Barak A. Pearlmutter

Barak A. Pearlmutter is a computer scientist and computational neuroscientist at Maynooth University (Brain and Computation Lab; formerly Siemens Corporate Research and the University of New Mexico, with joint neuroscience and electrical-engineering appointments) whose work runs across automatic differentiation, neural networks, acoustic source separation and localization, and programming-language design — his own homepage lists his interests as "adaptive systems, neural Networks, how the brain works, machine learning, acoustic source separation and localization, neuroscience, programming language design," and he is also a practicing ceramic artist. His 1994 paper "Fast Exact Multiplication by the Hessian" introduced the differential R-operator now known as "the Pearlmutter trick."

Within this vault Pearlmutter is the second author of the JMLR 2018 survey (Baydin, Pearlmutter, Radul & Siskind) that anchors much of moc-backpropagation-origins and the claim claim-ml-and-ad-communities-mutually-unaware — a citation the vault leaned on constantly before it ever examined the person behind it. His own cross-field career is a personal instance of the disciplinary bridging that survey both documents and performs.

References

• claim-ml-and-ad-communities-mutually-unaware — the mutual-unawareness claim resting on the survey he co-authored.
• claim-backpropagation-special-case-of-reverse-mode-ad — the survey's backprop-as-reverse-mode-AD identity, also co-authored by him.
• claim-pearlmutter-1994-r-operator-exact-hessian-vector-product — his 1994 exact Hessian-vector-product trick, still named infrastructure decades on.
• observation-ad-survey-second-author-personally-spans-ml-ad-divide — the observation that the survey's own authorship spans the divide it describes.