Sign in to follow this  
Palidine

New data on Flocking of Starlings

Recommended Posts

Cool article for those of us who play around with flocking behaviors: http://www.telegraph.co.uk/earth/main.jhtml;jsessionid=X2TAPSYKVGGJ5QFIQMFSFFWAVCBQ0IV0?view=DETAILS&grid=&xml=/earth/2008/01/29/scistarling129.xml The key point:
Quote:
Current computer models assume that each bird interacts with all birds within a certain distance. But the new observations, however, show that each bird keeps under control a fixed number of neighbours - seven other starlings - irrespective of their distance, which is the secret of how they stick together.
I'd love to read the actual paper that this article was based on. Anyone with uni access have a citation? -me

Share this post


Link to post
Share on other sites
I'm in the midst of searching up references for my own thesis, so I can oblige by tracking down that paper for you [smile].

I think this is the paper you're looking for: Interaction Ruling Animal Collective Behaviour Depends on Topological rather than Metric Distance: Evidence from a Field Study (PDF link)

Abstract:
Quote:
Numerical models indicate that collective animal behaviour may emerge from
simple local rules of interaction among the individuals. However, very little is
known about the nature of such interaction, so that models and theories mostly
rely on aprioristic assumptions. By reconstructing the three-dimensional position
of individual birds in airborne flocks of few thousands members, we prove that the
interaction does not depend on the metric distance, as most current models and
theories assume, but rather on the topological distance. In fact, we discover that
each bird interacts on average with a fixed number of neighbours (six-seven),
rather than with all neighbours within a fixed metric distance. We argue that a
topological interaction is indispensable to maintain flock’s cohesion against the
large density changes caused by external perturbations, typically predation. We
support this hypothesis by numerical simulations, showing that a topological
interaction grants significantly higher cohesion of the aggregation compared to a
standard metric one.

Share this post


Link to post
Share on other sites
That is very interesting if each bird is only looking at approximately 6 or 7 others in the flock of thousands. Assuming that their choices of other birds to look at is random, then the internal "interaction/follow" topology may not be fully connected and the flock may actually consist of multiple "topological" sub-flocks. This inference is based on the paper "The Number of Neighbors Needed for Connectivity of Wireless Networks" where the authors found that to guarantee connectivity in a randomly deployed sensor network, each node needs to be connected to at least 5.1774*(log n) neighboring nodes (where n is the total number of nodes deployed). So, in the case of a random flock of 1000 birds, to guarantee the internal "interaction/follow" topology is fully connected, each bird would really have to look at at least 15 others instead of 6 or 7.

Share this post


Link to post
Share on other sites
Have you noticed that flocks often do break apart into sub-flocks and reform, particularly when external influences are present (look at the way sheep respond to a sheep dog, or how small fish respond to larger predators as examples). I think the point is that nature wants there to be only a weak coupling throughout the flock so that it can break apart when needed, but reform when appropriate, irrespective of the distance. If, when breaking apart, all birds/fish/sheep went their own way (i.e., they ignored their rules governing interactions) each individual would be at higher risk. Having a flock break into two when a predator comes calling guarantees survival for at least one group!

Anyway, a very interesting article... 8)

Cheers,

Timkin

Share this post


Link to post
Share on other sites
The other limiting factor (or related anyway) is the "digit span" of birds. Like humans, other mammals can only focus on a limited number of things at a time. For humans that's about 7-10, birds is ~7, fish is 3-5. The authors note this in their article as one of the things that led them towards and/or reinforced their hypothesis (don't quite remember/too lazy to find the quote).

This model finds that the birds only cared about ~7 other birds; other cognitive studies have shown that a bird can only focus on ~7 object at a time. That's a very suggestive parallel (and perhaps obvious in retrospect [smile])

The one thing I couldn't get from the paper was whether the 7 chosen links were permanent or whether they shifted through time. i.e. am I just trying to stay with my friends, or am I trying to stay with any random 7 birds of the moment? Staying with a fixed 7 might suggest interesting things about their social structure; certainly in a human mob, you try to stay with your buddies as you move with the herd.

-me

Share this post


Link to post
Share on other sites
Quote:
Original post by WeirdoFu
This inference is based on the paper "The Number of Neighbors Needed for Connectivity of Wireless Networks" where the authors found that to guarantee connectivity in a randomly deployed sensor network, each node needs to be connected to at least 5.1774*(log n) neighboring nodes (where n is the total number of nodes deployed). So, in the case of a random flock of 1000 birds, to guarantee the internal "interaction/follow" topology is fully connected, each bird would really have to look at at least 15 others instead of 6 or 7.


The value of 5.1774 is what the authors of that paper rigorously prove will result in connectivity, but they also provide simulation results that suggest values above 1.5 are sufficient. Following just 6 neighbors should be sufficient for even a flock of up to 10,000 individuals, to keep the flock topologically connected.

Share this post


Link to post
Share on other sites
Quote:
Original post by Vorpy
Quote:
Original post by WeirdoFu
This inference is based on the paper "The Number of Neighbors Needed for Connectivity of Wireless Networks" where the authors found that to guarantee connectivity in a randomly deployed sensor network, each node needs to be connected to at least 5.1774*(log n) neighboring nodes (where n is the total number of nodes deployed). So, in the case of a random flock of 1000 birds, to guarantee the internal "interaction/follow" topology is fully connected, each bird would really have to look at at least 15 others instead of 6 or 7.


The value of 5.1774 is what the authors of that paper rigorously prove will result in connectivity, but they also provide simulation results that suggest values above 1.5 are sufficient. Following just 6 neighbors should be sufficient for even a flock of up to 10,000 individuals, to keep the flock topologically connected.


That is true, which would further reinforce why birds only need to look at ~7 neighbors, as it is sufficent for their purpose.

Share this post


Link to post
Share on other sites
Quote:
Original post by Palidine
The one thing I couldn't get from the paper was whether the 7 chosen links were permanent or whether they shifted through time. i.e. am I just trying to stay with my friends, or am I trying to stay with any random 7 birds of the moment?


Apart from bird friendship, changing the tracked neighbours from time to time would have a benefical effect from a flocking point of view: fluctuations like the random segregation suggested by WeirdoFu wouldn't last enough to affect the flock seriously.
It seems plausible for birds to elect more suitably placed buddies to follow when those they are tracking are too far or too near or too clustered; they might also lose track of their buddies spontaneously, e.g. when another bird occludes them.

Share this post


Link to post
Share on other sites
Naturally, we are all used to thinking of sticking together with the nearest neighbors. Since the quote was "irrespective of distance", I can't imagine what other criteria that they would be using other than a "starling-friend" one.

Share this post


Link to post
Share on other sites
The paper is talking about nearest neighbors. The "irrespective of distance" part just means that the density of the flock doesn't matter. Their behavior appears to be affected by their nearest neighbors even if the members of the flock are spread out or clustered together. The unanswered question is whether or not their nearest neighbors tend to remain the same birds over time, or if they are interchangable with other birds.

Share this post


Link to post
Share on other sites
i don't think this is anything strikingly new. Most flocking algorithms use a population-average for determining the individual variables. But i did see one demo that uses a nearest neighbor approach to greater effect. Can't remember where it was though.

Seems obvious when you think about it though. Most people are not influenced by the whole world as much as they are influenced by their friends.

The problem is that implementing a nearest neighbor / visual range approach to flocking is that it is CPU intensive with many boids. If anyone has a cute way to cut that down, i'd be interested. I've got a long range project that involves this very problem.

Share this post


Link to post
Share on other sites
Quote:
Original post by leiavoia
i don't think this is anything strikingly new. Most flocking algorithms use a population-average for determining the individual variables. But i did see one demo that uses a nearest neighbor approach to greater effect. Can't remember where it was though.


The 'newness' is that they did not start out with a k-nearest neighbour algorithm and show that this leads to flocking. They took real images of flocks of starlings, inferred the 3D positions and velocities of birds in the flock and then analysed this data to determine a likely hypothesis to explain the data. The result was that they believe the birds to be using a nearest neighbour method to maintain the flock. That's not to say this IS what the birds are doing... only that it is more likely than using metric distance. Certainly nearest neighbour algorithms are nothing new, nor their use on evaluating local behaviours in multi-agent systems. This is a paper linking the toy problems back to nature and in that it is important and novel.

As for the 'bird-friend' question... I doubt that it would be necessary... but it is possible. The old addage of '6 degrees of separation' has been widely studied across a range of disciplines and it always leads back to a basic principle of connectionism in (biological/social/engineered) networks that has been well studied (and certainly well before the results in the wireless network research mentioned previously... but then engineers are well known for not reading material from other fields before they publish their own results! ;) ). My personal feeling is that there would be too many times that a bird would lose track of several of its friends within the flock making it difficult for a bird to evaluate its position and velocity relative to the flock. If a bird brain could only keep track of 7 or so objects then it makes sense that it keeps track of the nearest 7 birds, rather than working out if the nearest 7 birds are also its friends and then having to assess more candidate neighbours. That seems like wasted computation to me and nature is very good and weeding that out.

On a slightly different but related note I'm interested by the result of 7 objects. Research has shown that humans instinctively understand up to 5 distinct objects (in other words, we can automatically detect how many objects are in a group if the number is 5 or less). We can obviously keep track of more than this, but we use higher brain functions to do it, rather than low level automatically encoded recognition techniques. If birds can recognise more instinctively (because I don't believe a bird would want to waste energy using higher brain functions for basic flight navigation), is that a result of them living in flocks that lead them to evolve more specialised brain functionality in this area? Is 5 an important number in terms of human evolution of social structures? Did we evolve to recognise 5 items because of our social behaviours and capabilities, or did these evolve because we can recognise 5 objects (and if so, how would society differ if we could only recognise 2, or 7)? Food for thought! ;)

Cheers,

Timkin

Share this post


Link to post
Share on other sites
I think that you can not equate the idea of "recognition" of objects and simply using them to keep track of spacing, etc. If we are in a crowd of people, for example, we can instinctively keep our alignment and separation from the people directly around us - which may or may not have anything to do with recognizing a specific 5.

Also, keep in mind that using 7 overall may only entail 3 or 4 or 5 in any one discrete time slice. If part of the "flocking conciousness" portion of the starling, however, involves checking various combinations of neighbors (different sides, for example), those instantaneous 3 or 4 or 5 may overlap to become an average of 7 over any given short interval. There are lots of different ways that the numbers can match up but short of having the Starlings fill out exit polls after a group event, we may never be able to discern the full process.

Share this post


Link to post
Share on other sites
I was just doing some thinking this morning about this issue. If the Starlings were to use standard flocking, they wouldn't string out the way they do at times. Standard flocking mentality tries desperately to pull everyone into a circular/spherical blob since they are trying to move to the center of mass of the entire group. Starlings, however (and, ducks, geese, etc.) will string out somewhat into more of a ribbon effect at times creating quite a procession. That is significantly different and is very easily the product of them only tracking a limited number of targets.



Video of starlings

(The video is amusing since there are so many of them they actually bend a cedar tree.)

Share this post


Link to post
Share on other sites
Quote:
Original post by InnocuousFox
I think that you can not equate the idea of "recognition" of objects and simply using them to keep track of spacing, etc.


I disagree on that point. An animal would want to be able to recognise an object to be able to implement the correct behavioural response to it. If it's another of the same species the reaction would be different to that of a predator. Certainly this doesn't mean that it needs to be able to identify individuals... and perhaps that's what you meant!?

Quote:
If we are in a crowd of people, for example, we can instinctively keep our alignment and separation from the people directly around us - which may or may not have anything to do with recognizing a specific 5.


There's been some interesting research in this area relating movement in crowds to thermodynamic principles. It's quite likely that our local interactions are guided toward a local optima for flow velocity in a positive feedback loop... rather than having anything to do with the fact that it's other people we are trying to 'flock' with.

Quote:

Also, keep in mind that using 7 overall may only entail 3 or 4 or 5 in any one discrete time slice.

Quite possibly... although neural systems don't work on time slices (only our imaging systems do). It's certainly possible though that there is some computation spread over a temporal cycle involving polling of individuals... but what reason would you give for the bird then choosing 7 others to poll? Perhaps minimise cycle time while maximising connectivity?

Quote:
There are lots of different ways that the numbers can match up but short of having the Starlings fill out exit polls after a group event, we may never be able to discern the full process.


True... but it might be fun to try and find out.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this