Topic outline

  • General

    Join the course slack group here: https://join.slack.com/t/aaltoaihealth-uct4660/shared_invite/zt-hctgfbbl-hhOT9j8rmCDnwQQEcx9Pwg
    Zoom link for the lectures: https://aalto.zoom.us/j/61235758006
    Zoom link for the exercises: https://aalto.zoom.us/j/65250762675

    Select a slot for your project work presentation here:

    https://docs.google.com/spreadsheets/d/1-Tj6ex8S5zSWrOIeXqzASUut63MD2LcY8PkHl7AnoTs/edit?usp=sharing 

    If all the slots are taken, send email to joel.jaskari@aalto.fi and more slots will be made.

    Learning Outcomes : 

    After the course, the student

    • can explain the fundamentals of neural networks.
    • is able to construct, design, implement, and run a deep learning project in medical domain.
    • has the skills to select and utilize the tools for applying AI for the different use cases needed for medical applications.
    • is able to identify and develop the data sets that provide the best results.

    Content : 

    The course will contain the basics in deep learning, introduce the student how machine learning is used in different medical domains, and what technologies are appropriate in different use cases. The work has mathematical and programming exercises, an exercise, and a project for designing and teaching a neural network for a medical use case.

    Lectures Mondays and Exercises on Fridays 8.15 on Zoom.

    • Lecture 1: What is currently possible with AI?
    • Lecture 2: Introduction to probability. Basics and what is needed for the course.
    • Lecture 3: Machine learning. Unsupervised, self-supervised, supervised, and reinforcement learning.
    • Lecture 4: Neural networks and medical data, Training of networks back-propagation, stochastic gradient descent.
    • Lecture 5: Models of data with spatial and temporal symmetries. Convolutional neural networks (CNNs). Medical imaging, object classification, data augmentation, recurrent neural networks. Physiological signals, like ECG and BCG. Modeling the cardiovascular system.
    • Lecture 6: Autoencoders, predicting the future, and de-noising signals. Generative adversarial networks (GANs)
    • Lecture 7: Learning with less data. Transfer learning, synthetic data. Legislation and ethics in use of AI.
    • Lecture 8: Natural Language Processing. Attention models. GPT-3. Generating text
    • Lecture 9: AI in embedded systems. Compressing deep neural networks. HW for AI.
    • Lecture 10: Challenges in using AI in health. Adversarial examples. Differential privacy
    • Lecture 11: What could we expect of the future?
    • Lectures 12 - 13 and Exercise 13: Project work reports
    • Lecture 14: Exam

    Grading :

    The course will be graded by Exercises (20%), Exam (40%), and Project work (40%).
    If you wish to do the extended Project work (graded as additional 5cr course), the course will be graded by Exercises (34%) and Exam (66%).