05

Course Title 

Introduction to Modelling Collective Behaviour

 

Course Description 

The goal of this course is to provide an overview of collective behaviour- and intelligence research. Titled “Introduction to Modelling Collective Behaviour” the seminar series invites internationally renowned speakers from the field of Biology, Computational Sciences, Modelling, Robotics and Ethics keeping collective behaviour in focus. The speaker series is accompanied by (1) an agent-based modelling workshop and (2) a one-day ethics seminar allowing participants to discuss ethical questions arising specifically from collective intelligence research. Lectures are prepared with (3) a journal club series to facilitate fruitful discussion between participants and invited speakers. Furthermore, to help the integration of existing and new research projects within the cluster, participants will be shortly introduced to (4) ongoing research on collective behaviour within our cluster.

 

The seminar series focuses on modelling embodied collective intelligence through the lens of natural- and computational sciences but also introduces artificial collective systems and complements the course “Introduction to swarm robotics” (in preparation by Mohsen Raoufi). The series starts with examples of collective behaviour in natural groups (such as social insects, birds or fish). Invited speakers from the field of natural sciences are encouraged to keep focus on three main aspects: (1) the evolutionary benefits of collective behaviour in nature, (2) the role of embodiment and physical constraints in collective behaviour and (3) the role of perception in widely observed collective behavioural phenomena. These aspects have a fundamental role in synthesizing collective behaviour, that is one of the main motivations of the Science of Intelligence Cluster. The course then shifts towards computational and agent-based modelling. Seminars in this block will cover essential understanding on capturing the main properties of observed collective behaviour through mathematical and computational models. Invited speakers will present various approaches and will elaborate on the types and taxonomy of computational models used to describe and predict collective behaviour, including bio-inspired and data-driven models.

The series will be continued with invited speakers from the field of bio-inspired artificial systems. Here, we encourage participants to look at robot platforms as realistic, embodied models of collective behaviour where – in contrast with computational models – constraints of a physical world cannot be lifted. Therefore, such model platforms can give key insights about the role of physical constraints on collective behaviour.

The lecture series will be closed with an ethical discussion, where participants will be encouraged to explore the ethical concerns of synthesizing intelligent swarms with an expert invited to guide the discussion. To facilitate scientific discussion between invited speakers and participants, lectures will be prepared with corresponding journal club events where the main contributions of distinguished speakers will be shortly discussed through selected publications.

At the end of the semester, participants will have the chance to implement one of the models during a practical modelling workshop using an agent-based framework (see GitHub). This has been previously proved as a useful educational material for the modelling and programming block course of the Institute of Theoretical Biology of the Humboldt Universität zu Berlin in two consecutive years (SoSe23 and SoSe22) with success. The framework is also used for active research at the Humboldt Universität zu Berlin and at the Science of Intelligence Cluster in project 34 (also see Videos), and project 51 and in an ongoing PhD and Master’s project. To present collective intelligence research at the cluster, internal speakers will be also invited (next to external speakers). Participants will be shortly introduced to the Kilobot and Thymio frameworks used to model embodied collective behaviour, the CoBe augmented reality system, and the OptiTrack motion capture system through short live demonstrations (“demos”).

Furthermore, participants will learn about current modelling and analytic approaches of collective behaviour within the cluster through short presentations. With this course, I believe, participants will gain a broad overview and enlightened perspectives in the field of collective behaviour and intelligence research. By systematically exploring example behaviours from biology, modelling, and robotics through captivating talks of renowned speakers, participants will map the strongly interconnected field of collective behaviour research keeping bio-inspired model systems and modelling natural, embodied collective intelligence in focus.

 

Course Organizer

David Mezey

 

Course Format 

The lecture format is hybrid and presentations will be recorded and provided as educational material for registered students.

 

Target Group 

The lecture series will be announced through several networks and academic mailing lists (HU, TU, MPI, BCCN, ECDF, HSMB) and some lectures will be open for the general public interested in research on collective intelligence (e.g. also announced via social media, such as X/Twitter, LinkedIn, Reddit, etc.). If technically possible, it will be further announced for national and international students from similar fields the same way. For that, possible mailing lists and networks will be initially collected with the help of the PIs. The target audience is anyone with an interest in modelling collective behaviour from the field of computational biology, neuroscience, bioinformatics, mathematical and physical sciences and engineering or equivalent. Prerequisite for the practical part is basic programming knowledge in python. (If technically possible, 3SWS could be provided for the pure theoretical track, and 4SWS for the overall course to allow participants with no programming background in python to participate.)

 

Course Structure 

Invited Speakers: 34 / 56 Units (60%)

  • 10 units on biological collective intelligence (3-5 Invited Speakers)
  • 12 units on modelling collective behaviour (4-6 Invited Speakers)
  • 6 units on bio-inspired artificial systems (2-3 Invited Speakers)
  • 6 units on ethical considerations (1 Invited Speaker, block)

 

Practice: 22 / 56 Units (40%)

  • 6 units for journal club (3 sessions before distinguished speakers)
  • 6 units for internal presentations and demos (block)
  • 10 units for agent-based modelling workshop (block)

 

Biological collective intelligence (10 units)

Observing collective behaviour in natural collectives gives the starting point of the seminar series. Participants will learn about different types of collective behaviour (collective movement, problem solving, evasion, migration) from different natural systems (locusts, ants, honey bees, birds, fish and humans) and corresponding mathematical or computational models through the work of internationally renowned speakers. During this part of the series, we focus on the evolutionary benefits of collective intelligence, the role of embodiment and environmental factors shaping collective behaviour, and also raise the question “Is collective behaviour always adaptive?”. The block will provide key insights and prepare the next part of the seminar series about modelling approaches of collective behaviour.

 

Modelling Collective Behaviour (12 units)

In this block of the seminar, the focus is shifted to different approaches of modelling collective behaviour. From inspirations of statistical physics to bio-inspired and data driven modelling, participants will see the state-of-the-art of capturing core elements of natural systems into tangible computational models to study collective behaviour. During this block we encourage invited speakers to (1) give a general overview about the history, motivation of modelling collective behaviour, (2) the types and taxonomy of models used to gain insights on collective behaviour, and (3) the role of the environment and perception in collective behaviour and how to model such constraints. By that, we set the stage for introducing bio-inspired physical modelling approaches using swarm robotics.

 

Bio-inspired Artificial Systems (6 units)

In the last block of the seminar series participants will get a short, high-level introduction into bio-inspired artificial systems. In the framing of this course, we emphasize two fundamentally different goals when designing collective artificial systems. (1) On one hand, how research in biology might influence and inspire approaches to engineering intelligent swarms (bio-inspired design), (2) on the other hand, how designing such applications might extend our knowledge about collective behaviour observed in nature. In the latter case, robot platforms serve as embodied models of collective behaviour where physical, perceptual and cognitive constraints cannot be lifted. Therefore, robot platforms can help us understand the effect of such constraints on the natural systems that inspired the application.

 

Agent-based Modelling Workshop (10 units)

During the workshop participants have the chance to learn more about agent-based modelling by implementing (in a guided environment) a simple model of collective behaviour such as (1) a zonal collective movement model or (2) a spatially explicit SIR model combined with agent movement. The provided framework allows participants to interact with their models via pygame, a widely used game-engine. Participants are encouraged to interactively discover the collective behaviour arising from the simple rules they implement. Finally, an assignment will be announced for the final evaluation.

 

Ethical Framing (6 units)

To provide participants the possibility of discussing the ethical questions of artificial collective systems specifically, we plan to invite an expert on the field of AI Ethics to (1) give a lecture about the ethics of engineered collective intelligence and (2) guide the further discussion between participants. Although the literature on the topic is rather scarce, some of the main points of this discussion could be inspired by this article by Edmund R. Hunt and Sabine Hauert (both on possible invited speakers list). In case the authors accept our invitation as well, next to their lectures, a panel discussion will be organized together with the invited expert from the field of AI ethics and possible other internal experts.

 

Internal Demos and Presentations (6 units)

To give participants an overview on collective intelligence research at our cluster, the semester ends with exciting presentations and live demonstrations in the framing of a SCIoI demo day (preferably within the same week as the agent-based modelling and/or ethics discussion). In the morning, presentations of PhD and PostDoc level researchers will set the stage for larger scale demonstrations using the Kilobot and Thymio frameworks together with the OptiTrack motion tracker and the CoBe system. Presentations will be ca. 15-30 minutes long and take 4 units. Demos take 2 units together.

 

Final Assignment(s) (ca. 20 units, not in presence)

The final assignment has two parts. First, in the theoretical part, participants will choose one lecture (that the participant found the most interesting) from the semester and write a short essay (max. 2 pages long) about the (1) general questions the lecture raises (2) the modelling approach used during the lecture and (3) the findings and results achieved through that specific modelling approach. As for the practical part, participants have to answer a set of predefined questions about the previously implemented agent-based model in the form of a short report (max. 2 pages without figures) following the main structure of a scientific paper. In both cases, if participants use large language-models to finish their assignments a supplementary note is to be written with the exact type and usage of the language model.