A 2006 research paper asked a question that has only grown more urgent since: Can machine learning be secure? Anthony D. Joseph was among its coauthors. He is a computer scientist at the University of California, Berkeley, where his faculty biography lists him as a Chancellor's Professor, and the University at Buffalo's directory of Black computer scientists identifies him with Berkeley.
Joseph earned his computer-science doctorate at MIT in 1998, a period when the internet was moving from universities and companies into everyday life. That shift raised new questions for researchers: systems now spread their work across many machines, any of which could fail or be compromised, and information traveled over networks that no single party controlled.
His research spans secure machine learning, distributed and mobile computing, wireless communications and genomics. Distributed computing splits a task across many computers; mobile and wireless systems add unreliable links and devices that move. Genomics brings very large biological datasets into the same picture. The common thread is the security of systems that process and exchange information.
His collaborative publications include a 2004 paper on Tapestry, a network overlay for deploying services. An overlay is a network built in software on top of another network, which lets researchers route messages in new ways without replacing the physical infrastructure underneath. The 2006 security paper examined what happens when the data used to train machine-learning systems can be manipulated, an early contribution to a research area that now has its own conferences and journals.
Berkeley's publication record credits Joseph alongside his coauthors, and his work continues through the university's faculty and its many research collaborations.
The questions in Joseph's portfolio have moved closer to daily life since he posed them. Machine-learning systems now help sort email, filter spam and recommend content, and each can be misled by someone who feeds it altered data. Distributed systems spread work across machines that can fail or be compromised, and security research asks what happens then. His genomics work adds the scale of modern biological data, where a single experiment can produce more information than earlier researchers handled in a career.
