Machine-Learning-MPs

Machine Learning-UIUC-CS course assignments

MP1: Linear Regression

• Implement the linear regression method using gradient descent.

• Implement linear regression by solving closed form.

MP2: SVM

• Training: Using projected gradient descent.

• Prediction: Using optimal dual solution.

• Kernel contour experiment.

MP3.1: Resnet

• Residual Block: Conv2d, BatchNorm2d, ReLU, Conv2d, BatchNorm2d, Concat(x), ReLU.

• Shallow Resnet: Conv2d, BatchNorm2d, ReLU, MaxPool2d, Residual Block, AdaptiveAvgPool2d, Linear.

MP3.2: Image overfitting

• __getitem__: Return the data element corresponding to the given index.

• Weight Initialization: uniform.

• train(): Implement main training loop.

• Activation: Experiment with ReLU, Tanh, and Sin.

MP4: Variational Autoencoder (VAE)

• Encoder: predict mean and variance of latent space.

• Decoder: given a Gaussian noise, generate objective.

MP5.1: Generative Adversarial Models (GAN)

• Discriminator: binary classifier optimized to distinguish real images from generated fake images.

• Generator: decoder optimized to generate high quality images that are close to real domain.

MP5.2: Diffusion Model

• Score-based generative model: Annealed Langevin dynamics.

• Encoder-Decoder: implement denoising loss.

MP6: Tabular Q-Learning

• Q-Function: update based on optimal one-step bootstrap.

• Initialization: Experiment with different Q-table value initialization.

Personal Notes

If you are interested in the theory aspects of Machine Learning, feel free to ask for access to my documents here:

Google Drive Folder

Disclaimer

THIS PROJECT IS INTENDED FOR WORK DISPLAY AND SHARING PURPOSES ONLY. THE CODE AND MATERIALS PROVIDED HERE SHOULD NOT BE SUBMITTED AS YOUR OWN WORK IN ANY FORM. USE OF THIS CODE FOR ACADEMIC ASSIGNMENTS OR EXAMS WITHOUT PROPER ATTRIBUTION OR PERMISSION MAY CONSTITUTE ACADEMIC MISCONDUCT INCLUDING PLAGIARISM. THE AUTHOR OF THIS PROJECT IS NOT RESPONSIBLE FOR ANY CONSEQUENCES ARISING FROM THE IMPROPER USE OF THIS CODE.