Sentrana Presents Scientific Research at GPU Technology Conference

Dr. Alireza Mahani, a member of Sentrana’s Quantitative Modeling team, presented new research on Gibbs Sampling at this year’s GPU Technology Conference (GTC) in San Jose, California
 
July 13, 2012 - PRLog -- Dr. Alireza Mahani, a member of Sentrana’s Quantitative Modeling team, presented his research on Gibbs Sampling at this year’s GPU Technology Conference (GTC) in San Jose, California.  The GTC advances global awareness of Graphic Processing Unit (GPU) computing and visualization. Dr. Mahani’s presentation entitled “GPU Parallelization of Gibbs Sampling” provided a forward-thinking means to accelerate the computation and lower the costs associated with Monte Carlo Markov Chain (MCMC) sampling.

In his presentation, Dr. Mahani discussed a strategy for accelerating MCMC sampling, using GPUs and Gibbs Sampling, with applications in estimating Hierarchical Bayesian models.  Typically the MCMC estimation is both time consuming and difficult to parallelize due to its sequential nature.  In order to accelerate Gibbs sampling, Dr. Mahani defined a taxonomy of variables and then provided GPU parallelization strategies for each variable type and their GPU implementation. When used for large problems, such parallelization approaches can accelerate estimation 30 to 100 times.  This provides the benefit of lowering computation cost, and also makes the modeling experience more interactive.

"Modeling should be a dynamic process,” explained Mahani.  “The GPU parallelization of Gibbs sampling enables our modeling team to significantly shorten the modeling lifecycle and maintain intellectual focus on the task at hand which translates into higher-quality analytics in a much shorter time frame for our clients."

To view a video of the “GPU Parallelization of Gibbs Sampling” presentation please visit the following link: http://sentrana.wistia.com/m/B676V6
End
Source:
Email:***@sentrana.com Email Verified
Phone:202-507-4531
Zip:2006
Tags:Gibbs Sampling, Gpu, Computation Cost, Modeling, Statistics
Industry:Business, Technology
Location:Washington - District of Columbia - United States
Account Email Address Verified     Account Phone Number Verified     Disclaimer     Report Abuse
Sentrana Inc News
Trending
Daily News
Weekly News



Like PRLog?
9K2K1K
Click to Share