As much of the world shut down in 2020, our digital reality continued to grow. So although the grand opening of the Hasso Plattner Institute (HPI) Research Center in Machine Learning and Data Science at UC Irvine was on hold for two and half years, work to advance artificial intelligence persisted.
This HPI at UCI partnership began in February 2020, with HPI — dedicated to pioneering research into information technology — opening its newest research school at UCI’s Donald Bren School of Information and Computer Sciences (ICS). “ICS is the only computing-focused school among California universities, which makes it an ideal place for this data science focused effort,” noted Erik Sudderth in his welcoming remarks at the long-delayed grand opening in November 2022. Sudderth, a professor of computer science and statistics in the School of ICS, serves as director of HPI at UCI. “The first students started in April 2020,” he continued, speaking of Ph.D. fellows at the center, “and if you think back to 2020, there was a little bit of bad luck there.”
Yet Christoph Meinel, director and CEO of HPI, Germany, had a different perspective. “[COVID-19] forced us to have a silent start,” he said during his welcoming remarks, adding that “it’s lucky that we had to postpone the opening to a day [when] the main persons, the Ph.D. students of this center, are among us.”
By the time UCI was able to host Meinel and other HPI researchers from Potsdam, Germany, including close to 25 HPI students, the event was more than a grand opening. It also was a symposium showcasing two years’ worth of innovative AI research.
HPI at UCI has already funded 16 students, with five starting in 2020, another five added in 2021, and six new fellows joining in 2022. Sudderth was pleased to report that, working with HPI faculty supervisors at UCI, this “very talented, productive group” has already produced more than 50 research papers. “In our virtual state, there was a lot of great work started.”
This work focuses on three main areas:
- online and adaptive AI systems that, with little or no human-annotated data, can adapt to real-world environments;
- robust and safe AI systems that are more “self-aware” when encountering situations beyond their expertise; and
- human-centered AI systems that effectively explain (fair) decisions, communicate their reasoning, and build trust with human users.
“The goal of HPI at UCI is to foster discovery in human-centered AI technologies, combining the strengths of HPI and ICS in the areas of machine learning, data analytics and information management systems,” said Marios Papefthymiou, dean of the School of ICS, at the grand opening. “The center is not simply aimed at better AI technologies. It aspires to advance transparent and explainable AI technologies that are designed around humans.”
Collaborative Visit & Symposium
Faculty and students in ICS have been involved in AI and machine learning research for decades, but the new HPI center is helping accelerate and advance that work by funding Ph.D. fellows and developing partnerships with similarly focused researchers in Germany. The in-person gathering aimed to help foster those partnerships. “We had some online meetings and interactions over zoom [during the pandemic], but there was no way for those to substitute for the interactions you get from an event like this,” explains Sudderth. “So it was great to have folks here.”
To help HPI faculty and students get to know their UCI counterparts, the visit began with a day of sightseeing and social activities. The more formal two-day grand opening began on Thursday, Nov. 3, 2022, in the new Interdisciplinary Science and Engineering Building at UCI. Faculty and students from ICS and HPI shared their research, starting with ICS Chancellor’s Professor of Computer Science Sharad Mehrotra. Talking about his work for the Testbed for IoT-based Privacy-Preserving PERvasive Spaces (TIPPERS) project and for COVID-19 location monitoring, Mehrotra highlighted machine learning problems related to privacy concerns and systems development. Gerard de Melo of HPI then shared his research into natural language understanding, natural language processing, and machine learning.
The two faculty talks were followed by more than 35 students giving 90-second lightning talks on topics ranging from deep learning for star/galaxy classification to reinforcement learning for revenue management. Next, a panel discussion on the future of AI and data science featured HPI’s de Melo and Felix Naumann and ICS Professors Sameer Singh and Roy Fox. Chancellor’s Professor of Computer Science and Statistics Padhraic Smyth of ICS moderated the discussion, which covered lofty expectations of AI in the next 10 years, touching on autonomous robots, self-driving cars and drone deliveries, as well as worrying trends, the persistent problem of errors in data, and the significance of ethical practices and interdisciplinary perspectives.
The gathering on Friday, Nov. 4, took place at the Arnold and Mabel Beckman Center of the National Academies of Sciences & Engineering. During opening remarks, Meinel talked about this being the first generation to have a digital world and about the “burden and responsibility” of shaping this new world “in a way that fits with our human values.” Arguing that it is not enough “to do business as usual,” he explained the need for globally interconnected networks in creating a digital reality that “makes our world a better place.” The talks, poster session, and keynote speaker that followed then delved into the specifics of building and managing that digital world.
Distinguished Professor of Computer Science Michael Carey gave a talk on “AsterixDB: Scalable Infrastructure for Big Data Analytics,” focusing on how to handle big data and adapt smaller-scale solutions to accommodate massive data sets. This multiyear research effort bridges the gap between the intuitive software tools that data scientists use to build and refine models, and the powerful but unwieldy interfaces provided by the huge databases arising in science and medicine.
Two UCI and two HPI students then presented highlights from their work, showing a natural overlap in research from the two organizations on topics ranging from explainability to reliability and safety. From HPI, Felix Grzelka presented his classification work for railway infrastructure safety, and Marta Lemanczyk talked about deep learning for computational biology. From UCI, Noble Kennamer discussed his research to optimize scientific experiments, while Dylan Slack talked about how machine learning models can be understood via natural language conversations.
Andrew Wright, principal software engineer of Disney Digital Technology, gave a thought-provoking keynote on the gap between current data analytics and the insights people actually need in order to achieve their business goals. Emphasizing the importance of how academics interface with nontechnical people, he stressed the need to communicate the “big picture” outcomes of research innovation instead of technical details.
Everyone then moved out to the atrium for the poster session and a catered reception. The posters allowed students to provide more in-depth information from the lightning talks the previous day, and attendees could ask questions and further engage in discussion.
Outcomes and Next Steps
“It was great to meet with a diverse group of researchers at HPI,” says Slack, a fourth-year Ph.D. candidate. “I gained a diverse set of perspectives; got feedback, new directions, and potential ideas to try out; and forged useful relationships that I can draw on in my work going forward.”
Establishing new relationships was one of the main goals. “In the first couple years of the center … we’ve done a ton of work on topics that are very relevant to HPI, and there’s some overlap, but many projects have proceeded independently,” says Sudderth. “We’re excited that we’re starting to identify deeper connections with people at HPI, and we hope there will be a lot more continuing in the coming years.”
Those connections expanded beyond the HPI event as well. “I had a wonderful time in Irvine,” says Lemanczyk, who extended her stay to also attend the Southern California Natural Language Processing Symposium (SoCal NLP) and Thirty-sixth Conference on Neural Information Processing Systems (NeurIPS). “Discussions about the differences and similarities between the countries when it comes to picking research topics and carrying out research gave me a much better perspective on academia in the U.S. I also enjoyed the time with Sameer’s group at the conferences (SoCal NLP, NeurIPS) as well as the activities after work. I am looking forward to next year’s visit of the Ph.D. students from UCI in Potsdam!”
Going forward, the expectation is that there will be two joint meetings with HPI researchers each year, and there are plans in the works for a student exchange program. “We aspire to accomplish a lot more,” said Papaefthymiou, establishing a closer and stronger relationship through extended stays of HPI researchers here at Irvine and UCI researchers over in Potsdam.” He hopes that by combining perspectives and talents from researchers across the globe, HPI at UCI will cultivate true, human-centered solutions to data science problems.
— Shani Murray