📌 Introduction

This repository is a collection of lab practices where I implemented Activation Layers, Machine Learning, Deep Learning models during the Introduction of Deep Neural Networks course.

💡 What I learned

• Machine Learning Basics, Deep Neural Networks Basics
• Overfitting and Model Generalization
• Evaluation Protocol and Metrics
• Backpropagation Algorithm
• Optimization
• Convolutional Neural Networks (LeNet-5, AlexNet, VGGNet, GoogLeNet, ResNet)
• Recurrent Neural Networks (LSTM, Seq2Seq, Attention Mechanism, BERT, Transformer)
• Deep Gernerative Models
• Generative Adversarial Networks
• How to implement Activiaton Layers, ML alogorithms with Numpy
• How to implement models with PyTorch

💻 Contents

• DNN_HW1
  • Implement Linear Regression with Numpy library (models/LinearRegression.py)
  • Implement Logistic Regression Numpy library (models/LogisticRegression.py)
  • Tune # of training epochs and learning rate to minimize MSE for 'Graduate' and 'Concrete' datasets.
  
  
• DNN_HW2
  • Implement activation layers(sigmoid, ReLU, tanh) with Numpy library. (Answer.py)
  • Implement fully-connected layer(FCLayer) with Numpy library. (Answer.py)
  • Implement the softmax layer* with Numpy library. (Answer.py)
  • Report test acuuracy on MNIST using three different activation fuctions with a given DNN architecture and parameters.
  • [Random Search on FashioMNIST] Optimize model architecture(# of hidden layers, # of hidden nodes, # of epochs, learning rate etc.) to achieve best results on FashionMNIST.
  
  
• DNN_HW3
  • Implement Multi Layer Perceptron Classifier with PyTorch framework. (model/MLP_classifier.py)
  • Implement Multi Layer Perceptron Regressor with PyTorch framework. (model/MLP_regressor.py)
  • [Random Search on House] Optimize model architecture(# of hidden layers, # of hidden nodes, # of epochs, learning rate etc.) to achieve best results on House Dataset.
  • [Random Search on FashioMNIST] Optimize model architecture(# of hidden layers, # of hidden nodes, # of epochs, learning rate etc.) to achieve best results on FashionMNIST.
  
  
• DNN_HW4
  • Implement AlexNet with PyTorch framework. (models/ALexNet.py)
  • Implement ResNet with PyTorch framework. (models/ResNet.py)
  • [Random Search on MNIST] Optimize model architecture(# of hidden layers, # of hidden nodes, # of epochs, learning rate etc.) to achieve best results on MNIST.
  • [Random Search on FashioMNIST] Optimize model architecture(# of hidden layers, # of hidden nodes, # of epochs, learning rate etc.) to achieve best results on FashionMNIST.
  
  
• project
  • Conduct Semi-supervised learning for image classification.
  • Build a machine learning model for image classification, where a few data are only labeled and most of the data are unlabeled.

🔧 Tech Stack

Language

Framework