Each year, 55 to 60 junior researchers earn the Computer and Information Science and Engineering Research Initiation Initiative (CRII) award from the National Science Foundation (NSF). In the past two years, three principal investigators from USC’s Information Sciences Institute (ISI) have received this award. While the more widely known NSF CAREER award applies to tenure-track faculty, the CRII program distinctly enables early-career research faculty to launch their research careers.
Filip Ilievski, a Research Scientist at the Center on Knowledge Graphs at ISI, was granted the CRII award on March 15th of this year. In 2021, Muhao Chen of the Artificial Intelligence Division and Loïc Pottier of the Computational Systems and Technology Division both scored one. This program seeks to provide essential resources and funding to enable promising, early-career investigators in launching their research careers, with sufficient funds for four years of graduate student support.
In laying the groundwork for resourcing their research goals, the NSF CRII award allows leading ISI researchers Ilievski, Chen, and Pottier to hire PhD students and grow their team to pursue their own research agendas. For them, it’s integrating common sense knowledge into AI technologies, automatically acquiring structured data from unstructured text, and uniting supercomputing and distributed systems to optimize machine learning algorithms, respectively.
Yolanda Gil, ISI’s Director for Major Strategic AI and Data Science Initiatives and a research professor in computer science, discussed the importance of the NSF program. “The CAREER and CRII programs have been the primary vehicles to nurture the research and education efforts of early-career investigators,” said Gil. “Given the incredible talent of young ISI hires who are in full-time research positions, the CRII program is a great fit.”
Gil has also been a reviewer in panels that evaluate proposals for the NSF CRII program. “I have been deeply impressed with the quality of the proposals submitted and the very competitive selection process,” said Gil.
Ewa Deelman, a research professor of computer science at ISI, has mentored Pottier throughout his research endeavors. “I am very happy for Loïc to have received the CRII award, this is a very good first step for him in developing his own research program centered around his HPC [high performance computing] interests.”
Pedro Szekely, Director of ISI’s Artificial Intelligence Division and a research associate professor in computer science, has guided Ilievski and Chen in various degrees throughout the proposal process. “The work that they do is not incremental, it’s innovative,” said Szekely. Successful applicants demonstrate that their idea is novel all while showcasing their plans for success in the submitted proposal.
Sometimes, the first time’s the charm. That was the case with Filip Ilievski, a research scientist at ISI. After submitting his first-ever proposal, he has become ISI’s latest awardee of the NSF CRII program, leading the field in creating AI agents that provide common sense explanations about open-world narratives.
Inspired by AI’s deficits in using common sense knowledge when they interact with people, Ilievski designed his research proposal with the eventual goal of improving AI assistants for elderly individuals with dementia and children with autism spectrum disorder (ASD).
His research proposal highlights the importance of integrating common sense into AI technologies.
Ilievski referred to the concept of ‘substitution’, an integral facet of common-sense AI technology. “When stories are dissimilar, but they rely on the same axiom, it’s called substitution.” For example, if a child is cold but there are no jackets available, the AI technology can assist them in understanding that another available object, such as a blanket, can serve the same function. Ilievski pointed to this example as illustrative of how this technology can be potentially used as assistance for ASD children.
Based on knowledge statements that experts have written, the machine is then able to see and learn from such statements in order to fill the gaps for future similar situations. “If you’re able to teach a computer a certain kind of story, it would then be able to understand different stories,” said Ilievski. “And once it does that, it could help human users do the same over time.”
For Ilievski, this project is just the beginning. He hopes to continue expanding his work in common sense technology to reach the ultimate goal of providing meaningful social assistance to vulnerable populations and in improving communication skills of AI systems.
Muhao Chen, a research lead at ISI and a research assistant professor in computer science, is using the award to supplement his study on automatically acquiring structured information from unstructured text. In AI terms, this structure is captured in knowledge graphs (KG), which provide both open-world and domain-specific knowledge representations that are integral to many AI systems. He is also looking at how to make structured knowledge transferable across different languages and scientific domains, such as biology, genomics, and proteomics.
“My long-term goal is to help the machine understand nature. If it understands nature, it understands what we say, what we know about the world, and how we communicate with machines,” said Chen.
The study proposes a framework that seeks to systematically improve the robustness of learning and inference for data-driven knowledge acquisition models.
And its real-world applications are vast, like helping physicians understand disease targets. “We are also applying this kind of technology to disease prediction,” said Chen. “It’s a way to help physicians observe patients.” In making predictions, it promotes the accuracy of designing drugs to target diseases, an important medical research area.
Chen also plans to expand automatic knowledge acquisition for software-related online forums, with the ultimate goal of lowering cognitive effort of software developers.
Loïc Pottier, a research scientist at ISI, was granted the NSF CRII award for his research proposal on advancing supercomputing and cloud computing to scale up machine learning.
But supercomputing is not a concept most are familiar with, so when Pottier explained it, he devised an analogy to help visualize its abilities. “Think of your laptop, and multiply your laptop by a hundred thousand,” said Pottier. “You interconnect all of these laptops [to process data together] and you have a supercomputer.” Supercomputers are found largely in governmental facilities like the U.S. Department of Energy in order to run scientific computations, such as climate change simulations or weather forecasting.
Pottier intends to combine supercomputing and machine learning to optimize algorithms. Machine learning (ML) algorithms have become key elements in many scientific domains over the last few years, and they exist virtually everywhere. When datasets become larger, machine learning requires extensive computational capabilities, and that’s where Pottier’s research comes in.
“I’m taking one machine learning algorithm, making it run faster on high-performance computing machines, and expending less energy and fewer resources,” said Pottier.
“If I can help them accelerate machine learning algorithms, it will help them run bigger and more accurate simulations,” continued Pottier. By improving and optimizing the base layer of data processing, major software benefits emerge that can accelerate research, like our understanding of COVID-19 simulations or climate modeling.
Published on April 14th, 2022
Last updated on November 17th, 2022