Machine Learning for Networks - Course Assignments

This course explores the application of Machine Learning techniques to wireless network problems, with a focus on Wi-Fi networks.

Students learn to apply both classical ML algorithms (Linear Regression, SVM, K-means) and advanced deep learning approaches (Neural Networks, LSTMs, Federated Learning) to solve real-world networking challenges such as performance prediction, pose recognition, indoor localization, channel allocation, congestion detection, and traffic forecasting.

The course combines theoretical foundations with hands-on implementation using industry-standard tools like PyTorch, scikit-learn, network simulators, and real datasets.

Repository Structure

SEMINARS/          # 5 introductory assignments on ML fundamentals
LABS/              # 3 applied projects with real-world datasets
requirements.txt   # Python dependencies

• Seminars: Introduce ML concepts and tools (Linear Regression, SVM, K-means, Neural Networks, Federated Learning)
• Labs: Apply learned techniques to solve practical Wi-Fi network problems

Technologies Used

• Language: Python 3.8+
• Environment: Jupyter Notebooks
• ML Libraries: PyTorch, scikit-learn
• Data Analysis: pandas, numpy, matplotlib, seaborn, plotly

Assignments

Seminars

Seminar 1

Introduction to Linear Regression using Wi-Fi simulation data. This seminar focuses on data preparation, analysis, and visualization techniques.

• Loading and structuring datasets from text files
• Computing statistical measures (mean, standard deviation)
• Visualizing data distributions and building regression models

Seminar 2

Wi-Fi Pose Recognition using Support Vector Machines (SVM). This seminar explores classification techniques applied to Channel State Information (CSI) data.

• Loading and visualizing skelleton points with pose labels
• Analyzing CSI amplitude distribution across subcarriers
• Building SVM classifiers with different preprocessing pipelines

Seminar 3

Introduction to K-means clustering for indoor localization using RSSI (Received Signal Strength Indicator) data from multiple Access Points.

• Loading and analyzing RSSI measurements from WiFi Access Points
• Visualizing 2D and 3D scatter plots of RSSI data by position
• Implementing K-means clustering for position classification

Seminar 4

Building Feedforward Neural Networks (FNN) with PyTorch to predict Wi-Fi network performance (throughput and latency).

• Performing correlation analysis between network features
• Feature selection for optimal prediction
• Designing and training neural networks for regression tasks

Seminar 5

Introduction to Federated Learning for collaborative model training across multiple clients without centralizing data with PyTorch.

• Loading and preprocessing distributed client datasets
• Implementing federated averaging algorithms
• Training neural networks in a privacy-preserving distributed manner

Labs

Lab 1

Multi-Armed Bandits (MABs) for channel allocation in Overlapping Basic Service Sets (OBSS) using the Komondor Wi-Fi simulator.

• Set up and run Wi-Fi simulations with Komondor
• Implement and compare different MAB algorithms: ε-greedy, UCB, softmax, Thomson sampling, and random.
• Analyze throughput performance and agent learning behavior

Lab 2

Congestion Detection in Wi-Fi networks using supervised learning. Training neural networks with Inter-Frame Space (IFS) histogram data from ns-3 simulator.

• Analyzing IFS histograms and collision rates for network saturation
• Implementing fully connected neural networks for binary classification
• Applying K-Fold Cross-Validation for robust model evaluation

Lab 3

Time Series Forecasting with Federated Learning for predicting user load across multiple Wi-Fi Access Points using LSTM networks.

• Analyzing and preprocessing time series data from multiple APs
• Implementing LSTM models for network traffic prediction
• Comparing centralized vs. federated learning approaches with performance metrics (MSE, MAE, MAPE)

Setup

Prerequisites

• Python 3.8 or higher
• pip (Python package manager)
• Jupyter Notebook, VS Code with Jupyter extension or PyCharm.

Installation Steps

1. Clone the repository

   git clone https://github.com/recuu7/AAX_SP.git
   cd AAX_SP

2. Create a virtual environment (recommended)

   python -m venv venv
   source venv/bin/activate  # On macOS/Linux
   # or
   venv\Scripts\activate  # On Windows

3. Install dependencies

   pip install -r requirements.txt

4. Launch Jupyter Notebook

   jupyter notebook

   Then navigate to the desired seminar or lab folder and open the .ipynb file.

Additional Notes

• Labs may require additional setup (e.g., Lab 1 uses the Komondor simulator).
• Most notebooks are configured to run also on Google Colab with Google Drive mounted. However, all of them can run locally.

Course information

• Enginyeria en Informàtica (Computer Science)
• Universitat Pompeu Fabra (UPF)
• Academic year: 2023-2024
• Instructor:

Group Members

• Xavier Cañadas
• Sergi Recuero
• Pau Reig