PORTALE DELLA DIDATTICA

Ricerca CERCA
  KEYWORD

GPU Based Parallel Algorithms to Solve Bayesian Networks

keywords DIGITAL SYSTEM DESIGN TEST AND VERIFICATION, GPU, NVIDIA, ACCELERATORE

Reference persons STEFANO DI CARLO

External reference persons Alessandro Vallero (alessandro.vallero@polito.it)

Research Groups TESTGROUP - TESTGROUP

Thesis type SPERIMENTALE

Description Bayesian Networks (BNs) are an efficient statistical model to represent multivariate statistical distribution functions.

BNs find application in several fields such as medical, artificial intelligence and reliability. Because of the great potential offered by BNs, solving them is extremely important. Unfortunately, find the exact solution of Bayesian network is an NP-hard problem, but to cope with this, in the literature, several approximation algorithms have been proposed.

The objectives of this thesis are:
- Analyzing algorithms for the approximate solution of BNs
- Parallelizing efficiently these algorithms by means of parallel computing
libraries, such as CUDA, OpenCL, OpenMPI, OpenMP, Posix, ...
- Analyzing performance of parallel algorithms

Don't miss the opportunity to work with one the most powerful GPGPUs ever built: NVIDIA Tesla K20 (3.52 TFLOPS, 2496 CUDA cores, 5 GB of GDDR5, Memory bandwidth 208 GB/sec)

Required skills C/C++ Programming


Deadline 04/07/2016      PROPONI LA TUA CANDIDATURA