University of Tehran - Artificial Intelligence Course

This repository contains the coursework and assessments for the Artificial Intelligence course at the University of Tehran. Each folder represents a different coursework assignment (referred to as "CA"), and additional resources are provided for reference.

Repository Structure

• CA1/ - Coursework Assignment 1: Search and Genetic Algorithms.
• CA2/ - Coursework Assignment 2: Bayesian Networks and Hidden Markov Models.
• CA3/ - Coursework Assignment 3: Clustering and Dimensionality Reduction.
• CA4/ - Coursework Assignment 4: Data Analysis and Machine Learning Algorithms.
• CA5/ - Coursework Assignment 5: Convolutional Neural Networks (CNNs) and Word Embeddings.
• CA6/ - Coursework Assignment 6: Reinforcement Learning and Deep Q-Networks (DQN).
• Extra/ - Additional Material: Extra material related to the course, including code snippets, research papers, or supplementary assignments.
• REF/ - Reference Material: Contains reference documents, papers, or other resources used throughout the course.
• Slides/ - Course Slides: Lecture slides provided by the instructors.

---

Coursework Assignments

Coursework Assignment 1: Search and Genetic Algorithms

This assignment focuses on search algorithms and the application of genetic algorithms to solve the Knapsack problem.

Project Overview

Part 1: Search Algorithms

1. BFS (Breadth-First Search) and DFS (Depth-First Search):

   • Implemented to find the optimal path on a given graph.
   • Compared both algorithms based on different types of graphs.
   • Included tracking of explored and frontier sets.

2. Uniform Cost Search:

   • Implemented to find the minimum cost path from a start node (S) to a goal node (G).
   • Calculated the total cost considering both the path cost and heuristic values.

3. A* Search:

   • Implemented a heuristic search algorithm to find the shortest path on a graph.
   • Evaluated the admissibility and consistency of the heuristic function used.

Part 2: Genetic Algorithm for the Knapsack Problem

1. Knapsack Problem:

   • Implemented a genetic algorithm to maximize the value of items packed in a knapsack without exceeding its weight limit.

2. Genetic Algorithm Implementation:

   • Chromosome Representation: Each chromosome represents a possible selection of items.
   • Population Initialization: Randomly generated initial population.
   • Fitness Function: Evaluates chromosomes based on total value and weight of selected items.
   • Crossover and Mutation: Used to produce new generations of solutions.
   • Selection: Best solutions are selected based on the fitness function to form the next generation.

---

Coursework Assignment 2: Bayesian Networks and Hidden Markov Models

This assignment covers Bayesian Networks and Hidden Markov Models (HMM), focusing on probability calculations, forward/backward algorithms, and model evaluation.

Project Overview

Part 1: Bayesian Networks

1. Bayesian Network Design:

   • Evaluated the probability of car accidents based on factors like car age, weather conditions, and maintenance.
   • Implemented probability tables for each variable.

2. Probability Calculations:

   • Computed joint and conditional probabilities using the Bayesian Network.
   • Implemented the Variable Elimination algorithm for specific probabilities.

3. Independence Analysis:

   • Analyzed independence and conditional independence of variables.
   • Justified dependencies based on network structure.

Part 2: Hidden Markov Models (HMM)

1. HMM Design for Music Genre Prediction:

   • Hidden states represent user emotions; observations represent music genres.
   • Configured transition, emission, and initial probability matrices.

2. Algorithm Implementation:

   • Forward Algorithm: Calculated the probability of an observation sequence.
   • Smoothing Algorithm: Computed the probability of a hidden state at a given time.
   • Viterbi Algorithm: Determined the most likely sequence of hidden states.

3. Practical Implementation:

   • Implemented HMM using both hmmlearn library and a custom implementation.
   • Compared results and performance.

Data Processing and Feature Extraction

1. Feature Extraction:

   • Extracted features like MFCC, Zero Crossing Rate, and chroma features from audio data.
   • Investigated their impact on HMM performance.

2. Evaluation Metrics:

   • Evaluated model performance using Accuracy, Precision, Recall, and F1 Score.
   • Implemented metrics from scratch and compared with library results.

---

Coursework Assignment 3: Clustering and Dimensionality Reduction

This assignment involves implementing clustering algorithms, feature extraction, dimensionality reduction, and cluster evaluation.

Project Overview

Part 1: Clustering Algorithms

1. K-Means Clustering:

   • Clustered points into three clusters using K-Means.
   • Calculated new cluster centers and visualized clusters.
   • Determined iterations needed for convergence.

2. DBSCAN Clustering:

   • Applied DBSCAN with different ε (epsilon) and minPoints.
   • Visualized clusters and identified core, border, and noise points.

3. Agglomerative Clustering:

   • Performed clustering using a distance matrix.
   • Visualized hierarchical clustering with a dendrogram.

Part 2: Practical Implementation - Clustering Images

1. Dataset:

   • Used an image dataset associated with different colors.

2. Feature Extraction:

   • Employed VGG16 for high-level feature extraction.
   • Researched techniques and preprocessing required.

3. Clustering Implementation:

   • Applied K-Means and DBSCAN to feature vectors.
   • Experimented with different parameters.
   • Compared clustering results.

Part 3: Dimensionality Reduction and Visualization

1. PCA:

   • Reduced dimensionality using Principal Component Analysis.
   • Visualized clusters in 2D/3D space.

2. Evaluation:

   • Used silhouette and homogeneity scores to evaluate clusters.
   • Compared results before and after dimensionality reduction.

Part 4: Analysis and Discussion

• Comparison of Methods:

  • Discussed pros and cons of K-Means vs. DBSCAN.
  • Provided insights on algorithm preferences.

• Cluster Evaluation:

  • Detailed calculation methods for evaluation metrics.
  • Suggested improvements for cluster performance.

---

Coursework Assignment 4: Data Analysis and Machine Learning Algorithms

This assignment covers data preprocessing, regression, classification, model evaluation, and ensemble methods.

Project Overview

Part 1: Data Analysis and Preprocessing

1. Exploratory Data Analysis (EDA):

   • Analyzed data distribution and characteristics.
   • Visualized relationships between features.

2. Data Preprocessing:

   • Handled missing values.
   • Performed feature scaling.
   • Managed categorical and numerical features.
   • Split data into training, validation, and testing sets.

Part 2: Regression Analysis

1. Linear Regression:

   • Implemented from scratch without libraries.
   • Used gradient descent and least squares.
   • Evaluated using MSE, RMSE, and R² score.

2. Polynomial Regression (Optional):

   • Captured non-linear relationships.
   • Compared with linear model.

Part 3: Classification Algorithms

1. K-Nearest Neighbors (KNN):

   • Implemented KNN for classification.
   • Tested different distance metrics.
   • Optimized value of K.

2. Support Vector Machine (SVM):

   • Implemented with linear and RBF kernels.
   • Evaluated using confusion matrix and other metrics.

Part 4: Decision Trees and Ensemble Methods

1. Decision Trees:

   • Implemented for regression and classification.
   • Explored pruning techniques.

2. Random Forests:

   • Used as an ensemble method combining multiple trees.
   • Analyzed hyperparameter impacts.

3. Ensemble Methods:

   • Compared bagging and boosting.
   • Implemented Random Forests and XGBoost.

Part 5: Model Evaluation and Optimization

1. Evaluation Metrics:

   • Used confusion matrix, precision, recall, F1-score, and ROC-AUC.
   • Performed hyperparameter tuning with Grid and Random Search.

2. ROC Curve Analysis:

   • Generated and analyzed ROC curves.
   • Explained AUC significance.

Part 6: Final Report

• Report: Detailed implementation, results, and analysis in report.pdf.
• Contents: Visualizations and discussions on model outcomes.

---

Coursework Assignment 5: Convolutional Neural Networks (CNNs) and Word Embeddings

This assignment focuses on deep learning techniques, specifically CNNs and Word Embeddings.

Project Overview

Part 1: Convolutional Neural Networks (CNNs)

• Introduction to CNNs:

  • Structure including convolutional, pooling, and fully connected layers.
  • Processing and feature extraction from images.

• Implementation:

  • Designed architecture and selected activation functions.
  • Understood roles of different layers.
  • Performed hyperparameter tuning.

Part 2: Word Embeddings

• Word2Vec and GloVe:

  • Transformed words into dense vector representations.
  • Captured semantic relationships.

• Implementation:

  • Used pre-trained models to extract word vectors.
  • Evaluated effectiveness in capturing semantic similarities.

Part 3: Text Classification with CNNs

• Building a CNN:

  • Applied to text classification using word embeddings.
  • Designed suitable architecture for text data.
  • Tuned context window size and hyperparameters.

• Model Training and Evaluation:

  • Trained on the given dataset.
  • Evaluated using accuracy, precision, recall, and F1-score.
  • Visualized training with loss and accuracy curves.

Part 4: Regularization Techniques

• Addressing Overfitting:

  • Explored Dropout and Batch Normalization.
  • Implemented within CNN models.

• Performance Analysis:

  • Compared models with and without regularization.
  • Provided detailed findings.

---

Coursework Assignment 6: Reinforcement Learning and Deep Q-Networks (DQN)

This assignment delves into reinforcement learning, focusing on MDPs, TD-Learning, and DQNs.

Project Overview

Part 1: Markov Decision Processes (MDP)

1. Value Iteration:

   • Calculated optimal policy for a grid environment.
   • Derived Value function and optimal policy.

2. Monte Carlo and Q-Learning:

   • Explored as model-free approaches.
   • Compared convergence and efficiency.

Part 2: Temporal Difference Learning (TD-Learning)

1. Implementation:
   • Estimated value function for MDP.
   • Examined learning rates and exploration-exploitation tradeoff.
   • Compared with Monte Carlo methods.

Part 3: Deep Q-Networks (DQN)

1. Overview and Applications:

   • Researched DQN applications.
   • Implemented a basic DQN model.

2. Practical Implementation - Snake Game:

   • Created an AI agent for Snake using Q-Learning and DQN.
   • Trained to maximize score through reinforcement learning.

Part 4: Model Training and Evaluation

1. Training:

   • Used various hyperparameters.
   • Implemented Epsilon Decay for exploration-exploitation balance.

2. Evaluation:

   • Assessed cumulative reward and episode length.
   • Visualized Q-values convergence and total reward.

3. Hyperparameter Tuning:

   • Experimented with configurations.
   • Saved and compared models.

---

Additional Directories

• Extra/ - Additional Material:

  • Code snippets, research papers, or supplementary assignments.

• REF/ - Reference Material:

  • Reference documents and resources used in the course.

• Slides/ - Course Slides:

  • Lecture slides from the instructors.

---

Acknowledgements

• Instructors: Dr. Fadayee and Dr. Yaghoobzadeh
• University: University of Tehran
• Course: Artificial Intelligence

---

License

This project is licensed under the MIT License. For more details, see the LICENSE file.

This repository was created as part of the coursework for the Artificial Intelligence course at the University of Tehran. All rights to the content are reserved.