machine learning project

01. Python tutorial and LaTex practice

• Learn usage of basic python syntax, numpy and matplotlib

02. Logistic regression for binary classification

• Classify two hand digit images into 0 and 1.

03. Logistic regression for binary classification

• Classify multiple hand digit images of 0 and 1.
• testing accuracy at iteration 900 : 1.0000000000

04. Classification for multiple classes based on softmax and cross entropy

• Logistic regression for multi-class classification.
• Classify multiple hand digit images of 0, 1, 2, 3 and 4.
• Use One-Hot Encoding.
• testing accuracy at iteration 900 : 0.9733333333

05. Classification for multiple classes with bias, weight-decay and stochastic gradient descent

• Logistic regression for multi-class classification.
• Classify multiple hand digit images of 0 ~ 9.
• Use One-Hot Encoding.
• testing accuracy (mean) at different mini-batch, weight-decay 0
• testing accuracy (mean) at different weight-decay, mini-batch0

06. Classification for multiple classes using Pytorch library

• Multi-class classification based on Softmax and Cross-Entropy using pytorch.
• Classify multiple hand digit images of 0 ~ 9.
• Constructed a neural network using a series of convolutional layers.
• best testing (mean) accuracy within the last 10 epochs : 97.7875000000

07. Image Denoising in a superivsed learning framework using Pytorch library

• Denoise the noised images.
• Constructed a neural network in the form of auto-encoder that consists of encoder and decoder.
• best testing PSNR (mean) within the last 10 epochs = 25.1513428898

08. Image Denoising in an unsuperivsed learning framework using Pytorch library

• Denoise the noised images.
• Constructed a neural network in the form of auto-encoder that consists of encoder and decoder.
• best testing PSNR (mean) within the last 10 epochs = 25.4595453543

09. Image Segmentation in a supervised learning framework using Pytorch library

• Get the clear boundaries of the cat images from the original.
• Constructed a neural network in the form of auto-encoder that consists of encoder and decoder.
• best testing accuracy within the last 10 epochs = 74.2580732318

10. Image Segmentation in an unsupervised learning framework using Pytorch library

• Get the clear boundaries of the sqaure images from the original noised sqaure images.
• Constructed a neural network in the form of auto-encoder that consists of encoder and decoder.
• best testing accuracy within the last 10 epochs = 97.8995329178

11. Image de-blurring by a supervised learning using Pytorch library

• De-blur the blurred images.
• Constructed a neural network in the form of auto-encoder that consists of encoder and decoder.
• best testing PSNR (mean) within the last 10 epochs = 24.1219267082

12. Image Generation via Generative Adversarial Networks

• Create square images by learning from square images.
• Constructed neural networks of a generator and a discriminator.
• best accuracy within the last 10 epochs = 96.0030833364