Tuesday, October 6, 2020

What is an Agent-Based Model, and how does it help us understand a pandemic?

Agent-based models (ABMs) are computer simulations capable of accounting for differences in individual (human or other organism) attributes. They can be used to predict how circumstances involving many individuals interacting with one another and their environment might unfold under a range of scenarios. Questions an ABM simulation might help us to answer include, How would motor traffic operate if we add a new road from one city to another?; How would people vote in an election if we spread misinformation over online social networks?; How will this virus spread through our city? 

How do these simulations work? What makes them capable of prediction, and what are their limitations? How are they applied to understand a pandemic?

For clarity, I'll place ABMs into the specific realm of social and human behavioural simulation. Each individual (real) person has unique cultural, behavioural, physiological, psychological and physical attributes. These differences all impact the way we make decisions, and the way we interact with the world around us. ABMs take these individual differences into account by explicitly representing an "individual" and the historical conditions it has experienced, as well as the local conditions it is currently experiencing. Here's a simple example... if I fell off my bicycle and my twin sister did not, I may develop a fear of cycling that my sister does not develop, even though our upbringing, genetics and environment may be very similar. I would then be expected to make different decisions to my sister following the bicycle accident with regard to my assessment of cycling safety. This would impact my future interactions with the world around me in various circumstances. An ABM would explicitly model these different circumstances within an individual "agent" that forms part of a simulation containing many such agents that interact with one another and their environment. The simulation is in effect a complete "virtual world" full of independent agents that make decisions based on their past, their present situation, and their goals for the future.

One way to think about an ABM is as a large computer game, such as Pacman, with thousands of "ghosts" and no player-controlled Pacman character. Each ghost is an "agent" in the software that has its own position in the world, and its own goals, direction of travel, colour and history of interactions, relationships, movements and experiences. Each ghost moves around the virtual world meeting other ghosts, making decisions about what to do when it encounters another, or deciding which way to turn at a junction in the road. The observer just watches the game unfold but can also establish and alter the conditions under which the virtual world operates. They might state explicitly key attributes of the simulation in response to questions such as, How many ghosts are in the world? What are their properties? How big is the world? How are the roads connected? Then, once the world is established, the observer can see circumstances unfold with some semblance to how they unfold in the real world. This can be used to test out ideas about how to improve the world, stop the spread of a virus, save more lives, save more jobs, or preserve a nation's economy.

An ABM for modelling human interactions in a city might be realised as a virtual world full of human agents and non-agent infrastructure such as transport networks, schools, workplaces and homes. To understand how a pandemic spreads we could set up our world full of human agents that have tendencies to wear face masks, or not; tendencies to socially distance, or not; likelihoods to catch viruses, become contagious, and pass on viruses to other agents nearby. The world can have virus agents too - these might only exist within the bodies of a human agent but be passed from human agent to human agent by close contact. Adult human agents in the virtual world might live in households with other children and adult agents of different genders and ages. They might go to work during the day, travelling on virtual transport networks and exposing them to situations where they come in close contact with agents from other households. This could put them at risk of catching a virus if the agent they meet is carrying one, but this will depend on how the two agents specifically handle the interaction - Were they wearing a face-mask during the meeting? Did they stay 1.5m apart? Was the carrier agent shouting or singing?

In building such a complex model, the modeller must always make decisions about what aspects of the real world can be left out. For instance, if eye colour is felt to be irrelevant to a pandemic, there'd be no need to worry about modelling it. Or if the clothing worn by a human was irrelevant, that too would not be modelled. The difficult trick is figuring out what must be included in the model, and what can be omitted. If a key feature of the world for understanding a situation is left out, the behaviour of the model will not bear a close resemblance to the real world system it is supposed to be modelling. For instance if the model doesn't include humans wearing face-masks, then we can't use it to understand what the difference is between a real world with face-masks and one without. Similarly, if we misrepresent the conditions under which a virus spreads (assuming it spreads by contact with droplets on surfaces instead of via aerosols perhaps), then the virtual virus will spread through a community in our model in a way differently to how it spreads in reality, making our model potentially misleading. Hence, our models need constant improvement as we come to understand more and more about the real world situation we are modelling.

ABMs are an extremely powerful way to help us understand the complexities of human interactions and disease spread. They require a lot of expertise to design, a lot of expertise to build and operate, a lot of expertise also to calibrate and validate against the real world. And their results need to be interpreted carefully by experts. They aren't a magic bullet, but they are proving extremely useful in the world's present situation. Without  computer scientists and epidemiologists, the experts constructing, operating and interpreting these models, it's fair to say we'd be running blind when it comes to handling today's pandemic. Sadly, when people ignore the science, well... the ramifications are distressing to say the least.

Extra reading: Here is an Open Access research article (I co-authored with some of the people leading Australia's current pandemic response modelling some years back) explaining an agent-based model, Synthetic Population Dynamics: A Model of Household Demography. This will provide some detail for those wanting to see how researchers use an agent-based model of human behaviour. ABMs are also valuable in understanding ecological interactions, here's a research article (work by an ex-PhD student I supervised recently) ABM simulating bee-flower interactions A-Bees See: A Simulation to Assess Social Bee Visual Attention During Complex Search Tasks.

Wednesday, September 24, 2014

How to review an academic paper

An inescapable part of being an academic, arguably even a PhD student these days, is the stream of requests to review journal articles, conference submissions, grant applications and book proposals. Leaving aside book proposals and grant applications for this article, please find below a list of things I consider important when I receive reviews of articles I have written, or when I solicit reviews for articles written by others.

0) A reviewer should always read the paper. Properly. Try to understand the paper. Properly. It is absurd that I would have to write this but I have read many reviews where I question whether or not the reviewer has read beyond the abstract, figure captions and conclusion. If you haven't time to read the full article properly, you should decline to conduct the review. Please don't say you will do it and then do a shoddy job. This helps nobody.

1) Cite me! Cite me! It is highly likely, even desirable, that reviewers of an article will themselves have published in the area where they are reviewing. It is of course not surprising that you, as a reviewer, feel your own work to be worth citing, and that therefore you would like to see its value recognised in the articles you review. In my opinion, this is fine. Ask for your work to be cited: if it is essential reading to provide background for the topic; or if it directly supports or counters a claim made in the article under review. Otherwise, please don't ask. If you do request a citation of your work don't harp on it, and please don't expect the authors to reference everything you have ever written throughout their article. Temper your enthusiasm for your own work.

2) Clearly articulate your requests. Clarify your arguments for insisting the authors make a change to their paper. Be explicit about what is wrong or in need of mending and make concrete, constructive suggestions as to how to improve it. This kind of criticism and information is the most helpful thing authors can receive. If you are a good reviewer, you will provide it.

3) Always be nice. Nobody likes to receive a "reject", but the least you can do as a reviewer is to be polite about it. Be especially sure you address point 2) above if you must reject an article.

4) Point out the strengths of an article. You might not find every submitted piece of research convincing, or well presented, or rigourously conducted. Even so, try to find something nice to say about it. Authors can build a good article and hopefully a worthwhile career from what you perceive to be their strengths. This is especially helpful for young or early career researchers. Be encouraging to those attempting to publish in your field. That is a great way to ensure your field welcomes new ideas and new blood, and it assists newcomers to understand how they can make a valuable contribution. Be a guide, not a gatekeeper.

5) Write the review you would like to receive from an expert if the article was your own and you had put a solid year or more of work into it.

Of course authors too have responsibilities. For instance they should proof-read articles carefully, write clearly and report on rigourously conducted research appropriate for their field. If authors meet their responsibilities the reviewer's role can be pleasurable. It is always frustrating to be asked to review an article littered with spelling mistakes, grammatical errors and unintelligible turns of phrase. If you receive such an article and feel that it is beyond repair, please suggest to the authors they edit it carefully (or have somebody do this for them) prior to resubmission. Personally, I think this is preferable to asking a reviewer (even if that reviewer is yourself) to spend their time on an article that is not yet ready for review.

More suggestions beyond points 1-5 above are welcome. Happy reading!

Friday, February 7, 2014

What is a good abstract?

In this article I will discuss abstracts for scientific papers and articles on art and technology.

There are probably many people better qualified to write this post than me, but as a regular reviewer for journals, conferences, grant applications, books etc. I guess it really matters to some people what I am looking for in an abstract, even if I am only one of many reviewers, even if my opinions are unconventional (for instance in the arts). So, what do I think are the properties of a good abstract?

I think a good abstract is (i) informative, (ii) clear and (iii) succinct. Obviously these aren't mutually exclusive properties, they are closely related. To be informative, tell me what I need to know. More on that in a second. To be clear, avoid jargon and keep terminology simple. Avoid references to other material (keep that for the introduction/background). The abstract is not post-modern poetry. Please explain yourself as simply and clearly as you can to engage as many people as possible. If you want people to read your paper, you must be understood by them from the beginning. To be succinct provide an overview but no detail, don't waffle.

What is needed for an abstract to be informative? Tell me directly what you have done. Tell me why you did it and convince me that it is important. Then tell me how you did it - briefly! This is the abstract, not the method/approach so keep it short. If the technique you used is well-known just name the technique. If the technique is related to a well-known technique, just name the technique and say that you used a custom or previously published variant of it. Lastly, tell me your conclusions, but just the main findings. This isn't a mystery novel. The reader mustn't be kept guessing until the Results section. Please include this information upfront.

That's it. I hope that isn't too much to ask. I must remember these guidelines myself next time I write an abstract.

Wednesday, October 5, 2011

cycling and sponsorship - pinarello and specialized comparison

Cavendish has a personal deal with Nike, while Sky's kit is supplied by Adidas, and Cavendish prefers a Specialized bike but Sky has a contract with Pinarello. [cyclingnews] It is complicated when you have people paying you buckets of money to use their stuff, and other people paying to use different stuff. Its tough when you don't want to use the stuff you are paid to use because it is not right for you.

So what's a rider to do? I wouldn't know as I have never been paid to use bike equipment before (although of course I am expected to wear my club racing kit when I compete... which I do!) But of course I have been asked to do things, by my employer even, that I personally felt were against the best interests of that same employer. What's a guy to do? When personal ambition is involved – such as in your own pride in winning a race – or (in my case) in teaching a subject well, or presenting an idea clearly, we are left with a conundrum. Tough call. If I were Cavendish I would take the Pinarello! :-)

Monday, September 26, 2011

mary shelley's moonlit window

"In August 2010, Professor Olson, two colleagues and two students went to Lake Geneva to discover when moonlight would have hit the windows, and penetrated the shutters, of Mary Shelley's bedroom." In this way, and by looking up their astro. tables, they aimed to date the birth of her famous tale, Frankenstein; or, The Modern Prometheus (1818) [The Guardian].

"It was a strong effort of the spirit of good; but it was ineffectual. Destiny was too potent, and her immutable laws had decreed my utter and terrible destruction" – wrote the poor Doctor in Shelley's tale (chapt 2).

If Shelley had her way, perhaps there would be no field of Artificial Life. If we took her text to heart, should we all stop now? Perhaps, like nuclear physics, the potential to make a mess of things is too great? And yet, here we are, pushing onwards in an effort to create life. Ahhh... what would a girl in her late teenage years know about the future of the world anyway?

Thursday, September 8, 2011

melbourne's classic cycling routes

A quick chat in the (miniature) bunch this morning with some friends got me thinking. What are our hottest spots to ride on the road around here? Obviously my views are limited since I usually ride on the Eastern side of the city. Still, here are a few of my favourites (in no particular order). Some are well know and possibly spoiled by rowdy riders or traffic. Others are less well known and still have the rural charm.

  • The Dandenongs (Basin to Sassafras but also the countless roads over the back)
  • Kinglake climb (St. Andrews to Kinglake)
  • Mt. Pleasant Rd. (Eltham)
  • Hussey's Lane (Park Orchards)
  • Beach Rd. (Brighton to Mordialloc)
  • Cottles Bridge - Strathewen Rd. (Cottles Bridge to Strathewen. Doh.)
  • Yarra Boulevard (Kew and Burnley)
  • Beverley - Banyule - Henty - Cleveland - Bonds - Old Eltham Roads (Rosanna)
  • The Esplanade (Mornington to Safety Beach)
  • Clintons Rd (into Smiths Gully)
  • The Alps (Falls Creek, Mt. Buffalo, Mt. Hotham)
  • The Great Ocean Rd. (It is long and almost entirely fabulous)
I am sure I have forgotten many, but that is a start!

Tuesday, September 6, 2011

bicycle headlight - moon X-power 500 review

I have recently taken to early starts at o'dark thirty as my previous post highlights. As I have been riding on unlit country roads, a proper headlight was in order. In my case, the Moon X-Power 500 – that's a 500 lumens headlight for a bike! I know brighter lights are available, but really, are they necessary for cycling? Perhaps for mountain bike riding in the dark?

The 500 is ample bright for unlit roads. So bright in fact that I have been riding it on its "standard" setting of 240 lumens (made from a selection of "Full after-burners engaged", High, Standard and Low) unless its really pitch black. The beam is a good design with ample reach and diffusion to provide a nice balance between seeing ahead at speed and around to give a sense of the space beside you. When the sunlight makes its debut, I switch to Flashing mode which, at a reported 380 lumens, is blindingly bright. Retro-reflective street signs flash at me from a kilometre away when I have this mode on. In pitch black, flash mode is disorienting. The whole world seems to strobe and the mode makes me dizzy.

The power/mode button is a flush press fit, not the best when wearing full gloves but manageable. I would also like a mode indicator on the light or a switch that shows by its position the current mode. It is hard to tell which mode you're in and since the battery life is reduced significantly in the brightest modes, I would like to be able to tell at a glance that I am in a lower intensity mode as I trundle along. The unit does flash red LED at you when it is running low on juice. Switching to a lower power mode can save you from complete blackout for awhile.

The mounting bracket for handlebar use is sturdy but the "quick-release" is not quick. You have to screw in the bolt and use the lever just to cinch it down. There's no way to get the quick-release to work as one since you can't get the clamping loop over the bars if the screw is in. Still, this is a minor quibble. The bracket clamps to my oversize road bars with no problems and the light is slid into place on (or removed from) the bracket with the press of a catch. The pitch and yaw of the headlight are adjustable easily.

A helmet mount (velcro strap and bracket) is included with the kit. I haven't tried it and I am unlikely to do so. There's nothing I hate more than a fellow cyclist looking me in the eye and blinding me with their head-mounted laser beam as they wish me good morning. I bet motorists hate it too. This system (IMHO) has no place on the road.

The light comes in a funky carry case with charger and USB cable. It charges okay in a few hours from computer USB or the wall USB charger provided. Only time will tell how many recharge cycles I get from the unit. It doesn't take a standard AA but instead a specially built NiMH battery enclosure slides inside the light. Hopefully a replacement is available when it comes time! Otherwise I would be really annoyed.

Be seen. Be safe.