|Math Coffee: "Higher-Order Automatic Differentiation Methods in MATLAB" Ben Altman ‘10
When: April 07,
Speaker: Ben Altman ‘10Ticket Required: No
Automatic Differentiation (AD) uses computer code to numerically calculate derivative values such that there is no approximation error. It is ideal when symbolic differentiation is cumbersome or the function cannot be expressed in a concise form (defined by a large program code). Standard operators are overloaded to return not just function values but Taylor polynomial coefficient values as well. Implementing higher-order AD of multivariate functions can be quite difficult. Enormous amounts of data must be stored and operated on repeatedly. While there is no approximation error, the large number of operations can increase round-off error. There are different methods to that attempt to minimize these complications. Three different methods of higher-order AD of multivariate functions were implemented in MATLAB. These methods were compared for their efficiency and accuracy on test runs.
We will gather in Math Hall at 3:45 for our traditional light snacks.
This project was also presented at:
- Math. Assoc. of America Southeast Region annual meeting at Elon Univeristy, March 2010.
- Davidson Science and Mathematics Research Poster Fair, May 2010.
- Oral honors defense before the faculty.
It culuminated in graduation with Honors in Mathematics with the thesis:
- Higher-order Automatic Differentiation of Multivariable Functions in MATLAB, in the Davidson library.
Contact: Prof. Donna Molinek