Researchers develop a new algorithm to solve complex problems at low cost

Researchers at Johannes Gutenberg University Mainz (JGU) in Germany and Università Della Svizzera Italiana (USI) in Lugano in Switzerland have developed an algorithm that can solve complex problems like weather programmes with remarkable accuracy even on a personal computer. The low-cost, scalable probabilistic approximation (SPA) algorithm, is a breakthrough as the computational cost is one of the major challenging factors till now when dealing with bigger programmes.

In the past 60 years, there has been an escalated growth in computer processing power (Moore’s law). However, this is expected to have a halt in the 2020s. The newer developments are based on artificial intelligence and machine learning. However, the exact running of many machine learning methods such as deep learning work like a “black box” according to the researchers. Professor Susanne Gerber, a specialist in bioinformatics at Mainz University along with Professor Illia Horenko, a computer expert at Università Della Svizzera Italiana and a Mercator Fellow of Freie Universität Berlin wanted to know how artificial intelligence works and liked to gain a better understanding of the connections involved in it. Thus the team developed an algorithm to implement complex calculations with high accuracy while maintaining a low cost. 

In their paper titled “Low-cost scalable discretization, prediction, and feature selection for complex systems” recently published in Science Advances. “, the duo researchers along with their co-authors explain their concept. 

The process is based on the Lego principle. Instead of solving the complex problems that include- discretization, feature selection, and prediction problems- separately, the introduced computational procedure (SPA) solves them simultaneously. In this method, complex systems are broken down into discrete states or patterns. With fewer components, large volumes of data can be analyzed and their future behaviour can be predicted.  

“For example, using the SPA algorithm we could make a data-based forecast of surface temperatures in Europe for the day ahead and have a prediction error of only 0.75 degrees Celsius,” said Gerber. It all works on an ordinary PC and has an error rate that is 40 per cent better than the computer systems usually used by weather services, whilst also being much cheaper.

In addition to weather forecasts, the research sees numerous possible applications such as in solving classification problems in bioinformatics, image analysis, and medical diagnostics