Cohere For AI - Guest Speaker: Dr. Adaku Uchendu AI Researcher

In recent years, text generation techniques have greatly advanced, especially in therealm of Large Language Models (LLMs), such as ChatGPT. These LLMs are able to generatetexts that are easily misconstrued as human-written. This, therefore, poses a security risk, asthese LLMs can be used to generate disinformation at scale with little cost. Thus, to combat thisnovel challenge, two new computational problems emerge - (1) “Deepfake” AuthorshipAttribution (AA) and (2) “Deepfake” Authorship Obfuscation (AO) problems, where the AAproblem is concerned with attributing the authorship of a given text to the true author, whilethe AO problem is aims to circumvent accurate attribution of a given text by modifying parts ofthe text. Therefore, in this talk, the focus is to call attention to the serious security risk of LLMs,and how these 2 computational problems propose solutions to mitigate this risk.

Place of work: MIT Lincoln LabBio: Dr. Adaku Uchendu is an AI Researcher at MIT Lincoln Lab. She earned her Ph.D.in Information Sciences and Technology at The Pennsylvania State University under theguidance of Dr. Dongwon Lee at the PIKE Lab in August 2023. While at Penn State, she was aButton-Waller Fellow, an NSF Scholarship for Service Scholar, and an Alfred P. Sloan Scholar.Her dissertation, titled “Reverse Turing Test in the Age of Deepfake Texts,” focused onunderstanding and detecting AI-generated texts. She earned a B.S. in Mathematics, with aminor in Statistics at University of Maryland Baltimore County (UMBC) in May 2018. While atUMBC, she was a McNair scholar and a member of Pi Mu Epsilon (the Mathematical HonorarySociety). Uchendu’s research interests are in Artificial Intelligence, Adversarial Robustness, andTopological Data Analysis in the application domain of Cybersecurity.

Add event to calendar

Apple Google Office 365 Outlook Outlook.com Yahoo