Automatic Differentiation of Higher-Order Functions
New Automatic Differentiation methods increase computational efficiency and reduce memory requirements for deep learning and other complex computing systems.
Researchers at Purdue University have developed methods of Automatic Differentiation (AD) to be applied to both rigid computations and arbitrary computer programs. This technology greatly increases the efficiency of these processes while also reducing the amount of required computer memory, allowing for more complicated deep learning systems.
Advantages:
-Efficient
-Versatile
Potential Applications:
-Application Programmers
-Machine Learning
-AI
TRL: 5
Intellectual Property:
Provisional-Patent, 2019-04-29, United States | PCT-Gov. Funding, 2020-04-29, WO | NATL-Patent, 2021-10-28, United States | NATL-Patent, N/A, Europe | EP-Patent, N/A, Switzerland | EP-Patent, N/A, Germany | EP-Patent, N/A, United Kingdom
Keywords: Automatic Differentiation, AD, deep learning systems, machine learning, AI, rigid computations, computer programs, efficiency, versatility, application programmers, AI, Computer Technology, Machine Learning, Software