|
Dr Bhekisipho Twala |
Recently, the use of artificial intelligence (AI) or machine learning algorithms has proven to be of great practical value in solving a variety of robotics problems including robot execution failure prediction.
CSIR research scientist, Dr Bhekisipho Twala, involved in digital intelligence studies, says. "An important field in applied mathematics is data, which use probability theory as a tool and allow the description, analysis, and prediction of phenomena where chance plays a role. Most techniques for predicting attributes of a robotic system require past data from which models will be constructed and validated.
One of the major problems of applying AI algorithms in robotics is the unavailability, scarcity and incompleteness of data, i.e. data for training the model. Most institutions do not share their data with other organisations and as such, a useful database with detailed data cannot be formed. In addition, surveys for collecting such data are sometimes small but difficult and expensive to conduct." Twala is determined to tackle this set-back.
He says, "My first goal is to improve the quality of the training data and in doing so, improve prediction accuracies produced by learning algorithms." An algorithm is a sequence of finite instructions where each variable affects the outcome.
Twala says, "In the South African context we do not have sound data that could be used for machine learning, which is also known as AI. As a result, the machine or robot will not be able to perform its required task. What I am doing is 'cleaning the data' to improve the quality and ultimately ensure that the application of AI for security and surveillance in South Africa is driven from informed data."
Numerical analysis investigates computational methods for efficiently solving a broad range of mathematical problems that are typically too large for human numerical capacity. "For example, a robot programmed to pick up possible threats. Data would have been initially fed into the system with various attributes having been determined as a possible threat; be it a physical object such as a bomb or a person wearing an overhead coat. If the data are not sound, the robot execution tasks will be limited," he explains. "We are presently using data collected internationally from CCTVs, videos and data from situations where the robots have failed in order to compile a set of informed data." Twala's current efforts in the quest for informed data will be modified at a later stage to the South African context.
Twala has over 10 years of experience in putting mathematics to scientific use in the form of data comparison, inference, analysis, and presentation to design, collect, and interpret data experiments in the fields of transport, medical, artificial intelligence, software engineering and most recently, robotics. He received his doctorate from the Open University in the UK in 2005 and did his postgraduate studies and MSc at the University of Southampton, also in the UK.
Enquiries: CSIR Communication
|
