General Information
| Course Code | XM_0102 |
|---|---|
| Credits | 6 EC |
| Period | P2 |
| Course Level | 500 |
| Language of Tuition | English |
| Faculty | Faculty of Science |
| Course Coordinator | prof. dr. A.C.M. ten Teije |
| Examiner | prof. dr. A.C.M. ten Teije |
| Teaching Staff |
prof. dr. A.C.M. ten Teije prof. dr. M. Hoogendoorn |
|
Part of programme(s)
Master
Use the arrow keys to navigate the list, Enter to select, and Escape to close.
|
|
Practical Information
You need to register for this course yourself. The (de)registration deadlines can be found on VU.nl
| Teaching Methods | Practical, Lecture |
|---|
Target audiences
This course is also available as:
Course Objective
The main aim of this course is to make the student familiar with the
combination of machine learning and reasoning techniques applied in
healthcare. We will focus on machine learning with prior knowledge
(“informed machine learning”), machine learning with symbolic output,
and explainable learning systems. We apply each approach to some of the
important tasks in healthcare, namely diagnosis, prediction, treatment
and prevention using patient data (electronic health record data or
trial data).
The course is focused on the following Dublin descriptors:
Knowledge and understanding: learning new relevant learning and
reasoning techniques.
- Applying Knowledge and Understanding: the aim is to apply algorithms
in
practice in a project.
- Making judgements: deciding which domain knowledge and learning and
reasoning techniques are appropriate.
- Communication skills: how to report on your approach, choices and your
results
Course Content
Machine learning with prior knowledge: The input of informed machinelearning consists of usual training data and additional prior knowledge.
The prior knowledge is independently of the learning task and can be in
the form of logic rules, simulation results, knowledge graphs, etc.
Topics:
Prediction with machine learning techniques (i.e., CART, LR, RF)
enriched with domain knowledge (ontologies). For example predictive
modeling of colorectal cancer using a dedicated pre-processing pipeline
on routine electronic medical records.
Reinforcement learning with initial policies based on domain knowledge
and/or constraints. Reinforcement learning can be used to select
sequences of interventions. This should however be done safely, and one
should not ignore existing knowledge on suitable intervention
strategies. For example, limiting the actions space per situation based
on rules and regulations and starting from the current treatments
regimes, and refining them based on data.
Machine learning with symbolic output: machine learning techniques can
be applied to symbolic structures (eg. Electronic Health Record) that
results in symbolic output (e.g. discovery of possible causes of rare
diseases, or simple guidelines).
Topics:
inductive logic programming (application eg. discovering possible causes
of diseases)
probabilistic logic (application in diagnosis)
Explainable learning systems: Machine learning algorithms are usually
black boxes, however physicians would like an explanation for the
decisions of machine learning algorithms.
Topics:
SHAPE and LIME algorithms for providing an explanation. Those algorithms
can be used for instance for explaining predictions of breast cancer
survival.
using background knowledge for deductively reconstructing an explanation
for the results of machine learning algorithms.
Project: exploiting a combination of reasoning and learning in health
care along the lines of the above themes (machine learning with prior
knowledge, machine learning with symbolic output, explainable learning
systems). For instance using several machine learning techniques and
knowledge sources to predict the occurrence of a particular disease
based on medical records.
Additional Information Teaching Methods
Lecture series, plus workgroup sessions for the assignment. In the firstpart of the period the emphasis is on the theory (lectures), the second
part of the period is a practical assignment in the form of a project.
The course ends with an exam.
Furthermore we aim to have one or two guest lectures by medical experts
(physicians).
Method of Assessment
Written exam (E) (50%) and practical assignment (A) (50%). For bothparts the grade needs to be sufficient to obtain a final grade.
No resist is possible for the practical assignment.
Literature
Collection of research papers.Additional Information Target Audience
Master students Artificial Intelligence (track AI for Health)
Master students Medical Informatics (UvA) (track AI for Health)
Custom Course Registration
Registration for medical informatics (UvA) students: Please check the
this document and then register for secondary courses via https://vu.nl/en/student/information-about-registration-and-enrolment