News
UMIACS Team Aims to Boost High-Performance Computing Software Development Using AI
UMD’s Tom Goldstein is leading a $7 million DOE-funded project to enhance AI-assisted software development for high-performance computing, aiming to improve the effectiveness of large language models in parallel programming and boost developer productivity.
Goldstein Leads Team to Invention of the Year Honors
UMD’s Tom Goldstein led a team that developed a technology, called Binoculars, which can detect text generated by large language models at an almost 90% success rate.
Addressing Gaps in Data Governance to Promote Trust in Generative AI
TRAILS partnered with the Digital Trade and Data Governance Hub to host a conference to identify gaps in data governance related to large language models and discuss ideas to address them.
Rising Stars Workshop Engages Machine Learning Community
TRAILS is a sponsor of the University of Maryland Center for Machine Learning’s 2023 Rising Stars in Machine Learning Workshop, designed to support researchers from underrepresented groups. This year’s committee included UMD’s Soheil Feizi, Hal Daumé, Marine Carpuat and Tom Goldstein.
UMD Team Wins DOE Award to Advance AI Using Supercomputers
UMD’s Tom Goldstein has won a competitive DOE award that will provide him access to some of the world’s most powerful computational platforms. His team will utilize these platforms to scale distributed AI training and develop new AI vision and language models.
Researchers Tested AI Watermarks—and Broke All of Them
UMD’s Soheil Feizi and Tom Goldstein are publishing studies detailing how easy it is to evade current methods of watermarking—and even add fake watermarks to real images.
Researchers Develop and Test AI Text-Detector Tools
UMD’s Soheil Feizi, Tom Goldstein and Furong Huang are publishing groundbreaking research on the reliability of AI text-detectors.
How ChatGPT Could Embed a ‘Watermark’ in the Text It Generates
Research on watermark detection by OpenAI and UMD’s Tom Goldstein was featured in The New York Times.
A Watermark for Chatbots Can Expose Text Written By AI
UMD’s Tom Goldstein discusses digital watermarks used to detect AI-generated text in a story published in MIT Technology Review.
UMD Machine Learning Experts Presenting Multiple Papers and Workshops at ICML 2022
UMD’s Hal Daumé III, Tom Goldstein and Soheil Feizi are among a mix of UMD faculty, postdocs and students presenting 18 papers and featured in 18 workshops at the 39th International Conference on Machine Learning in Baltimore.
$3.1M Award Aims to Prevent Attacks on AI Security Measures
Supported by a $3.1 million DARPA award, UMD’s Tom Goldstein and Abhinav Shrivastava are working to develop safeguards against deception attacks on machine learning algorithms.