New Algorithm significantly cuts back time needed for machines to learn and practice new concepts
A one of a kind algorithm has been developed that significantly reduces the times taken by computers to recognize simple concepts. Researchers from New York University and Massachusetts Institute of Technology are teaching computers an algorithm that is similar to learning techniques used by humans.
The step is significant in reducing the amount of time required for machines to learn and practice new concepts. Ruslan Salakhutdinov, an assistant professor of Computer Science at the University of Toronto said that it was very difficult to develop machines that require very little data as required by humans when grasping a new concept.
Replicating the concept is considered to be an interesting area of research connecting machine learning, statistics, computer vision and cognitive science. Therefore, the study researchers have developed a probability-based algorithm called a ‘Bayesian Program Learning’ framework.
The research paper published in the journal Science unveiled that the framework was self-programming. The algorithm came up with new code leading to a new output of the computer program that researchers wanted their software to learn in every variation.
The system is similar to the way humans learn. In experiments, it was found that the software was able to identify and draw a handwritten character after seeing just one example. The researchers then tested the accuracy by coming up with a database of the world’s written languages.
The researchers asked judges to compare the computer’s drawings to those made by humans. The researchers were majorly unable to distinguish between human drawings and computer ones.
Joshua Tenenbaum, a professor at MIT, said that it is the first time that they have a machine that has the ability to learn and use a large-class of real-world concepts. “Our results show that by reverse engineering how people think about a problem, we can develop better algorithms. Moreover, this work points to promising methods to narrow the gap for other machine learning tasks”, affirmed Brenden Lake from New York University.
