2024
Neural networks and deep learning
Name: Neural networks and deep learning
Code: INF13267M
6 ECTS
Duration: 15 weeks/156 hours
Scientific Area:
Informatics
Teaching languages: Portuguese, English
Languages of tutoring support: Portuguese, English
Regime de Frequência: Presencial
Sustainable Development Goals
Learning Goals
At the end of the course unit the student should demonstrate:
Broad knowledge of neural network approaches and techniques and deep learning, both in basic architectures and in those designed for specific problem types
Deeper understanding of the problems and possible approaches associated with the development of solutions based on neural networks
Knowledge about the design and programming techniques of these architectures and their evaluation
Knowledge about basic techniques needed to conduct deep learning research and big data
Broad knowledge of neural network approaches and techniques and deep learning, both in basic architectures and in those designed for specific problem types
Deeper understanding of the problems and possible approaches associated with the development of solutions based on neural networks
Knowledge about the design and programming techniques of these architectures and their evaluation
Knowledge about basic techniques needed to conduct deep learning research and big data
Contents
Basic concepts
basic architecture
perceptron
multi-layer networks
Activation and loss functions
Network training: backpropagation algorithm
practical issues
overfitting, vanishing, convergence difficulties
Feed-forward networks, recurring networks
Deep learning architectures:
cnn, lstm, transformers
Model explainability: intrinsic vs post-hoc, model specific or agnostic, global vs. local
basic architecture
perceptron
multi-layer networks
Activation and loss functions
Network training: backpropagation algorithm
practical issues
overfitting, vanishing, convergence difficulties
Feed-forward networks, recurring networks
Deep learning architectures:
cnn, lstm, transformers
Model explainability: intrinsic vs post-hoc, model specific or agnostic, global vs. local
Teaching Methods
Theoretical classes with introduction of concepts, accompanied resolution of exercises and clarification of doubts.
Practical laboratory classes with proposal of problems that accompany the theoretical material and clarification of doubts during their resolution. Exercises, of gradual difficulty, covering the topics taught, for students to practice the subjects.
Practical laboratory classes with proposal of problems that accompany the theoretical material and clarification of doubts during their resolution. Exercises, of gradual difficulty, covering the topics taught, for students to practice the subjects.
Assessment
Continuous assessment - consisting of 2 components:
* individual work on a specific topic in article format (30%)
* practical group work (70%)
Final assessment - consisting of 2 components:
* article (30%)
* report on group practice (70%)
The final grade is obtained through the weighted average of the 2 components. The student is approved if the final grade is equal to or greater than 10.
* individual work on a specific topic in article format (30%)
* practical group work (70%)
Final assessment - consisting of 2 components:
* article (30%)
* report on group practice (70%)
The final grade is obtained through the weighted average of the 2 components. The student is approved if the final grade is equal to or greater than 10.
Teaching Staff
- Teresa Cristina de Freitas Gonçalves [responsible]