蜜桃传媒

By Peter Murphy

The NSF Faculty Early Career Development (CAREER) program offers support to early-career faculty with potential to serve as academic role models in research and education, and those who lead advances in the mission of their organizations. The recipients in UB SEAS are Ian Bradley, assistant professor in the Department of Civil, Structural and Environmental Engineering, and Kaiyi Ji, assistant professor, Department of Computer Science and Engineering. Their CAREER awards total over $1.1 million. 

Harmful algal blooms (HABs) are responsible for countless public health concerns, and according to the National Centers for Coastal Ocean Science, the estimated economic impact of HABs in the U.S. is between $10–$100 million.

Not all algal blooms are hazardous, but HABS can produce toxins and cyanobacteria, a bacteria known as blue-green algae that could cause skin irritation, nausea, and liver or neurological damage. Even harmless algal blooms can cause eutrophication in water, a process where algae grow and then die after depleting the water’s oxygen. Bacteria feed on the dead algae, causing fish to die off and dead zones in the water.

Bradley, who is also a core faculty member in UB’s RENEW Institute, will use his $576,433 grant to investigate how wastewater treatment and agricultural discharge interact with HABs.

“We’re concerned about organic nitrogen and phosphorus that aren’t easily removed in wastewater treatment,” Bradley said, noting both are nutrients for algal growth. “What I want to understand is both how we can improve removing those in wastewater, and then also how they interact with harmful algal blooms out in the environment.”

Much of Bradley’s research focuses on algae and wastewater treatment. His PhD was in algal wastewater treatment, and his research group investigates algae cultivation and polyculture farming for biomass harvesting. The work in this CAREER award is the next step in his research on algae.

“Algae are great for wastewater treatment because they take up nitrogen and phosphorus very efficiently,” Bradley said. “Algae can take up even small amounts of nitrogen and phosphorus, that’s why we get algal blooms out in the environment. The idea behind all our previous work has been to take these blooms out of the environment and get them into an engineered system, and use algae to treat waste so that we have very low levels of nitrogen and phosphorus coming out of wastewater.”

This project continues those efforts and features a community outreach component as well. Bradley will work in wastewater treatment plants throughout Erie County, N.Y. He will also work with community and environmental groups in Western New York, including the Buffalo Niagara Waterkeeper. Bradley and his research group will connect their work to Buffalo Niagara Waterkeeper’s work measuring HABs and educating the community. 

AI, particularly large language models (LLM), like ChatGPT and big data applications—think 5G networks, and health care and finance models—must often fulfil multiple completing objectives at once. Multi-objective optimization (MOO) provides a framework for these models to identify the best trade-offs among competing objectives.

Ji’s $549,999 CAREER award aims to advance MOO theory, enhancing many systems that society interacts with daily. Ji is collaborating with UB faculty to implement the new theoretical framework into LLMs and robotics and is working with Amazon on ad recommendations using this theory.

“The research outcomes could have some impact on large foundation models, robotics and recommendation. In recommendation, users often have multiple requirements or requests,” Ji says. “For example, the user could request items that are cheap, high-quality and latest. These objectives are often conflicting, making the final decision difficult. My research wishes to strike a balanced solution that can satisfy users’ requirements as much as possible, instead of finding some items that focus only on one property like low price.”

MOO has been studied for decades, but in the era of ubiquitous large foundational models, it faces new challenges in scalability, stability and accuracy. Ji’s work aims to address these challenges and make MOO more effective in modern, practical AI-related applications. The broad impact his work could have on larger fields of artificial intelligence is significant, specifically with open AI platforms where users make multiple requests.

“How to find a balanced solution is important. Currently, multi-objective reinforcement learning from human feedback—reward soup—are very popular for multi-objective LLM modeling,” Ji says. “All of these tools involve dealing with multiple objectives at the same time. I believe my research will be useful there as well.”

Research efforts on this project are broken out into three complimentary thrusts where Ji and his research team will develop new theoretical and algorithmic foundations for stochastic MOO—optimizing multiple objectives when the solution is difficult to measure, propose efficient and scalable multi-objective bilevel optimization—a flexible and adaptive systems that works with MOO, and substantially advance MOO frameworks by incorporating innovative fairness concepts that are different than current approaches.

The project includes outreach activities to communicate research outcomes in educational and extracurricular settings to K-12, undergraduate and graduate students. Ji will work to develop experiential learning and undergraduate research opportunities and a course related to this work.