Berkeley Lab - Scientific Computing Seminar

Date:

Tuesday, May 20, 2008

Time:

1:00pm-2:00pm

Location:

Building 50F, 1647 Conference Room

Seminar Speaker:

Sanjukta Bhowmick
Penn State University

Title:

The Potential of Machine Learning Algorithms in Scientific Computing

Abstract:

The computational science is currently facing an "embarrassment of riches", with extensive computational memory and power, algorithms tailor-made to suit specific applications, and architectures ranging from uniprocessors to multiprocessors to CMPs. Machine learning is a popular tool to match a given problem to the set of resources that will solve it efficiently.

However the scope of machine learning in scientific computing extends beyond mere mapping of applications to algorithms. In this talk I will discuss how machine learning can also be used to explore the "why of the problem", i.e. understanding what parameters control this mapping. I will show how data mining techniques can be applied to relatively simple problems such as obtaining good cache locality in sparse matrix vector multiplications and will describe some challenges in applying machine learning algorithms that are specific to the scientific computing domain.

In the final part of the talk, I will show how we give back to the machine learning community, specifically by an example where graph-layout algorithms can be used to improve data mining results.

Sponsor of Seminar:

Esmond G. Ng

Scientific Computing

Contact Esmond G. Ng EGNg@lbl.gov