Deep Learning Fundamentals (en)
The course ‘Deep Learning Fundamentals’ offers an in-depth immersion into the field of deep learning, with a particular emphasis on the practical and experimental aspect. During the course, students will have the opportunity to explore fundamental concepts of neurons, neural layers, and neural network architectures, as well as learn techniques for designing, training, and optimizing both simple and complex neural networks. Through a combination of instructional labs, each student will be able to work to complete training exercises that will provide practical experience in using the tool for each of the topics covered during the course.
CODE: DSAI300
Category: Artificial Intelligence
Teaching methodology
The course includes educational laboratories in which each student will be able to work in order to complete training exercises that will provide practical experience in using the instrument, for each of the topics covered during the course.
Prerequisites
- Basic knowledge of mathematics, particularly linear algebra and differential calculus.
- Understanding of computer science fundamentals and programming.
- Familiarity with Python and its syntax.
- Understanding of fundamental concepts of artificial intelligence and machine learning.
The following is an overview of course content:
Neurons and Layers: This section delves into the fundamental building blocks of neural networks, covering the concepts of neurons and layers and how they contribute to the overall architecture of deep learning models.
Simple Neural Network: Participants will learn to construct and implement a basic neural network, gaining hands-on experience in building and training simple models for various tasks.
Coffee Roasting: The course introduces the concept of coffee roasting as a practical analogy for understanding the optimization process in neural networks, providing insights into the iterative nature of model training.
MNIST: MNIST, a popular dataset of handwritten digits, serves as a common benchmark for exploring classification tasks in deep learning. Participants will work with MNIST to develop skills in image classification using neural networks.
ReLU Activation: ReLU (Rectified Linear Unit) activation function is a key element in modern neural network architectures. This section covers the theory and application of ReLU activation functions in deep learning models.
Softmax: Softmax activation function is essential for multiclass classification tasks. Students will learn about Softmax and its role in producing probability distributions over multiple classes in neural network outputs.
Multiclass: Building upon earlier concepts, this section focuses on handling multiclass classification problems, where models must classify input data into more than two classes.
Evaluation: Participants will explore various techniques for evaluating the performance of deep learning models, including metrics such as accuracy, precision, recall, and F1-score.
Bias: The course concludes with a discussion on bias in deep learning models, addressing strategies for identifying and mitigating biases to ensure fair and equitable AI systems.
At the end of the course, participants will be able to:
- Understand concepts of neurons and neural layers.
- Create and implement a simple neural network.
- Apply machine learning in the context of coffee roasting.
- Use the MNIST dataset for classification.
- Understand and apply ReLU and Softmax activation.
- Handle multiclass problems.
- Evaluate performance of deep learning models.
- Understand and manage bias in deep learning models.
Duration – 1 day
Delivery – in Classroom, On Site, Remote
PC and SW requirements:
- Internet connection
- Web browser, Google Chrome
- Zoom
Language
- Instructor: English
- Workshops: English
- Slides: English
