Earth artificial intelligence (AI) has become a research frontier by leveraging AI techniques to understand the complex Earth system and help various Earth applications. Challenges for Earth AI include a large volume of available data, spatial-temporal high-dimensionality, incompatible data from multiple sources, data-driven causal understanding of the Earth system. This talk will present two related Earth AI studies. The first study proposes a Time-Series Causal Neural Network (TS-CausalNN) - a deep learning technique to discover contemporaneous and lagged causal relations simultaneously from non-stationary and non-linear Earth observation time series data. The second one studies how to leverage deep domain adaptation techniques and multiple satellite data to improve cloud remote sensing retrieval. Both studies use real-world Earth data to evaluate their advantages over state-of-art approaches.

Dr. Jianwu Wang is an Associate Professor in UMBC's Department of Information Systems. He leads the Big Data Analytics Lab (BDAL) and co-leads the NSF HDR Institute for Harnessing Data and Model Revolution in the Polar Regions (iHARP). He is also an affiliate faculty in CSEE and the Joint Center for Earth Systems Technology (JCET). 

The learned behaviors of AI and robot agents should align with the intentions of their human designers. Toward this goal, people must be able to easily specify, inspect, and model agent behaviors. For specifications, we will consider expert-written reward functions for reinforcement learning (RL) and non-expert preferences for reinforcement learning from human feedback (RLHF). I will show evidence that experts are bad at writing reward functions: even in a trivial setting, experts write specifications that are overfit to a particular RL algorithm, and they often write erroneous specifications for agents that fail to encode their true intent. Next, I will show that the common approach to learning a reward function from non-experts in RLHF uses an inductive bias that fails to encode how humans express preferences, and that our proposed bias better encodes human preferences both theoretically and empirically. For inspection, humans must be able to assess the behaviors an agent learns from a given specification. I will discuss a method to find settings that exhibit particular behaviors, like out-of-distribution failures. Lastly, cognitive science theories attempt to show how people build conceptual models that explain agent behaviors. I will show evidence that some of these theories are used in research to support humans, but that we can still build better curricula for modeling. Collectively, my research provides evidence that—even with the best of intentions— current human-AI systems often fail to induce alignment; my research proposes promising directions for how to build better aligned human-AI systems.

Serena Booth received her PhD at MIT CSAIL in 2023. Serena studies how people write specifications for AI systems and how people assess whether AI systems are successful in learning from specifications. While at MIT, Serena served as an inaugural Social and Ethical Responsible Computing Scholar, teaching AI Ethics and developing MIT’s AI ethics curriculum that is also released on MIT OpenCourseWare. Serena is a graduate of Harvard College (2016), after which she worked as an Associate Product Manager at Google to help scale Google’s ARCore augmented reality product to 100 million devices. Serena currently works in the U.S. Senate as a AAAS AI Policy Fellow, where she is working on AI policy questions for the Senate Banking, Housing, and Urban Affairs Committee. Her research has been supported by an MIT Presidential Fellowship and by an NSF GRFP. She is a Rising Star in EECS and an HRI Pioneer.

In an era where Generative Artificial Intelligence (GAI) is increasingly becoming a part of our daily professional and personal lives, understanding how to effectively interact with these systems is crucial. This workshop is designed to equip participants with practical skills and insights for engaging with GAI interfaces, specifically focusing on optimizing the quality of results, integrating GAI outputs into user responses, and adhering to ethical citation practices.  This workshop is being hosted by the Graduate Program in Geographic Information Systems. Friends of the program are welcome to join virtually.

Join UMBC Information Systems PhD student Rachael Kang as she shares her research journey of how she came to study AI, starting from her master's thesis on the utility of machine learning in predicting suicide risk to her current research interests of integrating a large language model into patient portals to increase patient health literacy. Rachael will also discuss how she has introduced ChatGPT as a learning tool for the students she TAs in Information Systems 303, the activities she has conducted with the students to demonstrate the strengths and weaknesses of ChatGPT, and the types of conversations she has with students about what is proper and improper use of ChatGPT.

Get more information and join the event here.

Here is a video of the session.

This week’s AI Lunchbox Series, held online 12-1 pm ET on Thurs. April 25, is on AI integrations in Full Circle Data Analysis: Integrating ChatGPT and Google Colab with Python for Business Insights. 

Dhruv Ratra shows how ChatGPT can generate Python code tailored to your data analysis needs, enabling you to interpret, cleanse, model, and validate your data with precision. You will see how the Python code executed in Google Colab produces results that are then analyzed and enhanced by ChatGPT, creating a feedback loop for continuous improvement and deeper insights.

Register for the free Lunchbox Series on the C4A site or through Meetup.

Swarm intelligence (SI) was first introduced and used in the context of robotic cellular systems and has become a subfield of artificial intelligence. The collective behavior and self-organization skills of different animal populations are used in the design of different SI techniques to solve computational problems, especially in the field of optimization. SI techniques, like Particle Swarm optimization or Ant Colony optimization algorithms are often used together with the Gradient Descent, the Gradient Conjugate and some Quasi-Newton methods to improve the effectiveness of high-dimensional optimization problems. In my talk, I would like to introduce a hybrid model that demonstrates how to avoid local minimum traps in a global, non-convex problem.