As the Covid-19 pandemic shows, we live in a richly connected world, promoting information and influence as well as efficient spread of the virus. What can we learn by analyzing these connections? This is a central question in network science, a field of study that models interactions between physical, biological, social, and information systems to solve problems.
The 2021 GraphEx Symposium, hosted by the MIT Lincoln Laboratory, brought together top network science researchers to share the latest advances and applications in this area.
“We will investigate and identify how the use of graph data can provide key technology enablers to solve the most pressing problems facing our country today,” said the Lincoln Laboratory AI. Edward Kao, the symposium organizer and technical staff of the Software Architecture and Algorithms Group, said.
Virtual event themes address some of the most relevant issues of the year, including disinformation analysis on social media, modeling of pandemic epidemics, and speeding up drug design using graph-based machine learning models. It was developed in the center.
“A special session on Impact Manipulation and Covid-19 in GraphEx reflects the relevance of network and graph-based analysis to understand the phenomenology of these complex and influential aspects of modern life. We’ll take a closer look at working with graphs, “said William Streilein, who co-chaired with Rajmonda Caceres at the Lincoln Institute.
Several presentations at the symposium focused on the role of network science in the analysis of impact manipulation (IO), or a systematic endeavor by national and / or non-state actors to disseminate the story of disinformation.
Researchers at the Lincoln Laboratory have developed tools to classify and quantify the impact of social media accounts, which are likely to be IO accounts, to deliberately spread false Covid-19 treatments to vulnerable populations. ..
“A cluster of IO accounts acts as an echo chamber to amplify the stories, and then vulnerable people are working on these stories,” he says, developing a tool called RIO or Reconnaissance of Impact Operations. Researcher Erika Mackin said.
To classify IO accounts, Mackin and her team trained algorithms to detect estimated IO accounts in the Twitter network based on specific hashtags or narratives. One example they studied was #MacronLeaks, a disinformation campaign targeting Emmanuel Macron during the 2017 French presidential election. The algorithm now labels accounts in this network as IO based on several factors, including the number of interactions with foreign news accounts, the number of links tweeted, and the number of languages used. I’m being trained. Their model then uses a statistical approach to score the level of influence the account has in spreading the story within its network.
The team found that classifiers were superior to existing detectors for IO accounts because they could identify both bot accounts and human-operated accounts. They also found that the IO accounts that propelled the disinformation story of the 2017 French presidential election largely overlapped with the accounts that influentially disseminate Covid-19 pandemic disinformation today. Did. “This suggests that these accounts will continue to move into the disinformation story,” says McKin.
Throughout the Covid-19 pandemic, leaders have made sound decisions in search of epidemiological models that predict how the disease will spread. Alessandro Vespignani, director of the Network Science Institute at Northeastern University, is leading the Covid-19 modeling effort in the United States and gave a keynote speech on this work at the symposium.
In addition to taking into account the biological facts of the disease, such as the incubation period, Vespignani’s model is particularly powerful in that it includes community behavior. To perform a realistic simulation of the disease epidemic, he develops a “synthetic population” built using a highly published, highly detailed dataset of US households. “We create a non-genuine but statistically real population and generate a map of the interactions of those individuals,” he says. This information is fed back into the model to predict the spread of the disease.
Today, Vespinani is looking at ways to integrate viral genomic analysis into this type of population modeling to understand how mutants are spreading. “It’s still a very interesting ongoing task,” he added, adding that this approach has helped to model the variance of the delta variant of SARS-CoV-2.
As researchers model virus spread, Lucas Laird of the Lincoln Laboratory is exploring ways to use network science to design effective control strategies. He and his team are developing models for customizing strategies in different geographic regions. This effort was facilitated by the discovery by researchers of the Covid-19 differences that permeate the entire US community and the gaps in intervention modeling to address those differences.
As an example, they applied the planning algorithm to three counties: Florida, Massachusetts, and California. Considering the characteristics of a particular geographic center, such as the number of susceptible individuals and the number of infections, their planners develop different strategies in their communities throughout the outbreak.
“Our approach eradicates the disease in 100 days, but it can be done with far more targeted interventions than any of the global interventions. In other words, the entire country needs to be closed. No. ”Laird adds that it provides a“ sandbox environment ”for planners to consider future intervention strategies.
Machine learning with graphs
Graph-based machine learning is gaining more and more attention as it has the potential to “learn” complex relationships between graphic data and extract new insights and predictions about these relationships. This interest has spawned a new class of algorithms called graph neural networks. Today, graph neural networks are being applied in areas such as drug discovery and material design with promising results.
“Now we have a much wider range of applications for deep learning, not just medical imaging and biological sequences. This opens up new opportunities for data-rich biology and medicine,” GraphEx said. Marinka Zitnik, an assistant professor at Harvard University, who presented her work at.
Zitnik’s research focuses on a rich network of interactions between proteins, drugs, diseases, and patients on a scale of billions of interactions. One use of this study is to discover drugs, such as Covid-19, to treat illnesses with little or no approved drug treatment. In April, Zitnik’s team used a graph neural network to rank 6,340 drugs for their expected efficacy against SARS-CoV-2 and identified four that could be reused to treat Covid-19. Published a paper on.
At the Lincoln Laboratory, researchers are also applying graph neural networks to the challenge of designing advanced materials that can withstand extreme radiation and capture carbon dioxide. Similar to the process of designing a drug, a trial-and-error approach to material design is time consuming and costly. Our lab team is developing a graph neural network that can learn the relationship between the crystal structure of a material and its properties. This network can be used to predict different properties from new crystal structures and significantly speed up the process of screening materials with the properties required for a particular application.
“Graphic feature learning has emerged as a rich and prosperous field of study for incorporating induction bias and structured prior information into machine learning processes: drug design, accelerated scientific discovery, and personalization. There are a wide range of applications, such as recommender systems, “says Caceres.
A vibrant community
The Lincoln Laboratory has been hosting the GraphEx Symposium every year since 2010, with the exception of last year’s cancellation by Covid-19. “One of the key points is that the GraphEx community is more vibrant and active than ever, despite the postponement and virtualization needs from last year,” says Streilein. “Network-based analytics continues to expand in scope, applying and influencing more important areas of science, society and defense than ever before.”
In addition to members of the Lincoln Laboratory, the Technical Committee members and co-chairs of the GraphEx Symposium include researchers from Harvard University, Arizona State University, Stanford University, Smith University, Duke University, the US Department of Defense, and Sandia National Laboratory. Was included.
MIT Lincoln Laboratory Convenes Top Network Scientists at Graph Utilization Symposium – India Education | Latest Education News | Global Education News
Source link MIT Lincoln Laboratory Convenes Top Network Scientists at Graph Utilization Symposium – India Education | Latest Education News | Global Education News