Breakout 2 Room 1 Summary

Edgar Lobaton (NCSU), Anuj Karpatne (VT)

On behalf of

AI2ASED Breakout 2 Room 1

Breakout room 1 participants

Moderator: Edgar Lobaton at NCSU

Scribe: Anuj Karpatne (VT)

Attendees:

• Animadhree Anandkumar (Caltech)
• Shuiwang Ji (Texas A&M)
• Mark Neubauer (UIUC)
• Xinghua Mindy Shi (Temple)
• Jing Gao (Delaware)
• Daniel Anglés-Alcázar (UConn)
• Peetak Mitra (Excarta)
• Carl Kingsford, CMU
• Madhav Marathe (UVA)
• Rajagopalan Balaji (Colorado)
• Mingyi Hong (UMN)
• Michael Littman (NSF IIS)
• Eric DeWeaver (NSF GEO/AGS)
• Jim Shank (NSF MPS/PHY)

Q1: What areas of science and engineering have been transformed by AI?

• NLP and Multimodal Data Fusion … Joint representations
• System side of Data Management
• Biomedicine (e.g., drug discovery)
• Imaging (e.g., from microscopic to astrophysics)
• Astronomy / Cosmology … Allowing to process large amounts of data

Q4: What are examples of research that could be used as concrete exemplars of the kind of work we’d like to see more of?

• Mining literature in biomedical sciences
• Extracting knowledge graphs
• There is more structure information that can be mined… It has not happened there yet.
• Need to see more applications

Q2: What are the key gaps in current scientific methods that can be filled by AI?

• We need to break the barrier of entry for collaboration so people can contribute to areas without having to get a Ph.D. in these areas.
• Our understanding of the physical phenomena is limited
• Physics models can be biased
• ML is very useful alternative and tools such as transfer learning make it more accessible but we still have a way to go
• ML models could be used as proxies for simulations

​

Q3: What are the priority research directions of data-driven discovery in science and engineering that data analytics, AI, and ML can enable or accelerate?

• Besides the topics in Q3, AI models need to be more robust and explainable. They also need to go beyond well structured datasets.

Q5: How can education and community outreach about data science, AI and ML speed up this process?

• Starting early with educating our broader scientific community and students on data science
• Educate people on using AI as a partner… Working “with” AI
• There is a need for a roadmap / national AI infrastructure for a birthright for students (and broader scientific community include low-resource institutions) to have access to computing resources
• We need a more structure set of resources that are supported and sustained by the community… perhaps even giving back from industry based on what they are gaining from the open source community.