10/12/2019

Unveiling materials through machine learning

Foto: Freepik

Reproduction from the IMPA Science & Mathematics blog, from O Globo, coordinated by Claudio Landim.

Adalberto Fazzio , National Nanotechnology Laboratory/CNPEM and researcher at INCT Nanocarbono

The technological advancements that developed intensely from the 20th century onwards culminated in the evolution of digital technologies. With the invention of the transistor and, consequently, of computers, the digital age began, and with it, the information revolution.

Since then, the quantity and speed of data generation has exploded, taking on diverse forms and across all areas, from entertainment and financial transactions to the social, biological, and natural sciences.

In the 1950s, the field then called "artificial intelligence" emerged, dealing with methods and algorithms created to give computers the ability to replicate human intelligence, even if in very specific tasks. Over time, these capabilities have evolved, until we have reached the present day, where artificial intelligence already surpasses human intelligence in many tasks, such as medical diagnoses or even winning championships in complex games like Poker or Go.

Due to these great advances and successes over time, various scientific fields have begun to invest resources in exploring these capabilities and converting raw data into information and knowledge.

In materials science, then-US President Barack Obama set a milestone in 2011 by launching the "Materials Genome" program. He emphasized that materials are a differentiating factor in a product, key to the economy and national security. This program aimed to create and utilize an experimental and computational infrastructure system for generating data related to optimizing the physical and chemical properties of materials.

Initiatives like these, combined with computational and methodological advances, have brought materials science closer to the world of "Big Data," where existing information can be studied by data science, which encompasses techniques such as data mining, sorting, and machine learning.

One of the best-known areas of data science and artificial intelligence is called "machine learning ," in which a fundamental component is the availability of datasets that provide the ability to "teach" machines. Among the most common capabilities to be taught are tasks such as classifying materials into different classes and predicting numerical values, for example, material properties.

Today, a significant portion of the physics, chemistry, and engineering communities invest in these methodologies to extract information and knowledge from accumulated data.

To read the full text, visit the newspaper's website.