Graduate Student Akihiko Nishimura Wins Laplace Award
The Laplace award is given to the best student paper submitted to the Bayesian Statistical Science section at Joint Statistical Meetings. Ten students are chosen as competition winners and the best of the ten recieve the Laplace Award: http://community.amstat.org/sbss/awards
Akihiko Nishimura explains his paper below:
My algorithm is called 'geometrically tempered Hamiltonian Monte Carlo.' As the name suggests, it draws ideas from a variety of fields; Riemannian geometry, probability theory, numerical analysis, as well as fluid dynamics.
Among many other applications, Hamiltonian Monte Carlo algorithm is widely used in statistics and machine learning to make predictions and draw inferences from observed data. For example, the popular open-source software package “Stan” for probabilistic modelling is based on Hamiltonian Monte Carlo. One main situation Hamiltonian Monte Carlo fails, though, is when the so-called “likelihood function” has multiple local maxima; this can happen when multiple probabilistic models explain the observed data equally well. My algorithm is an attempt to improve Hamiltonian Monte Carlo in such situations.
This is a highly interdisciplinary field where experts from different fields can contribute. I knew enough of everything to “prove the concept” but there is a lot of room for further development. With the help of fellow mathematicians, I hope my idea can develop into a practical tool for data analysis and other scientific inquiries. "