美國數(shù)學(xué)建模寫作須知

1、國際論文MCM/ICM寫作須知Anatomy of a Paper Summary This is without a doubt THE most importantpart of the paper. The difference between an honorable mention and a successful participant is that the judges never read more than the summary of an SP paper. The summary should be written LAST. Let me say that agai

2、n: You should not write the summary until the rest of the paper is done. In an ideal timetable, all of monday should be set aside for writing the summary. The summary must briefly hit all the main points and ideas of your paper. If you did anything creative, it must be here. Further you MUST put num

3、erical results in the summary: Our final algorithm performed 67.5% better than a simple greedy algorithm, and 123.3% better than a random choice. Ideal summary length is hard. You must include all your main ideas in the summary, but brevety is VERY important. I would try to make the summary around h

4、alf a page, definitely not more than 2/3. The summary (and ideally the whole paper should be written collaboratively: as a team. One way to do this is to have each person individually sit down and spend an hour (at least writing the best summary that they can. Then, come back together and read the s

5、ummaries out loud to each other (out loud is very important. After discussing them, set them aside and as a teamwrite a new summary together up on a blackboard. Introduction In the introduction, you want to restate the problem as you interpret it. Within an MCM problem, almost every team will find a

6、 different problem to work on. After the contest if you read the papers from other teams, you will be amazed by how differently they approach the problem. Often they wont even be working on the same thing! So what you need to do in the introduction is to clearly explain how you interpreted the probl

7、em, and what you decided to work on. Its also a place to give a little more background on the problem, and show what you learned while researching the problem. The introduction can usually be written first as a friday project. It can help make sure that all the team members are in synch about what t

8、hey are reallyworking on. The Model This is the first big section of the paper. I believe that every problem can be broken down into three parts: the model, the solutions, and testing methods. The model is what youll use to generate the data that youll test your solutions on. For the submarine probl

9、em, you would describe the methods of modeling waves traveling through water. For the velociraptor problem, this models dinosaurs stalking and chasing each other. For the MRI problem, this part of the paper would describe how you created simulated data representing biological tissue. In general, the

10、 modeling will occur on the computer, so the challenge is to translate the code into words, and justify each step. In order to create these models for continuous problems, I would recommend having a clear understanding of how to solve differential equations, however many C-problems have not involved

11、 this. For the animal population problems, you would want to be able to write differential equations describing the relationship between the predator and prey populations, then numerically integrate them. Know what a partial differential equation is - knowthe wave equation and the diffusion equation

12、 and what they mean. With this knowledge in hand, the concrete slab problem would have looked almost like an old friend! Remember, this is the mathematical MODELING competition, so do not gloss over this section. It may be simple - for the grade inflation problem, this section would merely involve s

13、imulating the actual grades for a class, then using some method to skew them due to inflation. In general, for the discrete problem, you want to be familiar with how to generate random number sets with different properties - this can be very useful in constructing the sets to test your algorithms wi

14、th. The programmer should be writing the code to create these models friday morning, andso this section should be roughed out on friday or saturday. The Solutions(PLURAL! The second BIG section of the problem. Here, we describe our methodsof dealing with the data generated by the first section. This

15、 is the section that actually describes how we solve the problem. In the submarine problem, this is our algorithm that takes the simulated data from waves traveling through water, and uses this data to guess at the position of the submarine. In the velociraptor problem, this is our algorithm which m

16、athematically states how the raptor tries to catch the thescelosaur, and how the thescelosaur tries to get away. (Occasionally a problem will explicitly give you the solutions they want you to test. In the Escaping a Hurricane problem, the solutions were to reverse lanes on the freeways and possibly

17、 the surrounding roads. You MUST have more than one solution. Let me say that again: MORE THAN ONE SOLUTION. In order to show that you have a brilliant method of finding submarines or cross-sectioning gridded MRI data, you need a baseline, something to compare your solution with. You want to start w

18、ith the simplest, most obvious algorithm possible, then gradually build on it, refining it until get to your best solution. Often for the discrete problem, the simplest solution may be merely to make random choices. For the meeting scheduling problem, you might want to have one algorithm which just

19、randomly makes up the schedules. Then when you compare your better solutions with it, they look good! You want to show that youve explored the problem thoroughly, and that youve tried many different approaches. Even if you started with your best algorithm, then tried a bunch of blind alleys, in the

20、paper you want to present things as if you started with the dumbest most basic solution, then gradually refined it and finally arrived at your best solution. What if you tried a more sophisticated algorithm, which didnt work well? Put it in the paper! Show all the angles you tried, even if your best

21、 solution is not the most complexand interesting one. In real life, that happens very often! Solution TestingMethods Sometimes the problem will state very clearly what the goal is, so itmakes your algorithm testing methods fairly easy. For the submarine problem, you create a simulation of sound wave

22、s propagating through water, bouncing off a submarine, then being received by an array of microphones. This data is passed to your various algorithms which all take a guess at where the submarine actually is.All you have to do is find out how far each algorithm was from the mark and you have an easy

23、 method of comparison. However, you usually have to make some decisions in how you compare the results of your algorithms. In the MRI cross-sectioning problem, you can compare your algorithms estimate of tissue density with the actual density created by your models for each of the thousand or so poi

24、nts. But do you just average the variance? Maybe you should look at RMS error? Are you concerned with making sure that no point is drastically wrong or that the overall error is small? With a lot of problems there will be many ways to compare your different algorithms, and theres good reason to use

25、more than one method to evaluate them. Evaluation methods shouldbe one area of brainstorming that you keep working on all weekend. Results Here, you need to actually present the results of the testing. This section should be very focused, because youve described everything else. If possible, you wan

26、t a lot of data to back up your conclusions. Have your models generate a whole bunch of different data sets, and see how your algorithms do. In general, youre going to end up with a lot of parameters to play with - in models, your algorithms, and comparison methods. Try to explore as much of this pa

27、rameter space as possible. You want to show that youve taken a mature approach to the problem, and probed all aspects of it as best you could. The specifics of data presentation are difficult. If you can make graphs, by all means do so. In the MRI problem, we made residual plots for each of our algo

28、rithms and each of our data sets. But ultimately if youve thoroughly explored the parameters of your models, algorithms, and comparison methods, youre going to have a huge quantity of numbers to present. Give then the numbers from all your results in tabular format, but dont expect them to read the

29、tables. You want to create a narrative in the text, going through the tables, pointing out the general trends, the exceptions to the rules, and the overall results. IMPORTANT: Many teams create a model, a solution, run a test of their solution, present the results, and stop. You must run MULTIPLE te

30、sts! You must find out if your solution is stable! Will it hold up under slightly different circumstances? At this point your code is debugged and generating useful information - take just another 20 minutes and run a few more cases, varying the parameters just a bit. If youre doing the Escaping a H

31、urricane problem, run your code with a few more cars, or a few less - see how flexible your results are. If youre doing the MRI Problem, test your routines on several different simulated organs. If youre doing the Air Traffic Control problem, throw the planes at your simulated airport a bit faster,

32、a bit slower, or take a runway out of comission and see how your program copes. Prove to the judges that your solution is flexible and stable, or honestly admit that your solution doesnt work as well against certain challenges. Your paper will look a whole lot better!Conclusions - S&W - Future Work

33、First, present the bottom line, evenif you just presented it in the Results section. Overall, algorithm A performed 34% better than B, and 67 % better than C.You need general numbers which summarize everything, so figure out a way to somehow average all the data and distil this into a few numbers wh

34、ich numericallyrank the algorithms. The results section is where you say Algorithm A worked better in these circumstances, but B had the advantage when we did this. In the conclusions section, you have to boil it down and say A is best. This is also what you will need to put in the summary, so be su

35、re you get specific overall results.Strengths and weaknesses is a great way to go over the key things - explain the good and bad points of your algorithm. I like to use a bulletized list. Nothing new should go into S&W - you are summarizing. The main points of your results section should be here, al

36、ong with the shortcomings, and any limiting assumptions as well.In order to show that you are mature in your approach to the problem, a future work section may also be useful. What would you do if you could work on this problem for the next few months? Were there any algorithms that you thought woul

37、d be great, but just could not impliment on the computer? The MCM is very limiting, so this is a place to show that you can see the big picture.Overall Recommendations: An skimmable paper The first thing to keepin mind is that the judges are looking at a lot of papers in a very small amount of time.

38、 In their first pass, (in which half of the papers are thrown into the successful participant pile!, the judges can only give FIVE MINUTES to each paper. This means that theyll read your summary, skim through the rest and then your paper gets tossed in one of two piles. Thats it! Half of the papers

39、never make it past this stage. Even worse, in the second round of judging, your paper will STILL get only about FIVE MINUTES. Only by getting through the first two rounds will your paper be read in any detail. This means that your #1 goal is to beat the five minute read.Weve talked about the summary (THE most important part of the paper, now what about the skim? Things that may get noticed are headings, bulletized lists, tables, diagrams, and figures. What yo