Homework answers / question archive / CSE 625 Parallel Programming Term Project 100 points (30% of the final grade) Due: December 7 (Wed) midnight (Submit your project report and any related source code to the Blackboard

CSE 625 Parallel Programming Term Project 100 points (30% of the final grade) Due: December 7 (Wed) midnight (Submit your project report and any related source code to the Blackboard

Computer Science

Share With

---

CSE 625 Parallel Programming Term Project

100 points (30% of the final grade)

Due: December 7 (Wed) midnight (Submit your project report and any related source code to the Blackboard.)

1 (30 points)

In the CodeBlocks project, All_Pair_distance, it implements three functions using C++ multi-threads to compute the pair-wise distance matrix of MNIST train images (loaded from train-images.bin). These three methods are:

 

                1 block_all_pairs (C++ multi-threads - block work distribution)

2 block_ cyclic_all_pairs (C++ multi-threads - block cyclic work distribution)

                3 dynamic_all_pairs (C++ multi-threads - dynamic work distribution)

 

Use OpenMP to re-implement these three methods.

 

1.1 Implement block work distribution in OpenMP and compare its computing time with  that of C++ multi-threads block_all_pairs implementation. List your OpenMP implementation in the report and put the computing time results in the following

      table.

 

Matrix Size          400         800         10,000   20,000   30,000   60,000

C++ Block

12 threads                                                                                          

OpenMP block

12 threads                                                                                          

 

1.2 Implement block-cyclic work distribution in OpenMP and compare its computing

      time with that of C++ multi-threads block_cyclic_all_pairs implementation. List

      your OpenMP implementation in the report and put the computing time results in the

      following table.

 

Matrix Size          400         800         10,000   20,000   30,000   60,000

C++ block-cyclic

12 threads

Chunk size 2                                                                                      

OpenMP block-cyclic

12 threads

Chunk size 2                                                                                      

 

 

1.3 Implement dynamic work distribution in OpenMP and compare its computing time

      with that of C++ multi-threads dynamic_all_pairs implementation. List your

      OpenMP implementation in the report and put the computing time results in the

      following table.

 

Matrix Size          400         800         10,000   20,000   30,000   60,000

C++ dynamic

12 threads

Chunk size 2                                                                                      

OpenMP dynamic

12 threads

Chunk size 2                                                                                      

 

2 (70 points) Individual term project

The topic of the term project must relate to solving computationally intensive problems

using any parallel computing platforms such as C++ multi-threading, OpenMP, CUDA, python libraries (e.g, PyTorch), etc.

 

Select your term project from one of the follwig list of project topics:

 

1 sp500 stock price Similarity Computation Efficency and Analysis

 

   The goal of this topic is a study of computation of similarity matrix of time series

  (based on various similarity measurements) and their applications.

 

   Code example:

 

         sp500.zip (available on the Blackboard)

 

   References:

 

   [1] Measuring Financial Time Series Similarity With a View to Identifying Profitable

        Stock Market Opportunities

        (https://arxiv.org/pdf/2107.03926.pdf)

 

   [2] A similarity measurement for time series and its application to the stock

         market

         (https://www.sciencedirect.com/science/article/pii/S0957417421006503)

 

   [3] A Fuzzy Approach for Similarity Measurement in Time Series, Case Study for

        Stocks

        (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7274751/)

 

2 Celeb Faces Dataset analysis

 

   The goal of this topic is a study of computational efficiency of large image dataset.

  

    Code example:

 

      celebA.zip (available on the Blackboard)

 

    The example uses the Large-scale CelebFaces Attributes (CelebA) Dataset, which can be downloaded here: https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html

 

    CelebFaces Attributes Dataset (CelebA) is a large-scale face attributes dataset with more than 200K celebrity images, each with 40 attribute annotations. The images in this dataset cover large pose variations and background clutter. CelebA has large diversities, large quantities, and rich annotations, including

 

•             10,177 number of identities,

•             202,599 number of color (R, G, B) face images, and

•             5 landmark locations, 40 binary attributes annotations per image.

 

3 MNIST written digit classifiers Implementation and Analysis (C++)

 

   3.1 OneNN classifier and

   3.2 NN_Softmax classifier

         (Download link: https://louisville.box.com/s/9hpbooqc7n8xr5jk740byr2gaih94tv9)

 

  The goal of this topic is a study of classifier computation efficiency ans analyze and compare the accuracy of the classfiers and the misclassified digit images.

 

 

4 CUDA applications using CodeBlocks gcc compiler

 

5 PyTorch (CPU and/or GPU) applications of reasonably large-scale computing

   (e.g., computing all-pair distance matrix using various distance metric) 

 

 

 

Project Report Outline

 

1 Title page

 

 

CSE 625 Term Project Report

 

<Project Title>

 

 

<Name>

 

mm-dd-yyyy

 

 

2 Project statement and objective

 

3 General description of the approach (including required platforms)

 

4 Implementation details (showing important source code snippets in the report)

 

5 Contributions

 

6 Reference