News Release

Scientists create a new ultrafast, low-power machine learning algorithm for big data processing

The new solution dramatically improves one of the reference methods for data classification using Machine Learning techniques.

Peer-Reviewed Publication


Ultra-fast, low-power solution

image: Scientists from CiTIUS Research Center, in Spain have created a new algorithm which improves Support Vector Machines (SVMs), a reference method for classifying data sets with an accuracy that is virtually identical to that of humans. view more 

Credit: Andrés Ruiz / CiTIUS

Barely a few people are still unaware that Artificial Intelligence (and more specifically, Machine Learning), are living their moment of maximum splendor. The research developed in this field in recent years, together with the enormous computing power achieved by computers and the huge amount of data available to train algorithms, has revolutionized our lives and all sectors of the economy.

None of this would have been possible without the curiosity and efforts of the scientific community, which over the last few decades has contributed to the progressive sophistication of machine learning techniques, an increasingly precise area of research that allows machines to be trained to solve a wide variety of problems having to specifically program them for every purpose.

There are many strategies proposed to achieve this; one of the best known, and most powerful today are the so-called 'Support Vector Machines' (SVM), created by scientists Isabelle Guyon, Bernhard Schölkopf and Vladimir Vapnik in the 1990s. A major contribution recognized, among other awards, with the BBVA Foundation's Frontiers of Knowledge in 2020.

Essentially, Support Vector Machines (SVMs) are a method for classifying data sets, with an accuracy that is virtually identical to that of humans, and in some cases, even better. SVMs are one of today's best performing classifiers, which have amply demonstrated their effectiveness in recognizing data of diverse nature: from texts, voices and people’s faces to cancer cells or fraudulent uses of a credit card.

However, there is no infallible or best method in all cases, or for all circumstances. Thus, SVMs have proven to be considerably slow when dealing with problems where the number of data is very large, which is particularly problematic when working in Big Data environments; on the other hand, their memory consumption is sometimes unacceptable and can invalidate, in practice, this type of solution.

Now, a research carried out at CiTIUS (Singular Research Center on Intelligent Technologies of the University of Santiago de Compostela, Spain), has overcome these limitations with the development of a new 'Fast Support Vector Classifier'(FSVC), which has many advantages over the standard method. First, it is much faster - between 10 and 100 times - than traditional approaches. In addition, this new classifier operates with much less memory, "thanks to which it is able to develop optimal solutions with much less powerful and expensive computers," explains Manuel Fernández Delgado, director of the research.

CiTIUS researchers highlight this issue as one of the essential contributions of the work: "memory saving is of great importance," says Ziad Akram, predoctoral researcher at CiTIUS and first author of the paper, "since by improving efficiency we can solve, with much more modest equipment, problems for which we would normally need a supercomputer. "All this translates into a huge reduction in cost and energy consumption," says his colleague Eva Cernadas, principal investigator at the center and co-author of the paper.

Another of the architects of the work, Senén Barro, points out that "one of the keys is to have managed to develop an analytical solution for the design of classifiers, which avoids using iterative learning methods on data sets, the main cause of computational inefficiency and resource consumption of Machine Learning". The CiTIUS scientific director clarifies that "with this new approach, it is as if we could memorize a huge set of cases at once (think of faces, for example), without the need of having to see them over and over again until they are recorded in our memory". "The speed and savings in memory and computational capacity is enormous, which means savings in money and, more importantly, in carbon footprint," concludes Barro.


Z. A. Ali Hammouri, M. F. Delgado, E. Cernadas and S. Barro, "Fast SVC for large-scale classification problems". IEEE Transactions on Pattern Analysis and Machine Intelligence, doi: 10.1109/TPAMI.2021.3085969.

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.