How to industry sign banking michigan ppt now
all right I think we should we should get started so let me just take a minute to welcome everybody to this special seminar series on data science for the kovat 19 problem as as all of you know data science is of use of value in essentially every discipline across our university and the many disciplines that bring value to addressing different aspects of our current pandemic are no exception and so we find data science at the forefront of addressing these issues and and so I like this with we're kicking off a seminar series as well as sort of mini workshops that dig into a few of the topics in a little more detail and you'll find information about all of this on the midis webpage but today I'm delighted to have our seminar series kicked off with presentation by Professor Chen Chen from the department of industrial and operations engineering the next talk is next week where professor Brahma Mukherjee of the department of biostatistics so sir Shen is has been an active affiliate member of Midas she's also an associate director of the Michigan Institute for computational discovery engineering she got her bachelor's degree from Ching WA her PhD from University of Florida she's won a number of honors including video a early career award and IBM Smarter planet innovation faculty award she works in the area of optimization primarily and risk analysis and when the covered 19 pandemic hit the world she realized that there were many logistical challenges that the work that she had been doing and others in her field would be of importance as so many people struggle to put various systems together under very stressful conditions and we're delighted that she's take the time to tell us something about that today so the chenchen please take it away invitation for having me here and so today I'm going to as Jack says I'm going to show some of the things that has been down in particular in the area of operations research industrial engineering system engineering and see you know anything that you know we can push them further to address some Kobe nagging related issues so I'm going to share my screen now so that you can see my slides and I'm going to start it so doing the presentation if any of you have any questions because I think I mean those sort of like a full-screen version I couldn't see you either you nod your head or raise your hand so please enter your questions in the chat and then at the end of the talk I'm going to go through the chat and read off the questions and then answer them so that would be something that we will try to do during this talk all right all right so yeah so as I said you know sort of like this is basically it's not it's today I'm going to give a very you know technique all right light talk basically we're now going to dive into a details of the techniques and so and so forth for specific problems but it's more like you know like we try to cover as many aspects of Cobie knighting related issues as possible and try to present some tools and models we have out there and the outline of my talk is our following I will first start with some overall lessons like why motivate us to think about these problems and then basically we try to put into problems by categories so the problems that very necessary to address during the disease prevention phase versus if you you know your disease already had an outbreak and how we control them and as you can see there are many different parts that we can you know kind of I think about as system engineers and then try to you know find ways to to improve the current situation and then you know like hopefully as the disease comes down we will need to also deal with some other issues in a recovery and and an impose a recovery planning phase and with the conclusion so one thing I'd like to mention is that my talk is basically based on this so if like a document I have also put it on my website this is a developing document so you know if you have other reference or any other paper start dressing singular issues that I didn't including my talk welcome to send me email and I can include into this and and before I start I just want to like thank you all the people who gave me help especially these are PhD students that they have their own project to work home by at the same time they just devote their time to help me collecting some reference than making a slides so thank my former and current PhD student in fact a lot of formal PhD student gave a lot of help as well because they are working in a very firm line as you know research scientists in Google and Amazon and then and they have a lot of real-world problems they are addressing daily right now so they gave me a lot of ideas alright so this is a picture on McKenzie and basically this picture date is pretty outdated as we can see now it's a modern month ago and that was a time that when you know the disease virus kind of like jumps out of China and start spreading out of the whole world and as you can see back then United States is not really seeing has a big problem back then you only has 1678 infected cases but we know how bad it is now so so I think you know you can see how quickly this disease spread out definitely from this picture but one thing I would like to mention is that in this picture it reflects you know different countries that can be at different stages in terms of the disease outbreak anomaly you can think them as you know if you are in stage one those are sort of like you know small numbers of cases identifying out of local transmission and observe yet and then you know like and and as you know time goes if we didn't do well in terms of controlling disease we it can get worse and you can get into these you know stage two three four and Stage five basically represent this recovery base that the disease is almost under control and then but then you still see some new cases appears for instance you know China has been that stage for quite a while right now and we can talk about how you know what out of problems they are facing right now in terms of being Stage five so so I want to come as I show this picture just to you know emphasize that if you want to look at a problem you need to first examine what stage you are in because the problems are very different depending on the other stages and and the virus outbreak situation so before I start with the you know the modeling part I want to emphasize first like what data science can help us right so so even you don't want to make any decision at this moment you just want to see what our systems looks like at this moment there are a lot of things can be done using machine learning data analytics tools so I gave you a few examples like you know this kind of like picture is being drawn using a lot of like tools behind it so for instance you can kind of I could see the trajectory of the disease the confirmed cases in different countries how they evolve and so on so forth and you can guide you through a few you know decision making process later if you want to kind of I figure out a more detailed sort of like setting you can see that back then when I was updating this slice so that was like April 2nd that's a date and and you can quickly see that our you know United States spend eyeing us for FICA increasing mode at that moment as a comparison for South Korea and China has already inquired a little bit so that kind of I gave us a little more information and again there are many things you can do so one thing that you know people have come up with and that's something example I find out in lie and I believe that was a talk actually give a few weeks ago through my tasks as well so Michigan researcher also you know think about how we can predict the disease trajectory especially for instance in Michigan or in County it doesn't have to be a country level so one one observation we have from the past data before is that oh there is some you know kind of like patterns and in some cases we call them exponential growth so if we know that's sort of like the trend based on the current data then we can solve I predict the future right and then that's exactly if you go to this website are developed by Michigan researchers that you can find more results over there regarding the the prediction part and and if you're digging to it it's not only about prediction of how many confirmed cases we can even divide it by you know the zip code they leaving the demographics of the population and so on so forth okay so in this talk I would like to focus on the other side which is if you have the prediction function and also some tools that you can use to know what the future data may look like even if you have some certain uncertainties can we put them into actions so so basically you know I would like to change from the data apart to the action part and then to the solution part we cut we have been talking about flatting the curve you know again and again so what we can really do in order to fathom occur from what's the mathematical models behind them in particular I would like to put them into prevention and intervention control and recovery phases I'm and then I you know because my background is in ioe so I can only be sort of like giving you more information about Oran and I Industrial Engineering ie tools but you know like different different domains as Jack was mentioning it at beginning are trying their best to use their domain knowledge to help the very emerging situation right now so in the prevention face the first thing I would like to mention is testing so definitely testing is very very important actually you know like later we we see that a lot of it good examples about if you do testing well you can you know kind of like integrity the disease in a very early phase and hopefully control it so so this testing can be started even we see the outbreak happens so being the ideal case that you know if you predict that the see the the new season is coming and your wave is coming how we can actually allocate our testing facilities that's one question and for that you knowing in our domain basically there is a very well known traditional problem called foresee location problem FL key that has used to be you know kinda help that we to solve their supply chain problems so they need to know where to locate their you know kind of like manufacturing our plants know so how to allocate their retail warehouses the retailer stores and so and so forth and the goal here is that Kawai use this location to satisfy the demand spread out everywhere and I do not know exactly what a demand will be but I have a prediction just like the case that we have for the Coby 19 case and then you know in addition to the location design you also need to ship a testing kit from your production facilities to the testing facilities and then from testing facilities to less so there's a lot of like a shipment design as well you have to you know cafe operating a database so for this type of problem the objective here usually will be you know minimizing the total cost if you talk about the cup you know the contacts we're facing here the cost will be how convenient it is for local people to travel to the design for City locations so you can sort of I try to see the ways that they can you know kind of like travel or just the total distance for them to go to these facilities also you want to take into account the cost of shipping as well for the test kits so overall speaking it's a long-term process but then you have to make the decision now and you operate over time okay so this is the the model that usually people have formulated for this type of problem and in this model basically the key here is for most of the aw ma dough's we try to use linear constraints because then you can have a lot of algorithm for solving them it can be mixed up very low so you can have some continuous variables representing your shipping this season but if you want to say okay where do I locate my first three T's those are zero one binary decision so you have to deal with some sort of a we called mixing feature in mixing two your linear programming model which usually are empty hard so there are some solvers liver work for that per person solving them efficiently so then this is a real-world example in fact Singapore has done models in terms of you know deciding their testing facilities and we all know that they didn't have it very bad you know disease outbreak before but still if you log online you can find a nearby you know contract facilities you can get tests and also they show you how many units are still available in those you like the testing kids ask you a variable in those facilities so then you can choose the nearby one to go so these kind of information I feel if we can get them in place as already as possible that the public see them it will our maximum ly reduce the chaos we we may have later and then as I said you know like if you want to build a mathematical model normally there are several questions we need to ask ourselves in order to build a model I just show you what we know about it the situation so in this case the situation is we would like to you know beautiful cities and so we need to know the potential number of in fact their cases population density characteristics and also you know like the travel distance from the manufacturers individuals to potential facilities what are we trying to decide here is those are those binary decisions that I just show you as the X variable so where to locate and the options are you know you have all these CVS Walmart parking lot that you can use so then they will be associate with binary variables to be integrating to your model and the testing capacity capacity also sometimes that's a decision variable will have to make and also the shipment volume the goes here we can maximize the total shipment cause or minimize the total shipment cause or maximize the travel convenience so I would like to you know show you one reference that's kind of like a test book very well cited in this domain area talking about network design foresee location design a lot of other things that has been used widely in retailing but now you can think about in terms of the designing for the coffee nineteen case again I would like to emphasize that how important testing is by giving you example about Singapore you know the information is very very uh transparent if you need to test a starting from the very beginning you can find it nearby locations and so and so forth and then they still if you extend this model a little bit right so the model I was showing you doesn't have this ext and what that index means here is that you have multiple time periods to make the decision so you can think about your decisions not only made once but then you need to periodically re-examine your decision and decide whether there is any new facility I need to open where any facility who closed if you are in a much good shape so then you know like we call this 4c location but then you you normally if you want to extend it to a multi-stage case you you need to incorporate this uncertainty so this is sort of like a time series model about the demand that you you will have and the demand is highly relevant rather than to the infectious status that you will see basically using the data prediction about how the community is going to you know like a spread and then you can use that to predict the testing demand and use that testing demand feeding to this model to decide the best strategy were you know sequentially allocate facilities so that's some other things that we can think about not only that we can even you know locating or key hospitals like if we want to direct the public to certain hospitals instead of you know so if I get all kinds of hospitals then what would be the ones that you know we should we should try to to locate and also prepare in our PV ease in those hospitals so these all can be done in the prevention phase but but as you see here I'm mentioning again in the outbreak space because this decision just needs to be done so open and so periodically so there you know like in terms of the the testing the key is how can we increasing the testing ability and to me that's not only you know like the number of testing we can do I think evolves a lot of like systemic
ays and then you know strategies that we can implement even with limited testing we have and then and then the example is again you know so if I for from countries like South Korea and Singh they are so small so that they can say okay we just like to a massive testing and make sure everyone gets the test United States were not even near that so then you know like to think about testing as a more sort of like a system engineering problem and design testing more effectively rather than you know sort of like UPS widespread I think that's also the key that we need to think about here so I'm citing you know sort of like some literature about you know confirming that testing is central to a DZ outbreak response and and you will help you to detect early and minimize further spread so so I cannot emphasize you know how important testing is I would like to move on to another problem now but then there are actually a lot of models related to testing that I could not cover in my slides but you know you're welcome to examine my arm my online document and it won't meet some comments so the other part is if we see the disease already out spread then we need to make this decision right which is lockdown so the lockdown can be done sequentially it can be done by different types of facilities that we would like to close just like what we're doing now reopen the economy we can also do it sequentially by we open you know each part but not all of them together so then you know like this chart tells you different types of events activities what's their chance of getting you know disease spread out and so and so forth so basically you know the idea here is that you know these change the chances and you know what's existing activities out there the question is how much you can afford in terms of locking down so you have certain budget like you know now if they're the very beginning ah you know if the situation is really bad you'll probably just should lock down everything so then this was the the prevention versus the intervention idea you can do that the planning much earlier so that when you have to reactivate you can quickly commit to your policy before or you can do it dynamically as you go and as you can see this is a picture of New York City locking down all right so to build a lockdown model we ask how what data we have our decision we need to make and what other goals right and then as you can see there are many ways to characterize the problem one way I present here is a very sort of like a static way and a naive way but I think he starts the model right away so basically you have these locations and facilities or the activities you know the frequency how people visit them and how important they are to people's daily life by some number and then we need to decide which one to close with certain budget and what are the goals here is definitely you know me maizing the total infections and and inconvenience of you cause the by lockdown and also the economy impact so very quickly you can build some sort of like we call knapsack model so the idea of a knapsack is you have a bag and you have many items that you can see that I'm putting the bag each item has their value but at the same time the bag has their own capacity so how you can actually select a subset of item in this case a subset of lockdown facilities such that within this capacity you maximize the value of all the items you select and this case can be even more complicated because we have this probability so locking down something doesn't necessary say a hundred percent you will you know reduce the infection by so much right so there are certain probability about you know if I jointly lock down one facility and the other facility and then you know because there overlap there are many populations going to these facilities have their overlap the total probability will be need to be calculated in terms of the total induction a reduction of the infection so I direct you to this paper that we publish actually quite already go about this how can we design this lockdown strategy in the prevention phase and intervention phase and now it's a paper done by my former PhD student and a collaborator who actually graduate from Michigan and so then you know like we can also think about you know individual : tying model has some sort of like a graph as well so we can you know the question here is maybe you have a network social network there are certain people in a social network are very highly contagious or if they get affected infected and so the question here is if I need to you know how I protect the certain groups or quantize certain groups who are the ones I I need to you know have I take action home and and then the idea here you can kind of I look at something like this so we we did some modeling for disconnecting a network so if you are network looks like this if I give your budget say you can delete one node from this network which know who you were deleting order to maximum li disconnect the network and you can come a pre quantify as what do you mean by disconnection so here our definition of the disconnection is that you were resulting the maximum number of components so for instance if you delete this black node you will see that these parties are still connected so they are one component but these three guys are disconnected so there are three components so you have small components as a results instead if you delete this node gray node then everybody's still connected somehow right so then in this way you still have only one component still so the question here is what's a strategy to do this kind of like you know quarantine or interdiction work or isolation problem normally you would think why don't we just do a greedy way which is counting how many connections every now it has before the maximum connection node ID and this picture already shows you that's not going to be working for some cases like this so then in this case we need to build a model basically solve the problem and try to find what would be the best way to to phone type of course you you may have multiple budgets not rather than blocking one node you can rock multiple and then the same question you can use for deciding your treatment plan if you have certain groups that only some sort of like a special treatment meaning that for instance you can allocate additional facility in those communities and making sure that these people in those communities can get treatment as fast as possible where are you going to rotate them and then the same graph you can you can do basically for the current I can be used for treatment okay and then there was some other models we did before sort of like saying you're cleaning strategy so for instance if you have Network there are some you know comes I know they're already being kind of like contagious ah you can pull some cleaning agent in the network and then basically these coming agent will chase these node and then the virus will will kind of I spread out but at the same time your cleaning agent will be spreading out as well and you can define certain rules how they clean for instance if they post arrive at the same node then the virus will get clean and and then that way you can so if like a map your whole place as some sort of like a network and try to use some formulation like this to to decide the best cleaning strategies so these are talking about the testing lockdown Kranti are one big problem that we are facing in the in the coving nineteen related problems are health care systems overwhelming right and of course you know like how can we reform our health systems make it resilient and also like improve our daily operations so that we can reduce the the infection that's a key issue and even with without kobe nagging this has been dying the health system for quite a while so i can show you some some problems that you know like or IE people have been thinking about in terms of the health care operations management so the issue here is as you can see for kobe 19 specifically is the high demand we have and they're very tight resource that we see in a very short amount of time so so it's kind of like a big pig that you see in the demand in terms of the health systems facility resource need and then a go here also is how can we avoid over Cronus and cross infections so there are many things are people have been thinking about for the ICU bed if you'll have many people waiting for ICO Pat how can you allocate them for even like patient emission part how do I a meet patient and try out them so that I can minimize the cross infection how do we plan staff members and how do we dispatch ambulances and and again you know these are very small snapshot about the whole picture so there are a lot of like health care problems this was a I put all the slides about about a month ago so I actually pull out the the one of the picture shown by a newspaper that when the virus was initially started in United States they actually claimed in a newspaper that United States should be fine with the wisdom virus because if you look at e ICU bed her number of people each country has United States is leading the world so even Germany is lower than United States and then we all know that even the situation in Europe was pretty bad Germany was kind of like much better situation than the the rest of the countries in Europe so that was very optimistic but then the question here is that to wait can we just use this figure to say where we're verifying with the ICU beds and so that's fine with you know like the high number of critical symptoms patient that we see in Kobe nettings case what's not reflected in the figure is that how much in balance we actually have in terms of the healthcare resource allocation in different states in different countries right and also the there are some places like New York City has a very high population density and then the the normal ICU bad demand it probably is already pretty high here and then so you can't really allocate extra and also the extremely high health care costs and some ICU costs in the u.s. so those are the other factors are causing the problem these days as we can see so forty I feel bad sorry for so I will mention the ICU about problem a bit later so if you think about the whole health care system the first part starts from people come and say I need to check in myself and at that moment you have different types of patients they are conditions are very different so how can I do the triad your Mission Control they'll be better so what we need to decide here is prioritize these patients to be tested middle and and also triage them into different departments and assign appropriations to doctors and nurses so so when we do these decisions we hope that these can be integrated into our objective so patient safety operational efficiency and also the risk that the health care providers are exposed to infectious patients so as you can see using some math mathematical models already we have this some sort of like guideline in terms of how do you triage patients as a flow chart and basically these are all down you know like behind the thing they're all like math models deriving them and I show you a y example that has been done actually by our Michigan people so so this is a very well cited paper talking about you know how do you prioritize patients and this is not for Kobe 90 just like in general there's some certain type of patient arrival so you can look at this hack sort of like a queueing serum system so so so you know like a hospital or bank there's so many kind of systems that you can model this way you have multiple queues arrival you just need to try out them making some decisions and eventually they will become flows into different accounts in the in the system so you can write some sort of like a stochastic processes to guide the decision okay if you make it more complicated you can integrate more sort of like a decisions and also the complication of the departments into your into into your setting so originally there was a paper 2005 which popular one talking about we should use this typical 5 level charge system so the author thing this paper proposed a new new way to design the triage and use a lot of like testing since the real data to show that they are more effective okay um and then you know like the the ICU bed I was talking about there are a lot of like departments in the hospital not the only ICU but then other departments Nizar allocation of resources meaning the best that you have and then in doctors and nurses so as you can see during the COBE 19 you know like a hospital is very overwhelmed a lot of like beds are actually in the hallway so how can we better design these I see use capacities and there are many ways you can do you can convert one single loom to doubles you can discharge some patient who doesn't need the care anymore so then that decision is also critical as the same as the triage decision now you need to try out them to some lower priority department in the hospital and you also need to reduce the a mission for the non covin 18 cases who doesn't have to come to a hospital anymore for their you know regular check and things like I you need to decide that carefully and also define how to allocate gas technology and resources and provide training to healthcare workers all these can be done using some models to guide the decision so I mentioned one as an example to you this is very recent paper coming out to one of our sort of like a flagship journal called operations research so the paper says allocation of intensive so ICU beds in high demand and and so the idea is that they model this whole process as some sort of like a Markov chain Markov decision processes and make decisions along the way so so you have actions and then your actions can cause the transitions of the states and so on so forth the other big part of the problem here is the dispatching of the ambulance again Oh like you have many stocking different facilities and basically you need to send them out to cover it in and so that the demand is arriving randomly so you don't really know next minute what kind of call you're going to get right so in this case there is a series of literature down by you know professor Laura Albert who is a spurt of you know like ambulance dispatch in Wisconsin and and basically it's talking about you know how can you design the coverage model as well as a real-time model and then a commonly used approaches comics nature programming formulation so as you can see here still everything you know the constraints you have in terms of how many ambulance you have and so and so forth and then the the flow balance comes from can all be written down as linear constraint but then you may have two integer variables meaning how to dispatch them and and then these kind of models can be solved very well in the of the show of like commercial solvers okay the other kind of problems in healthcare related domain is this telemedicine service scheduling that arise pretty much during the Kobe 19 case because many patient cannot visit the hospital they need to you know like talk to their doctors online and the doctors has certain time limit so they can only do the shifts like off the duty and then they tell you like what time I can do it and then your basically needs to put them into schedule right so so this is a very well established a scheduling problem down in the literature for industrial engineering but then it just like has a new use today for this telemedicine so this all right some other things we can think about is for some of the the the elderlies we may send nurses to their home to treat them rather than having them come to the hospital and getting affected because cross infection has been observed as a big problem in the early time in wuhan when the disease first outbreak many people actually get infected when they wait in the hospital so then you know sort of like these medical home care in every problem can be think about it so the essential idea here is that you have you know that you have many health care providers information so where they are what's the Esper T what kind of patient they can cover you also have the location of the patients we need to decide is how do you match the
providers with the patient as well as telling them the route that they need to do every day so then you know like you would tell the patient when some people arrive their home and you would tell the the providers like how do you you know what's the sequence you visit the patients and Asthma describing this problem it's all ready to you seems very similar to this you know kind of like a post delivery or UPS problem and that's exactly you know kind of a what has been done in the literature in terms of modeling vehicle routing problem via P matching there are very well known problems you know in terms of finding you know come by routes schedules for you know delivery you can have like one tipo a multiple Depot multiple vehicles and so of course and the basic model is very simple but you can add more things like time windows and everything to this model and and then you know you can use similar ideas to design this medical home care delivery so I think the supply chain is also a big area that people have been you know kind of worried about and mentioning about for the Cobie 19 you know impact so the the surprising problem is very challenging because we have seen very abnormal demand patterns these days and then for the sort of like ER overall you you can look at this book it's a very sort of like a well-known well cited book in supply chain management it talks about supply chain risk management as well for this kind of a situation and the whole idea here I would like to give a sort of like a quick snapshot about the the bigger picture is why the supply chain is so you know kind of frustrating these days is because some in fact people observing the suppression called boo will prefect so if you don't do anything and it's supposed that you look at your hosts whole supply chain from a customer side to the manufacturer side if there is a small you know kind of a turbulence in the customer side it will get amplified towards the end so your manufacturer will order much more if they see a small jump in a custom order and then the same if the customer has a little bit dropped then your manufacturer in fact were decides their production plan much in a much bigger sense in terms of reduction in the plan so this is just some sort of a phenomenon that people observe and then people have been thinking about how can we you know kind of avoid this bullwhip effect and there's a notion called the safety stock and a lot of things being introduced in order to prevent that you know like people just get panic and and and then you know so if like make their decision very unnecessary so using this observation basically motivates a lot of researching the supplies your area I put two out there just as an example something that we do is how can we jointly make location design and decision together to mitigate the supply chain risk and also some other things you can think about it's introducing this uncertainty true model and design a more robust model and the idea here is that you can do some risk pooling so if you only have one supplier if that's acquire gets some problem then you may you may see a big failure in your system so the idea is can we actually have a hierarchical order of the supplier I go to and basically you'll have a central supplier and the secondary and maybe third supplier and then for the secondary and third basically the purpose is maintaining well relationship with them so that if you have some back holders and you can quickly react to it again you know surprising is a big tip demand and these days it also gets evolved into this online retailing problem more because we have more ecommerce and as you can see here now the supply chain problem gets integrated it was this vehicle routing problem I talked before because not only people with shop they will also ask you to deliver so you have to desire delivery system as well as your supply chain inventory problem together so so that becomes a even more challenging problem and you as you can see many retailers are failing this test right now because just we can't respond to that big search in the in the online retailing these days and and again I'll for the time sake I'm not going to build the details how you design these type of model but as I said you know there are literature's talking about you know the modeling part and and eventually there are two big problems we need to solve one is how do we control the inventory inside a store and also how do we deliver them to the customer using some vehicle routing problem I talked about before okay a lie is also facing a big challenge to operate their system these days even the the Kobe 19 and for airline management and airport screening again we have some models established before for different purpose and now we can re-examine them and try to make them work for Kobe nettings case so I'm showing you basically you know this picture shows you the decline of the the trips of the airline face these days and they have to basically adapt to everything they have to this you know decline of the trip so rearrange all their flights and everything and this picture shows you that you know what's going on in the airport the I think it's Dallas Airport when the u.s. announced this you SEO travel ban so that day basically everyone trying to you know come back flew back to the US but the airport didn't get prepared in terms of how they should implementing a different screening strategy so that everyone stuck in the airport for many hours and because of that you know you can see that there's no social distance here there's no one wearing any masks here except to some some sick people probably so then you know sort of like you are getting exposed to the virus in a very high risk so so can we actually have these policies in place before we make reactions that's that's that's something I think we should think about so for the airline fleet management there are many questions the airline needs to solving a daily routine base so how do they put their schedules depending on demand between different cities and also once they have a schedule for each schedule they need to use a different types of air plan depending on how big size small size of the the fleet can work right so d and then once a while I need to pull the aircraft out for maintenance and so how so how they should decide that schedule and also for every fly they need to cook crew scheduling and then they need to control the OD pairs of their whole network so the typical way that you know like the research has been going is to design this so vai very sophisticated again they're integer programming models with linear constraints to solve this large size airline scheduling problem and and because the size of this there are many algorithm called like column generation approach a lot of things has been developed for speed up the combination of this algorithm okay so then you know also for the a li you know staffing sent constant or staffing there has been a you know comes like a wide research being developer for this type of problem the main theory is a call queuing theory and and then basically is for you know handling large volume of calls which is frequently happen these days the screening I was talking about there has been some research done in terms of a multi-level passenger screen the idea here is basically the result of that is actually this so the paper was publishing 2006 and that was like after nine one one and then people been complaining about they have to go through the whole process even you know like the chance of catching the the terrorist attack attacker are pretty low so then you know how can we actually prioritize people and then as you can see today we have this TSA PreCheck that you know you get pre checked so that you can go faster in the airport and then you know some sort of like this clear program so they're all sort of like a reads out of this multi-level passenger screening approach which can be actually down also in in the Kobe 19 case when we need to screen people later for possible infections I just want to mention one thing that all the models I was discussing I think there is a big use of them in the lo comp low-income and underserved communities and as you can see today there are people who need to help the most these days you know ranging from a variety of things and then so basically we can think about a for instance I I'm I'm happy to learn that another public school has this food distribution plan to satisfy kids who had free lunch at school but now because the school is is not open so that they kind of get healthy food at home so what happened is that the school still cook food every day and and then they have the school bus to deliver them to the people who needs them but then you know behind it you need a you know mathematical model basically tell you how can you design the bus routes and everything because all the food has to be delivered location for people to pick up during a lunch time so the time we know it's very narrow and you'll still have to design now routes to cover that so that's a very challenging problem and they have something down uh behind the scene and I have talked to the to them they have done pretty well but only ala Hanna I don't think of many public school United States can do that so that would be something that we can do put online and then sort of like this that they can use it for all free okay the last part I would like to quickly mention a lot of questions we need to also think about when we do that when we are in the recovery and post recovery phase and that's a lot of problems that actually if I you read a news that's what China is facing right now ah so so how to continue allocate the essential hospital resources in that face it's very important and also you need to attract the reinfected patients and most of them actually have no symptoms for Kobe 19 and then how to continue to perform multiple level passenger screening hospitals in airports and so on and so forth how do you control the inbound flights from different countries who may still be not in the recovery phase so these days there's a lot of issues with the airline scheduling and everything in China just because a lot of people flip fly back but then you know they are coming from countries with the highest status so then you know they may read they may they may increase the search again so how do you gradually unlock our facility in school sending events and how to adapt to new customer behavior and travel behavior you're no more supply chain models and these are all valid questions and quotient questions to ask and and the conclusion is we don't have to invent something new in fact a lot of research related to these facilitation queuing problem integer programming vehicle routing that I have been talking about in this talk can be modified and Southey some of these problems okay so with that I would like to conclude my talk as I said I put this talk a month ago so it's still there this is a snapshot of the the the W Oh sort of like tweet her account say sad it took 67 days for the first reported case to reach the first 100k case and then the nest 100k case was reached 11 days and in the nest one was four days and that was stayed March 23 and we all know that in fact worldwide we have almost 3 million in Factor cases and about 210,000 - cases and about a third of them is actually in United States so that's very hard to read but then I feel like there is a lot of lessons we can learn from other countries or even from the research we have done before even not at the scale that we're facing right now and we should utilize them policymakers should really utilizing this approach that scientist has already been developed for years to fight the disease so with dad thanks for listening to my talk and then if you have any suggestions about other tools I didn't cover um please send me email and I'm happy to include it in my document thank you and I can take questions for now you can either use the chat or you can just you know like a mute yourself and ask you don't whine it's fine for me you you so well we give such an a minute to catch a breath that was a really very nice talk search and you covered a lot of ground and give those of us who are not experts in your field a very nice overview of the many different places that our techniques can be useful in addressing the Corbett crisis so thank you for that and I wanted people to go ahead and put questions in chat if they have any or to just yeah so if someone can help me that would be nice because I couldn't find the chat oh my screen if there's already a question there there isn't any question yet oh here the first question has just been posted on chat in person game theory concepts for Kobe 19 oh yes so myself hasn't and I should be clarify here that you know the models I was reviewing to you some of those what I have done before and then some of those are not they're just you know contacts summarize what's in the area so I definitely think that you know like game theory models you know in fact I show you a snapshot of this interdiction model we talk about these sequential game that you can think about but then there is also some sort of like National Curriculum and stuff like that you can be down here as well so like we can we can you know like maybe talk offline or by email to talk about more specific context that we can apply this you know in-person game it's what I call me 19 but I also know that there is a lot of like CSS research being done in terms of studying you know kind of a social behavior how people cooperate with the the order of stay home and then and then you know this basically can be model as game as well so there is another question about is there any model for distributing medication medication trials and so on so forth so I know that there's a lot of research being done in terms of the drug supply chain management so not only for Kobe 19 you know just the FDA Jack's like how do you you know cuz I distribute them to different channels and that's kind of like a supply chain question and actually I have colleagues in my department who is a spurt of that I can do back to you some papers that they wrote these days I think wow the interesting question I heard is the repurposing of some of the drugs so you know it's it's very time consuming waiting for some drugs together fda-approved for just Kobe 19 but then you do have some other drugs getting taking some effects and and people have statistics about showing their their their performance right so then you know the question is how can you take combination of the drugs that already been proved for other type of disease to treat some of the Kobe 19 cases that's something people have been talking about and that can that can use some models behind it the models that depend on theta they are concerns about data quality and lack of transparency and a coordination between garments how can data science community tackle these issues and I think that's an awesome question and and I didn't you know come by dive into the deeper about that you know these different models what they're like a new status out because the models I show you are very traditional old models but the question here is like the parameters we need in these models right we often assume if you take any classes most of the you know kind of undergraduate classes we tell you we give you these parameters we let you know what the costs are what the demand are and so on so forth you're plugging to your model and you can solve it there is a bigger research area in optimization called stochastic optimization and robust optimization are dealing with the issues that you don't know the parameters and in this case if you don't know the data can be completely trusted what you can do is basically you can think about it may follow certain distribution but you don't really know the information of that distribution exactly so it won't be just for sure no more distribution and so so forth so so something that you can do from the modeling side to to get a better solution is to do something we call that you know so like a distribution robust optimization or stochastic optimization so you have to make decisions under uncertainty and with respect to all kinds of possi
ilities that the data could be the true data could be and then you basically minimize the regret or something like that so so that would be some sort of like way the modeling side can handle the other ones that you can think about is you can make your model and the data collection process iterative process so meaning that some part of your model is actually decision dependence and that's an example example in testing case because if I test this region then I know how many people get infected in this region and I will use this data to get prediction but if I don't test this region I wouldn't know right so so some of the application is like this feature like oil like exploration now you have to actually in the order to observe the data and then so the action you make come from a model that takes data but then that data can be model as a function of your decision so so that's something that you can also think about using something we call decision dependant on certainty to model the problem thank you can you say a few model predictions and it's authentic about a tech daughter so I'm sorry so I'm not an expert of the the teri medicine thing I think the the so when you say model predictions I would think that you want to predict it like how many demand we may see in terms of people who need a doctor online and that part I think that data can be pretty trusted like you have a historical data how people go into hospitals and how often they need to go like for those chronic Oh disease patient and you can use those to predict the demand for a model itself it's just sort of like some scheduling problem we open do you have doctors who can do this time you have patients who need these kind of doctors how do you pair them together and then make it work so so that's some scheduling problem and again you know you pay you can send me email and I can send you some papers on that okay so I think about the time and thanks everyone for asking the question and thank you yeah thank you very much professor Shen I this was this was a very nice kickoff to the series and everybody please tune in next week Wednesday at 12:30 for the second talk and a series by Professor Brian Walker G of five statistics
