Course Information
| Course Code | YPZ312 |
| Program | Artificial Intelligence Engineering |
| Language | Turkish |
| Level | Undergraduate |
| Type | Elective |
| Prerequisites | Machine Learning |
| Theory | 2 Hours |
| Practice | 1 Hour |
| Laboratory | — |
| Credits | 2.50 |
| ECTS | 6.00 |
| Delivery | Face-to-face |
| Instructor | Asst. Prof. Ramazan Özgür Doğan |
Course Objective
The aim of this course is to provide students with the fundamental conceptual and mathematical foundations of artificial neural networks (ANN). Students learn classical ANN models ranging from the artificial neuron, perceptron, and multilayer networks with backpropagation to radial basis function networks, self-organizing maps (SOM), LVQ, and Hopfield networks; comprehend network structures, learning rules, and training processes; select and implement suitable network architectures for classification, clustering, and prediction problems through hands-on practice; and, by the end of the term, are introduced to convolutional neural networks (CNN), preparing them with a solid foundation for the Deep Learning course.
Course Content
Introduction to artificial neural networks and fundamental concepts; biological inspiration and the artificial neuron model; activation functions; learning paradigms and learning rules (Hebb, delta); discrete and continuous perceptrons, linear separability; multilayer perceptron and forward propagation; backpropagation (generalized delta rule); factors affecting learning and generalization; radial basis function (RBF) networks; unsupervised learning, self-organizing maps (SOM) and LVQ; recurrent networks, Hopfield and associative memory, ART; transition to deep networks and the fundamentals and architectures of convolutional neural networks (CNN).
Weekly Schedule
| Week | Topic |
|---|---|
| 1 | Introduction to ANNs; fundamental concepts, history, biological inspiration, and application areas |
| 2 | The artificial neuron model and fundamental concepts; activation functions |
| 3 | Learning paradigms and learning rules (Hebb, delta/Widrow–Hoff) |
| 4 | Classification and the discrete (threshold-activation) perceptron; linear separability |
| 5 | The continuous (sigmoid) perceptron and non-linearly separable patterns |
| 6 | Multilayer perceptron (MLP) and forward propagation |
| 7 | Backpropagation — the generalized delta rule |
| 8 | Factors affecting learning, generalization, and performance evaluation |
| 9 | Midterm Exam |
| 10 | Radial basis function (RBF) networks |
| 11 | Unsupervised learning; competitive learning and self-organizing maps (SOM/Kohonen) |
| 12 | Learning vector quantization (LVQ) and clustering |
| 13 | Recurrent networks; Hopfield, associative memory, and ART |
| 14 | Transition to deep networks and the fundamentals of convolutional neural networks (CNN) |
| 15 | CNN architectures (AlexNet, VGG, ResNet, YOLO) and project presentations |
| 16 | Final Exam |
Assessment
| Component | Weight |
|---|---|
| Final Exam | 40% |
| Midterm Exam | 30% |
| Project | 20% |
| Assignment | 10% |
In-semester studies contribute 60% and the final exam 40% to the success grade.
Course Learning Outcomes
Upon successful completion of this course, students will be able to:
- Explain the biologically inspired artificial neuron model and the fundamental components of a neural network (weights, summation, activation, threshold).
- Distinguish supervised, unsupervised, and reinforcement learning paradigms and the basic learning rules (Hebb, delta/Widrow–Hoff, competitive).
- Comprehend the working principles of discrete and continuous perceptrons and the multilayer perceptron; derive the backpropagation algorithm and train a network.
- Manage training factors such as over/under-fitting, generalization, learning rate, and network sizing, and evaluate model performance using appropriate metrics.
- Apply radial basis function (RBF) networks to function approximation and classification problems.
- Solve clustering and vector quantization problems using self-organizing maps (SOM), LVQ, and competitive learning.
- Explain associative memory with Hopfield networks and the stability–plasticity approach of ART.
- Comprehend the need for the transition from classical ANN to deep learning and recognize the fundamental building blocks and architectures of convolutional neural networks (CNN).
Recommended Textbooks
- Öztemel, E. Yapay Sinir Ağları. Papatya Yayıncılık.
- Haykin, S. Neural Networks and Learning Machines. Pearson.
- Course lecture notes and laboratory/practice materials.
Supplementary Resources
- Hagan, M. T., Demuth, H. B., & Beale, M. Neural Network Design.
- Fausett, L. Fundamentals of Neural Networks.
- Goodfellow, I., Bengio, Y., & Courville, A. Deep Learning (CNN chapters). MIT Press.