Artificial Intelligence and Computer Networks

The course “Artificial Intelligence and Computer Networks” explores the role of AI in analyzing, optimizing, and managing network infrastructures. It introduces core machine learning techniques, from decision trees and clustering algorithms to neural networks and reinforcement learning, with applications in QoS management, anomaly detection, traffic load prediction, and node energy management. The course combines theory with hands-on Python/Colab labs, where students apply algorithms to flow data and network topologies. Its goal is to provide both a conceptual understanding of AI principles and practical skills for developing intelligent and efficient approaches to modern network management.

Upon completion of the course, students will be able to:

• Understand how Artificial Intelligence is applied to optimize networks and edge/fog infrastructures.
• Analyze and preprocess network data for training Machine Learning algorithms.
• Apply classical algorithms (Decision Trees, K-NN, SVM, Clustering) in QoS and flow analysis scenarios.
• Use neural networks (CNN, RNN, GNN) for traffic load prediction and anomaly detection in networks.
• Implement reinforcement learning methods (Q-Learning, Deep Q-Learning) for routing and node energy management.
• Design intelligent and efficient strategies for managing modern network infrastructures.

1. Networks and Edge Computing
   •    Basic principles of edge computing
   •    Use cases in network infrastructures
   •    The role of AI in the edge

2. Introduction to AI and Network Optimization
   •    Definition of AI and main categories (supervised, unsupervised, reinforcement learning)
   •    Relationship between AI and computer networks
   •    Optimization examples with applications in edge/fog environments

3. Network Performance Metrics and Data for Machine Learning
   •    Throughput, latency, packet loss, QoS
   •    Collection, preprocessing, normalization of data
   •    Train/test split and network-related datasets

4. Decision Trees and k-Nearest Neighbors for QoS Classification
   •    Decision Trees: basic principles and interpretability
   •    K-NN: distance metrics and QoS prediction
   •    Applications in flow classification

5. SVM and SVR for Network Performance and Latency Prediction
   •    SVM for QoS class classification
   •    SVR for latency prediction
   •    Examples of applications on network data

6. Clustering Algorithms for Node Grouping and Traffic Patterns
   •    k-Means algorithm
   •    Hierarchical clustering
   •    Applications: grouping nodes and traffic patterns

7. Ensemble Methods for Network Flow Analysis and Diagnosis
   •    Bagging, Boosting, Random Forests
   •    Robustness to noisy network data
   •    Feature importance and SLA failure analysis

8. Training and Shallow Neural Networks for QoS Classification
   •    Backpropagation, loss functions, optimization (SGD, Adam)
   •    Shallow NNs (MLP) for QoS classification
   •    Overfitting issues and regularization

9. Convolutional and Recurrent Neural Networks for Traffic Prediction
   •    CNNs: traffic matrices as images, anomaly detection
   •    RNNs: traffic load prediction (time series)
   •    Examples on network datasets

10. Graph Neural Networks for Network Structure and Routing
    •    Architecture and basic principles
    •    Computer networks as graphs
    •    Applications: routing prediction, anomaly detection

11. Self-Organizing Maps for Network Traffic Anomaly Detection
    •    SOM theory and visualization (U-matrix)
    •    Application in anomaly detection on flows
    •    Combination with unsupervised methods

12. Q-Learning for Node Energy Management
    •    Basic theory of Q-Learning
    •    Definition of states, actions, rewards for routing/recharging
    •    Application in dynamic routing policy

13. Deep Reinforcement Learning (Deep Q-Learning) for Dynamic Network Optimization
    •    From Q-table to Deep Q-Networks (DQN)
    •    Applications: dynamic spectrum allocation, energy-aware routing
    •    Advanced scenarios in edge/fog networks