This March, amid the sharp rising number of COVID-19 cases, a call to action to the artificial intelligence (AI) community was issued by leading research institutions alongside federal policy officials. It advocates for new AI-driven methods to be developed, in order to address urgent scientific questions and curb the spread of coronavirus.
At the Lawrence Berkeley National Laboratory (Berkeley Lab), a group of material researchers answered the call. They came up with a solution in a week — a search engine called COVID Scholar. Powered by a supercomputer, COVID Scholar uses its natural language processing algorithms to scan coronavirus-related literature and machine learning ability to make connections that aren't obvious from a human perspective.
Within a month of its operation, the AI-driven site has collected over sixty-one thousand higly relevant research reports, with 8,000 of them specifically targeting COVID-19 and the rest on other viral species and pandemics in general.
Natural language processing (NLP) is the cross-section of seemingly unrelated fields, including linguistics, computer science, and information engineering. It's a popular subject in artificial intelligence because well programed NLP tools can mine and investigate useful information from large piles of natural language-based data. A commonly seen application are the chatbots that websites use to provide customer service without any involvement of real human employees.
Machine learning (ML), the second utility of COVID Scholar, is the use of computer programs to extract useful experience from well studies data, and make prediction of unencountered scenarios. ML can help email servers to filter out junk mails and scams, video streaming services to predict user preferences, and radiologists to improve the reading of diagnostic scans .
What's extra intriguing, the researchers behind this spontaneous innovation have zero background in medical science. Led by Gerbrand Ceder, a Standard University material scientist, the main focus of the group is to design, develop, and test novel materials for energy storage using computational approaches. The supercomputer used in COVID Scholar is hosted by the National Energy Research Scientific Computing Center (NERSC), a Department of Energy (DOE) facility located inside the Berkeley Lab.
The material research group previously built MatScholar a similar search engine that mines knowledge for material research. According to a published report from last year, their algorithm was able to predict the discoveries of new thermoelectric materials and identify previously unknown material candidates, by combing and analyzing the abstracts of 3.3 million research papers in the relevent field.
By joining the fight against COVID-19, the researchers also advanced their own knowledge in mining scientific texts. Commented on their experience in the COVID Scholar project at a press interview, Ceder said: "This is a test case of whether an algorithm can be better and faster at information assimilation than just all of us reading a bunch of papers."
Russ Altman: Artificial intelligence takes on COVID-19 (Standford)
Source: SciTechDaily