery experiments.

Noise Filtering of New Motion Capture Markers Using Modified K-Means

In this report a detailed description of a new set of multicolor Illuminated Contour-Based Markers, to be used for optical motion capture and a modified K-means algorithm, that can be used for filtering out noise in motion capture data are presented. The new markers provide solutions to central problems with current standard spherical flashing LED based markers. The modified K-means algorithm that can be used for removing noise in optical motion capture data, is guided by constraints on the compactness and number of data points per cluster. Experiments on the presented algorithm and findings in literature indicate that this noise removing algorithm outperforms standard filtering algorithms such as Mean and Median because it is capable of completely removing noise with both Spike and Gaussian characteristics. The cleaned motion data can be used for accurate reconstruction of captured movements, which in turn can be compared to ideal models such that ways of improving physical performance can be identified.

Minimum Message Length Criterion for Second-Order Polynomial Model Disovery

This paper proposes a method based on the Minimum Message Length (MML) Principle for the task of discovering polynomial models up to the second order. The method is compared with a number of other selection criteria in the ability to, in an automated manner, discover a model given the generated data. Of particular interest is the ability of the methods to discover (1) second-order independent variables, (2) independent variables with weak causal relationships with the target variable given a small sample size, and (3) independent variables with weak links to the target variable but strong links from other variables which are not directly linked with the target variable. A common nonbacktracking search strategy has been developed and is used with all of the model selection criteria.

Clustering Experiments on Big Transaction Data for Market Segmentation

This paper addresses the Volume dimension of Big Data. It presents a preliminary work on finding segments of retailers from a large amount of Electronic Funds Transfer at Point Of Sale (EFTPOS) transaction data. To the best of our knowledge, this is the first time a work on Big EFTPOS Data problem has been reported. A data reduction technique using the RFM (Recency, Frequency, Monetary) analysis as applied to a large data set is presented. Ways to optimise clustering techniques used to segment the big data set through data partitioning and parallelization are explained. Preliminary analysis on the segments of the retailers output from the clustering experiments demonstrates that further drilling down into the retailer segments to find more insights into their business behaviours is warranted.

A Two Tiered Finite Mixture Modelling Framework to Cluster Customers on EFTPOS Network

This paper proposes a framework to build a clustering model of customers of the retailers on the EFTPOS network of a major bank in Australia. The framework consists of two clustering tiers using Finite Mixture Modelling (FMM) that segments customers based on their probabilities of generating transactions of different categories. The first tier generates the transaction categories and the second tier segments the customers, each with a vector of the fractions of their transaction categories as parameters. For each tier, we determine the optimal number of clusters based on the Minimum Message Length (MML) criterion. With the premise that the most valuable customer segment is one that is most likely to generate the most valuable transaction category, we rank the customer segments based on their respective joint probabilities with the most valuable transaction category. By doing so, we are able to reveal the relative value of each customer segment.