06

Course Title 

Quantitative approaches to behaviour

 

Course Description 

This course provides participants with a hands-on introduction to the end-to-end analysis of animal behaviour, from processing video recordings to characterizing & modelling behaviour. Participants will be introduced to the mathematical tools necessary to understand the basics of tracking (in 2 dimensions and 3 dimensions), state estimation, and characterization of behaviour on a week-to-week basis in the form of lectures. Since these topics cover many different disciplines, the focus will be put on gaining a high-level understanding of each of the topics. Furthermore, various renowned scientists will be invited to present state-of-the-art methods, research, and widely used open-source tools in computer vision, computational ethology, and behavioural sciences. These lectures serve to teach and inspire participants about what current research has to offer, as well as to demonstrate fruitful collaborations in highly interdisciplinary science.

In addition to preparing and attending the lectures, participants will select one out of two projects and work on it in pairs, which are based on active research within SCIoI. The participants are expected to apply the methodologies that they learned in the lectures to the data they received for their project.

Participants are expected to deliver a report and give a presentation at the end of the semester. The main objective of the projects is to get the participants acquainted with behavioural data and gain practical experience with the mathematical tools.

The projects are a central element of this course. Not only will participants gain hands-on experience with analysing real data, but they will furthermore learn about the research conducted within SCIoI.

 

Course Organizer

Marcus N. Boon

 

Course Format

tbd

 

Target Group

The course will be taught at the master’s level (TU, HU, and FU students are all welcome to take the course). The students will obtain 3 ECTS upon successful completion of the course. Solid knowledge in Mathematics (linear algebra, ordinary differential equations) and basic programming skills (Python) are required to attend this course. Basic knowledge in artificial neural networks is a plus, but not mandatory.

 

Course Structure

Lecture 1: Introduction to the course and the analysis of behaviour

The first lecture is used to provide the participants with an overview of the course. The lecture is furthermore used to provide a high-level overview of current research topics in behavioural science and to motivate students about interdisciplinary science. Any remaining time will be used to explain the perceptron, which will be continued in the next lecture.

 

Lecture 2: Deep learning

In this lecture the perceptron, (deep) artificial neural networks, and convolutional neural networks are introduced. Emphasis is put on their application in computer vision. Any remaining time is used to give the participant a high-level understanding of unsupervised learning methods.

 

Lecture 3: DeepLabCut (potential speaker: Mackenzie Mathis)

In the first guest lecture the participants are introduced to DeepLabCut, a widely used open-source software package used for tracking animals and/or objects in videos. We are planning to invite Mackenzie Mathis to give a talk about DeepLabCut.

 

Lecture 4: Presentation of the projects

The projects are introduced by the researchers within SCIoI that work on the topics. In the case for the first project, this will be either Katharina Hohlbaum, Niek Andresen, or myself, and for the second project this will be Ulrike Scherer or Sean Ehlman.

 

Lecture 5: Computer vision in 2D and 3D

This lecture serves to teach important key topics in computer vision for calibrating and triangulating camera video data. The lecture will be given by an “in-house” SCIoI PI (Olaf Hellwich, confirmed)

 

Lecture 6: State estimation

In this lecture the participants are introduced to the fundamentals of state estimation. The central topic of this lecture is the Kalman filter and its nonlinear variants. Similar to the lecture on computer vision, since in this lecture a key concept is explained, there will be no time to discuss current research. The lecture will therefore be given by an “in-house” PI (Henning Sprekeler, confirmed).

 

Lecture 7: Modern state-space models (potential speaker: Scott Linderman)

After the participants learned about state estimation, we delve deeper into modern state-space models. Examples from active research will be incorporated in the lecture.

 

Lecture 8: Characterizing behaviour: Key point Motion Sequencing (potential speaker: Sandeep Robert Datta)

The second guest lecture focuses on going from continuous key point data (e.g. obtained from state estimation techniques from last lectures) to behavioural modules. To this end, Sandeep Robert Datta will be invited to talk about his recent work on “Key point Motion Sequencing” (Weinreb et al. 2023).

 

Lecture 9: Temporal organization of behaviour (potential speaker: Gordon Berman)

In this lecture we focus on how complex behaviour might emerge out of stereotypical behavioural modules that are based on smaller time scales.

 

Lecture 10: Behavioural individuality (potential speaker: Benjamin de Bivort)

Individual animals with identical genomes that are placed in identical environments nevertheless exhibit behavioural differences. In this lecture the rise of this behavioural variability is discussed.

 

Lecture 11: Modelling of decision making (potential speaker: Jonathan Pillow)

The last lecture of the series closes off by discussing models for characterizing decision-making behaviour in animals.

 

Students are given 2 weeks of working on their projects. During this time, there will be no lectures but instead the students will have additional time for project questions. Presentation sessions (1-2 depending on number of participants) Each group is expected to give a short presentation of 10 to 15 minutes about their findings, followed by a round of questions of 10 minutes.