Projects

This page records part of my class projects. It is unfinished yet, more projects details will be added soon. Right now, I separate the projects into:

Parallel Computing

Accelerate GPU Convolutional Neural Network (CNN, one deep learning strategy) with Auto-tuning

Convolutional Neural Network (CNN) is a powerful tool in machine learning area, it can handle the problems in image classification and signal process. However, it requires a large mount of the traning time for this system. Thus, there are tons of efforts to speedup the traning process. Because of the property of convolutional calculation, GPU acceleration is applied to speedup the CNN. However, in the different layers, the size of input images varies largely, which decreases the efficiency of the convolution.

In this project, we concatenate multiple input images into one batch. Thus, we create larger matrix per cuBlas call. In order to do the auto-tuning, we create an extra-parameter v, which represents the new size of images. Auto-tuning v for peak performance in different layers can optimize the system for different computer. We speedup the Caffe (Berkeley CNN framework, the fastest open source one as we know) 1.5-2 times.

Accelerate image blur and scan

Implement high-performance and memory-efficient image blur using SSE intrinsics, OpenMP, loop unrolling, register blocking, and cache blocking by C. Implemented reduce and efficient scan with GPU speedup by openCL.

General Machine Learning

Bayesian Network Model to predict the highway network's performance

More ...

Nature Language Process (NLP)

Probabilistic Parsing

details will be added soon

Speech Recognition

details will be added soon

Word Alignment

details will be added soon

Image Computing

Drawing Process Reconstruction from Single Image

details will be added soon

Gradient Domain Fusion

details will be added soon

Stitching Photo Mosaics

details will be added soon

Statistic Data Analysis

Brain Responses to Visual Images

details will be added soon

Predict the Traffic State for Bay Bridge

details will be added soon

Cloud Detection

details will be added soon

Othes

Compiler

details will be added soon

Design of CPU

details will be added soon

Chips Game

details will be added soon