|
Ant colonies organize their foraging as well as other activities using indirect communication by laying pheromone trails (stygmergy) or using direct communication between individuals (invitation behavior). Mass recruitment by trail laying species has been widely studied (see, for example, this paper), is reasonably well understood, and has served has served as an inspiration for the solution of various computational problems (e.g. ant colony optimization). A excellent overview of this is given in Self-Organization in Biological Systems by Scott Camazine et al. (eds.) and in Information Processing in Social Insects by Claire Detrain et al. (eds.).
Recent studies suggest that recruitment by invitation behavior results in different properties on the colony level, potentially allowing the colony to adapt more quickly to changing environments. The project aims to understand (some of) the differences between these two forms of organization. We compare the two behaviors by performing parallel experiments with two different species in the lab (you can actively participate) and attempt to model the behavior with simulation models and/or analytic dynamic system models. |
|
Recent studies have shown that ant colonies, somewhat counter-intuitively, use "noisy" communication to their advantage (paper). The studies suggest that noise allows self-organized systems generally, and ant colonies in particular, to adapt flexibly in changing environments. This can best be explained using an information-theoretic approach to quantify how quickly information about the environment spreads through a colony. We will study this phenomenon by developing simulations of known biological experiments that allow us to directly measure the information throughput. We will extend these simulations to predict results of yet-to-be-conducted experiments. |
|
Recently it has been discovered that the True Slime Mold Physarum Polycephalum, an amoeba-like unicellular organism, is "smart enough" to solve the shortest path problem in mazes (news article). We are interested in whether slime molds can also flexibly adapt when the maze changes. We will explore this by building biologically realistic simulations of their behavior. |
A large number of biological systems can only be understood through computer simulations. Prototypical examples in popular science are insect swarms and ant colonies, but the same holds true for many other biological phenomena. For example, the progression of a disease in a host tissue (cell invasion), the propagation of viruses, or the formation of bacteria films can often only be predicted using simulations. From the perspective of computational modelling, all these systems are similar: a large number of individual elements move, proliferate or change state according to simple rules that are only based on local information. While simulations of such systems are often conceptually simple, they put extreme strain on computational resources. It is not unusual for a single simulation run to take many hours or even days and often thousands of runs are required. This currently renders many interesting simulation experiments infeasible in practice. GPU computing promises a solution to this problem. Graphics procesesing units (GPUs) have special purpose architectures that perform computer graphics operations in parallel at extremely high speed. Somewhat surprisingly, the parallel processing capabilities of GPUs can also be used to speed up many types of scientific computations by several orders of mangnitude (at extremely low cost!). The project will explore such parallelizations for common simulation algorithms, such as the Gillespie method, and use these algorithms to build simulations for selected problems in swarm behaviour and tissue modelling.
The Causal Robotics Group and the Natural Computation Lab are building a swarm of micro robots that will eventually live in building 63. This is an open-ended long term project. Its aim is to experimentally explore self-organizing "intelligent" behaviour as we find it, for example, in ant colonies and swarms of bees. This is part of an ARC project investigating the principles of self-organized collective decision making, often somewhat more spectacularly termed Swarm Intelligence. The swarm will initially comprise about 20 micro robots that are customized from extensible off-the shelf units, specifically iRobots and/or ePucks. There is a variety of different honours projects available within this long-term initiative, including the development of control algorithms, simulations, and specialized sensor and actuator software/hardware. Talk to me about details if you are interested. You must be a competent programmer and in some cases an inclination to tinker with hardware, maybe even knowing how to use a soldering iron would help!
The discovery that the True Slime Mold Physarum Polycephalum, an amoeba-like unicellular organism, can solve the shortest path problem in mazes (news article) has recently served as an inspiration for a new shortest path algorithm, termed Physarum Solver (paper). While this algorithm is theoretically interesting, its computational complexity limits its practical usefulness. Unlike other nature-inspired algorithms, such as Ant Colony Optimization or Genetic Algorithms, Physarum Solver does not use stochastic sampling to reduce the problem space. We will explore whether it is possible to extend Physarum Solver in this form and how this compares to other nature-inspired search methods.
A classical mathematical problem is the "Stable Marriage" (see also this survey paper). The challenge is a to find a pairing of a group of men and women into couples such that the overall stability (happiness?) of the group based on individual preferences for partners is in some sense optimized. We will attempt to solve this problem with a self-organizing algorithm, and we are specifically interested in the question whether this enables the group to reorganize optimally when new partners become available (or existing ones become unavailable).