Inside one of the University of Washington’s computer science buildings on Tuesday evening, students showed off smart earrings that monitor health metrics, and earbuds that measure blood pressure.
On the floor below, an assistive dexterous arm picked up pieces of fruit as part of a robot-assisted feeding system.
Others demoed their research on the implications of facial recognition technology and security of government websites.
The annual Research Showcase and Open House at the UW’s Paul G. Allen School of Computer Science & Engineering offered a glimpse at the current state and potential direction of computing — demonstrating the growing impact of artificial intelligence as both a focus and a tool for computer science breakthroughs.
At the outset of the research process, generative AI tools such as ChatGPT and Google Bard are dramatically accelerating the process of synthesizing and summarizing existing computer science literature, while also helping to brainstorm potential questions to study, said UW computer science professor Shwetak Patel.
“Getting to a candidate research hypothesis is so much faster now,” Patel said. Before, he explained, “it would take months.” But now, “You can do this in an hour.”
Focus on AI advances
Many of the UW researchers are pursuing — and achieving — AI breakthroughs.
Seattle venture capital firm Madrona Venture Group each year recognizes teams that demonstrate top research with strong commercial potential. This year both the winner, a project called QLoRA, and runner-up, dubbed Punica, are working on different ways to more efficiently fine-tune large language models.
The picks reflect the recent boom and attention on generative AI and LLMs.
“While a lot of exciting news comes from industry right now, the research presented today shows some of the importance and impact of academic research in this space,” said Magdalena Balazinska, the Allen School director.
The event also highlighted the emerging disparities in the field, and efforts to overcome them.
The luncheon keynote speaker, Hanna Hajishirzi, a UW associate professor and senior research manager at the Allen Institute for AI (AI2), gave attendees the latest details on OLMo, an AI2 initiative to develop a transparent, open large language model.
“The challenge that we’re facing is that all these state-of-the-art models nowadays are being developed by private companies. And all of these models are proprietary,” she said. “So it’s very hard for AI researchers to actually understand and analyze what is going on behind the doors of these large language models.”
While researchers can use existing large language models as part of their work, funding and access to the immense processing power needed to train their own LLMs is an ongoing challenge, said the UW’s Patel.
“We just literally don’t have the compute,” Patel said. “We have to think about research problems that can inform foundational models, or think about application areas. But academia and industry have to co-evolve. And it’s hard, honestly, to train these models in an academic context.”
Assistive and embedded technology
Many of the UW projects showed how tech can be used for good.
Madrona’s second runner-up award went to a team working on wireless earbuds that can perform hearing screenings.
The “people’s choice award” went to the group building the robot-assisted feeding system, a project aimed at helping those who are unable to perform essential tasks live more independently.
“Robots can really represent an extension of one’s independence and extension of one’s ability to act in the world,” said Amal Nanavati, a UW Ph.D. computer science student on the team. “I think we need more people focusing on projects like this, to take cutting edge technology that we are actively developing and apply it to the needs of people who have been underserved by technological progress so far.”
Several projects demonstrated how technology is becoming smaller, faster, cheaper, and more embedded.
“It’s passively there and helps you get better health,” said Shirley Xue, a Ph.D. student who helped develop smart earrings for health monitoring.
UW students aren’t just focused on developing better software or hardware. They’re also thinking about the implications of the technology on society.
Ph.D. student Rachel Hong is part of a team researching racial equity in facial recognition software. The work focuses on data collection methods that power such models, which have sparked controversy.
“With all the push for machine learning and LLMs, they work well a majority of the time — but when they don’t, it can be incredibly consequential,” Hong said.
View a list of the presenting teams for the poster sessions here, and those who gave presentions during the day here.
