Welcome back to the Oxford Mathematics Public Lecture Hall Edition. My name is Allegory and I'm in charge of external relations for the Mathematical Institute. As usual, special thanks to a sponsor execs market execs market are leading quantitative driven electronic market maker with offices in London, Singapore and New York. This support makes it possible for us to provide quality content. As you know, we are now entering a new phase of the coveted nineteen crisis following
a lockdown and the hope that the numbers of new infected case may be manageable. Governments around the world are tentatively trying to unlock the countries at the mathematical level. The first phase of the disease was governed by population dynamics of infection, a topic that was discussing approach previous public lecture that you can find online. The second phase of the crisis will depend on our ability to manage and understand social
interaction and human mobility. As you may have heard in recent news, one of the main ideas is to track individuals through their forms. What kind of data do we need? Or do we use it? What type of mathematics is involved in this technology? And what are the ethical and legal challenges regarding privacy? To discuss these ideas? We have invited Professor Hono
lobbyist tonight. Originally from Belgium, Reneau is no professor of networks and non-linear systems at the Mathematical Institute here in Oxford and a fellow of Somerville College. Reneau is one of the top network scientists in the world, and he has been working on mobile phone network for many years, is now leading a collaborative effort to put anonymized data from multiple countries and service providers in a coherent
way so that it can be integrated in various models. If you want to ask a question, please send it in via social media and we will collate them and send out answers in the next couple of days. Thank you, Reneau. Please start. No. Hello, everyone, my name is Robert and then the professor in the Mathematical Institute of the University of Oxford. So today I'm going to talk to you about phones, smartphones and how to use them to
fight against coving 19. So clearly, if you want to steer a country out of the very complex and tragic situation in which it is right now, you can proceed blindfolded. You. You need data. You need data to guide your actions. And you did that to estimate the prevalence of the disease with sufficient tests. But you also need data to capture the way each of us is moving and interacting. S At the end of the day, we are the ones carrying the disease and propagating them to our contacts
by our actions. So as we see today, mobile phones and more precisely mobile phone data are a very nice opportunity for us to try to capture these mobility and social contacts. And the end can be useful in different ways that can be useful to try to understand better and to model better dynamics of the disease that can and will be useful in order to assess the impact of some social distancing measures on mobility,
for instance, such as lockdown and to track it drill down. And they can also provide some data driven support for exit strategies by government and decision makers. So all this talk is going to be organised as follows. We're going to start with an introduction about networks in epidemiology. Will then talk about the use of mobile phones to estimate contacts and mobility and the very important question of privacy.
We will then look more specifically of the use of mobile phones in order to understand and fight against coving 19. And then we're going to conclude. For the last few months, there's been intense research effort worldwide in different disciplines to understand better coming 19. These efforts score from the understanding of the incubation period of the disease and the proportion of asymptomatic people after they're infected.
To understanding the physical way by which the spreading takes place, mostly by Arrison or droplets or surface contamination. But a very important aspect as well is the social component associated with the disease and trying to capture all of these social contacts that are existing inside the society and trying to understand the way these contacts and these physical contacts are actually either slowing down or accelerating
the spread of the disease. And rather, you, in your situation about the importance of contacts and social and social contacts and networks to understand the spread of a disease. Let's look at a very different example taken from Kevin, 19, but taken from sexually transmitted diseases. So in this example, in this author. Well, in this paper, the authors went to a high school and by means of interviews and by means of question, as they build what they call to be
sexual network. And in a sexual network, you have the nodes represent students and the links represent sexual relations. And when you look at the sexual nitto, you clearly see that there's a lot of diversity in the way that people are interacting with each other. But you can also see that's two people may actually infect each other. Not by means of direct connexions, but by means of indirect connexion to
a path of connexions. And because of this notion of bath, you can define the notion of connected component, which is a very good at critical aspect, because it will tell you that if a node belongs to such a connected component, it will have the possibility to infect a large part of the whole population. So clearly, networks have an opinion, it's a very important concept and very much related with each other.
But on top of a network, you also need a model for the way the disease is evolving. And I would not say too much about it. There is a very nice talk by my colleague Robin Thompson will give a talk and there is a link here just below towards this talk. But let's just have the basics of it. So when you try to move the disease spreading, typically the way you do is you divide the population two compartments and these compartments represents a different different states of the evolution
of the disease. You start by being susceptible, you potentially become infected, and then you go into a recovered state. And in this example, you just have three states. But actually
these mothers can be generalised. And this is an example of a mother that people use for coordinating 19 with many more compartments to try to predict more accurately the different path of the stages that the but that a person may follow from being susceptible to being in the final state of being recovered or deceased in this case. And so clearly, this mother is probably more accurate, but it also enforce many more parameters.
And it raises thoroughly the question of what would be the right data to use in order to fit those parameters. So when you when you have a mother for your population and for your disease, well, then a very important quantity is the Soka, the so-called repetition. Repetition? No, these are not that you might have seen quite, quite often, actually in the newspapers. And what is not, it's the typical number of infections caused by an individual. And clearly, if this number is above one,
we would expect to have an exponential growth of the disease. If it is below one, we would expect of an extension of citizens. And for Kobe 19, this number is estimated typically between two and three, meaning that if you want to stop the epidemics, we need to try. We need to find measures in order to bring this onward below one. And one of these measures is social distancing, trying to decrease the contact between the people and the hands to bring this number to one.
Now, when you want to take some, like, strategies against the virus. Well, typically you first need data. And the most basic data would be this one here about the total number of reported case by day. But clearly, between this single curve, you have a very complex system with many people living their life, interacting with each other and possibly propagating the disease. And as we see what networks would be very important, you know, to capture
this richness that is behind this car. And when you try then to act upon the disease, well, that is two different types of approaches. First one one's looks looks from today and goes back into the past and tried to trace back the contacts that's infected people had in the past, while another Bush tried to predict the future course of the epidemics. But before we discuss these two different
options. Well, let's talk about contact networks for for us for a minute. We've seen a sexual network, but what about the contact that talks about the physical meetings that each of us has every day? Well, clearly, my interactions on the scene every week, the people I meet that I met last week will not be the same as the ones that are really next week. Some might be the same, but some others not. There is a lot of varied in the contacts
that I have. Do I think the symbols? Do I see the same share and the same shop and so on and so forth? So it means that we clearly have a first issue is that how would be be predicting a network in the future if we only know the network from the past? We also have the problem of uncertainties. Is the fact that how do you define a contact clearly? Typically we define it to be an interaction of a certain of it so that people are within certain number of metres, within a certain number of seconds.
But you have a threshold here. And whereas the threshold is matter of discussion and also this threshold may depend on the environment by your in a bus or outside in the park. And finally, even if you have a very good definition of a contact, how do you measure these contacts in large populations? It seems unrealistic to have the exact contact of the whole population. The UK
on every day. So as we'll see what mobile phones can help us to measure contacts but will not try to match exactly these contacts will try to measure certain aspects of the resulting networks that will be important for the disease. So let's talk about the victories. So, as I said, contact tracing, what you try to do is actually to find to go back into the past. And so let's assume that today you found find out and someone show some symptoms of of a disease. Well, very rapidly.
You try to identify the contacts that you had in the past in order to isolate this person before himself has the chance of transmitting the disease. And at times. Time is the essence here. So this can be seen from this blood on the left where you can see the number of infections that someone causes as a function of the of the days after, which is infected, depending if this is as a symptomatic a situation,
symptomatic environmental or asymptomatic. But the most important here is to see that if you were to be able to identify someone and taking out by quarantining them just after one, two, three days after you is infected, actually many of the infection that he would have caused otherwise when that happened. So it means that this is a way to break the spreading chains of the disease and make Anacleto decreased. On the note and then repetition number from its natural value to a value that would be
below what? So another approach is not to look into the past, but to look towards the future and to look at the situation. As of today and to try to predict what would be the future course of the disease. So clearly, this is something that is very important for decision maker to know what will happen tomorrow in a week if she wants. She wants, for instance, to
want to put the right resources, the right place at the right time. But actually, these mothers are also used to try to imagine what would happen in the future if certain decisions are taken. Opening schools, for instance, or banning certain types of shops. Now, when you want to build these mothers based on this, I are, for instance, when you will need some sort of a model of the connectivity of the craft. And in it's more simple scenario, you will assume that
the system is basically fully connected. It's a mean field situation where everybody is connected to everyone. So this is an approach that has the advantage of simplicity with just a few parameters. But this is clearly very interesting. I am very unlikely to meet with the same protein with people living nearby than people living in Manchester, for instance.
Then on the other side of the spectrum. What you try to do is to catch the exact network of contacts and then to do some agent based modelling about whether people are going to be infecting each other through these contacts. So this is clearly something that is very much refined. But as we say, it's there's a problem. What is this network to be used? We know the network from the past, but we don't know the network of the future.
We know that there is a lot of viability. And to try to capture until well, to get rid of these very weighty one ways to protect somewhere in between. Not to look at the exact network and exactly what is connected to, but instead to to group nodes according to some information. For instance, here the doctors together, the children together and
so on and so forth. And instead of trying to estimate what is the connexion of one specific doctor to one specific child, a child, you're gonna be trying to look for the total number of connexions between the doctors and the children. And by doing so, you're going to be building some contact matrices between different rows or between group, between different groups of people that can be stratified either by these roles like here or by that age
or by their agenda. And even if we know one, we assume to have a lot of IBT four pairs of nodes because of the law over the large number. We can expect to have that between groups. We will have numbers that would be readily stable a time. So there is a second aspect that is extremely important. You want to mother the spread of the disease and it is
geographic space. So if you were to look to go back into the 14th century and to to study the way the Black Death was basically propagating across Europe, you would see a wave, a wave black in the sea where you would have this wave from coming from Turkey entries through France and then to the north. And you have this very nice but deadly wave propagating inside Europe. Well, nowadays, we travel and move very differently than we did
back then. We have in particular planes that make us take shortcuts from one place on earth to another one. And actually, these shortcuts are going to change radically the way these diseases Miss Pretz spatially. And to illustrate this, there's a very nice article by the Bruckman where they were looking at source that started
in China to. And they were trying to understand what was the what would be the relation between the time when a disease would appear in a country and different measures of distance. What you can see from the left is that actually if you were to try to probe this time of the first appearance to the
physical distance and kilometres, you don't seem to have very much of a signal there. What if instead you were to probe this time with respect to a natural distance, a distance measure measured in a network where you would have cities as nodes and weighted edges based on the number of passengers, aeroplane passengers going from one place to another one would actually have the network distance gives you a very,
very good linear relation, meaning that networks seem to be more predictive than physical distance to understand and predict the spreading of the disease. So so actually, when people try to incorporate this spatial information into a modelling format, quite, they will now consider other networks and not contact networks as before, but what is usually called meta population networks where the nodes were represent locations
and the edges would represent flows of population between nodes. And what do we basically assume now is that you would have no people would be physically moving inside this meta population network, interacting with each other and potentially spreading the disease. So what have we seen so far? We've seen so far that actually. Mothers are becoming much more and more complex. And clearly, because of their complexity, we will need data in order
to parametrised and to fit the parameters of these mothers. We also need data. If you want to monitor the effect isn't the effectiveness of distancing measures. We take measures. Oh, well, are they implemented? What is their impact on the actual behaviour of the people? This is something that needs to be measured either to strengthen these measures or to replace
them by a measure that would be more appropriate. And finally, as soon as you think about mothers and parameters, you need to have day to day updates on these parameters because clearly the weather people move and the way the people interact is very different today than it would be tomorrow. And the other infant last week, due to the way that people adapt to the disease and to all of these
distancing measures that are implemented by governments. So let us know. Look at the different ways in which people have been using mobile phones as a way to capture at a large scale such mobility and contact patterns. The ones that you need to use actually in order to fit these images and mothers. And that's also discussed
very important question of the privacy associated to these data. So clearly, for the last 20 years, more and more of our behaviour and actions are leaving traces, digital traces in some databases. Every time you click something on Facebook, you send an email. You you just use your G.P.S. Well, some of this information is going to be stored or marked by
INS in some databases online. And actually, researchers have been trying to explode all of these digital traces for the last 15, 20 years in a field called computational social science, in the interface between sociology, computer science, applied mathematics. Now, there's been wonderful research that people have been doing showing all sorts of ways in which these datasets could help us to improve society.
But it's also been recognised that these data as it can be dangerous and there can be very dangerous for our own privacy and even for our own personal liberties, freedom. And for this reason, a few years ago were implemented by women who were pushed by Europe, was proposed a very strict regulation called general data protection regulation in order to protect to protect individual people from, for
or for the use in the unwitting use of their data. And I know very strict regulations on what you can or cannot do if you want to collect, used and or share personal data sets. But I still wanted to say a few things about privacy that are quite important for the rest of this stock. So the first one is the fact that's actually data that could appear to be very harmless,
can still reveal a very personal aspect of your life. And to start this discussion, let's have a look at a blog post by the famous physicist Stefan Wolfram, where he looked at all of the emails that he sends over a period of 10 years. And what he looked at wasn't the content of the message. He just looked at the
time of the day when the email was sent. And you had this very nice picture here where you see that those are picked up by every team, but also some regularities that you see, for instance, that between 1996 and 2002, there was a shift when you were starting to write more and more emails during the night and keeping modern. Did they actually interpret? That's a very, very active field of his of his life when he was working on a book for a couple of years and completely shifting
its working schedule from a data shift to a nightly shift. So I could not resist the will to try to do it on myself. And I didn't look at 20 years of data. I just looked at two weeks of data, one week from January before the lockdown and one week of May after it was implemented. Now, you clearly see here that there's been a completely complete reorganisation in the way I organise my working schedule, where in January
I was sending most of my emails in the morning because in the afternoon, even less in the evenings. Now, I sent almost no emails in the morning anymore. Many in the afternoon and also a lot quite late indeed. And actually, the mechanism that can explain this change of behaviours, in my case, simply homeschooling. Schools are closed in the mornings. My wife fiche to our kids, meaning that most of my work has been shifted from the day more towards the night, basically.
And actually, it's kind of revealing. That's even if the data is very harmless. It's just about the time when I send my emails. Still can't say something about my own life, right. If you were to look at these data, you could physically assess or you could predict that I do have
kids. But also, if you were working in advertisements, well, you kids think that I might be slightly sleep deprived due to my late working hours and I might be interested in a new brand of I don't like fancy coffee and sending it advertising that I might be willing to respond to. So, as I say. The time when I send an e-mail is not very important policy, but still it can be sensitive.
And this is even more so with even more private data and mobility and contacts are even more private because that can really be really very intimate aspects of our life. So a second point that is very important is that no one is neither. And actually, you could share your data. Even without knowing it and without winning it. And this is exactly what happens when country to country to Analytica could get access to a huge number of users without actually having contacts with a
small number. And so basically what works is that about three thousand users took a creese organised by companies that Analytica providing their own personal data to the firm. But actually, by the way, Facebook worked at the time, it did not only provide their own information, but also the information of their friends. And by this network effect, basically Cambridge Analytica was able to capture to capture personal data, about 90 million people,
while only. Targeting directly a very small number of its users. And these network effect is very important because, as we'll see, this is something that could play a role. When you start to collect data about contacts and also about location. So a third bond, which is also extremely important, is a point of anonymization, and indeed, usually when you deal with personal data, everything's anonymous. So you should doubt that about
me instead of having Roman objects. Would just have a random number. But still, it's sometimes possible to do that general mass data. If you start to couple different data sets together. A very good example here is the one from the New York City taxi day taxi data. So about five, six years ago was released. A huge data sets about the way taxis were travelling. New York City, you would know that the taxi started from a certain time, going to be an at another time
and you would know the price paid but paid by the person. Also, the tip that you gave away seems pretty harmless, right? And actually, this data was used by research and it could be used
in very positive ways. And for Eisen's, researchers showed that by means, by using this data, that it would be possible to organise some shared taxi rides and almost where everyone was basically having had almost the same tablet travelling time, but reducing the number of trips by a very significant amount, hence decreasing pollution and traffic. But at the same time, people also tried to capture that data sets to a
paparazzi dataset. And knowing that a certain actor could carpet a certain place and living the cab at another place, they could look at fairly private information about the tips that these stars were given to get High-Tech Tips or notice. So, of course, here, this is not very important. Right. But this shows that it's it's impossible. It's in principle possible to determine, again, mass data as soon as the taxi
trip is unique in the data sets. And you can somehow identify this trip with some excellent information that you may have access to. So now what about maybe. Well, actually, when you tried to track lability by means of phones, you don't always needs smartphones. And actually, let's start with very basic, very basic functioning of the phone, the mobile phone. So actually, whatever your phone, the phone you use and or Nokia or the latest iPhone, well, is
going to be interacting with cell towers. And these interactions are actually some of these interactions are going to be captured in databases of mobile phone providers. And these are usually called call data recorders. And every time you make a phone call, receive a phone call, send a message that's in your line that's going to tell you what does the cell to which you are connected to. There's going to be information about what you're collecting. There's new information about the time
when this section was made. And in other case, we're interested in location and mobility. Well, clearly, knowing the cell to which you attach, attach gives you information about your location. And this information is going to be more or less precise, depending on the density of the cell towers. And clearly, in cities, in urban environments, you have a high resolution of the order of 100 metres. Why? In rare rural areas, you would
have a resolution of about one kilometre. But still, by tracking down all of your locations during the day, you can track down some of the mobility and your mobility behaviour. So. So, actually, well, it's too late. I'm sorry. It was last year, July 2019, vulcanise a conference where researchers used and analysed such datasets in all sorts of context. Some of them more industrial ones, others purely about arithmetic, but also others
in terms of the potential applications for the social good of these datasets. And just to give you an example of the kind of positive application that could be made for them. Well, that was a very famous paper in 2011. Of researchers using coded records in order to predict the wave population would move after the earthquake in Haiti. And using those in order to sort
of predict the way cholera would be propagating inside the country. And this is just one paper, amongst many others, where people try to use coded records as a way to track population movements and hence to track the way diseases would be spreading inside countries. Let's go now to smartphones. And for smartphones, well, now you may have apps, but also G.P.S. data that would become available. So one first interesting approach is when people are
using participatory syndrome surveillance. And basically the idea is to get instead of nap on your phone, where regularly you would be telling whether or not you have some symptoms of a certain disease of flu, for instance. But these datasets are very regular, regularly updated. You have a regular update of the status of patients that basically sensors inside your country and you can have in real time maps of the way symptoms are propagating inside
the country. Symptoms that may be precursors of a certain disease. There is also a very nice project that I would like to mention. It's a project that was basically organised and done in Copenhagen in Jitu. And indeed, You, which is a technical university, is a group of researchers getaway 1000 phones to students, freshman students, the deal being that you receive a phone, but you give away some of your datasets. And these datasets could be information about your location,
information about your Bluetooth connectivity information. We are actually in social media and so on and so forth. And actually, this data says this dataset has been extremely influential because it was one of the first times that that there was a lack of comprehensive data that was shared
with the community of researchers. And that allowed to explore the different ways by which people were moving and communicating with each other, but also the techniques that could be used in order to infer some information from mobility and contact data. So I just my childhood broke with actually Bluetooth is actually something that people use very often,
they look to estimate contacts in population. How does it work? But actually, if you have a mobile, a smartphone, your laptop, typically it has some form of both those implemented. And Bluetooth is a way for information to be spread on very small distances to connect your phone to your earphones, to connect your computer to a keyword and so on, so forth.
And the thing is that because it has this short distance radius, it's also something that people have been using or trying to use for quite a long time, 10, 15 years at least, in order to try to trace contacts between people. If my phone detects another phone nearby, it's probably due to the fact that there's a contact between person and person. You should just keep in mind that's route without just an opportunistic. Solution, we are using Bluetooth. Not because it's been designed to estimate
contact. It's not been designed for that. But we use it because it's implemented by default on many, many faults. It means that we are trying to use a technology that is available in order to estimate contacts. But states also know that none that it's very difficult to estimate precisely the contacts. What is the exact distance between people, for instance, or whether or not they are actually looking the same direction or looking to each other?
So it means that this is one way to measure contact, but it's clearly in that idea and actually in the literature, people have been trying to use other ways to capture contacts. And a very nice one is this project called Social Patterns, whether they were using RFID where they gave away some badges to people and actually these badges us as done in such a way that only one people are sufficient to close to each other and face to face with an interaction be captured.
So this is a very nice data set. There've been experiments done in hospitals and schools in many situations, and it's really also improved a lot of understanding about the way the people are physically interacting with each other. But unfortunately, it requires some specific devices that people don't have and it couldn't be implemented at a population scale. So to finish this kind of talk, I just wanted to mention something else and the fact that,
as I say, it's mobile phones have G.P.A. And actually the truck position fairly regularly, regularly on time. So not every five seconds to drain your battery. Very, very fast. Actually, there are some some tools, existing operating systems that were basically the phones captures significant changes in maybe in position. And basically, you have a sequence of posts where the person has been in the course of the day. Now, usually these data can be collected by certain apps. If you give, you
get your agreement for that. Now, there is a company called Kubic that basically collects these datasets from different apps and uses them in a marketing environment. Now, it can be using for marketing purposes, but it can also be used for good. And they have a programme called Data for Good where they give away such G.P.S. tracks, geep just G.P.S. position of people for researchers
while, for instance, interested in repeating energy. And this we'll see many of the solutions that people have been implementing in in recent times against coving thinking are actually based on such Cubitt data. So just so that, you know, keeping data in the US, it gives us about five to 10 percent of the others in the country. This number is much lower in other countries, including in the UK. So so what about the use of such technologies in order to understand and to fight
again? Coving 90? So we're talked about contactors and actually it's been proposed and to use mobile phones in order to accelerate and to improve on connectors. So usually contact tracing is done manually by means of questionnaires, which may slow down quite a lot. The process of the identification of potentially infected people. And actually, there've been some some applications, mobile apps, Devlins, for instance, in Asian countries that have been used
in order to accelerate this process drastically. And there's a very nice paper from our colleague Christopher is over, actually. They look at this. This contract is by apps from a mathematical point of view. And sure, that indeed it is possible to decrease the opposition number below this value of one by using this automated contact tracing applications. Now, clearly here in
in their ordinary paper. So they make a creation, but they also estimate that you would need about 60 percent of the population to use the app for this application to be sufficient in order to to to make this repetition number could go to zero zero and ends for the epidemics to die off naturally. 60 percent is a huge number. So just to give you an idea, in Singapore, the project traced together, committed it, completed it about 20 percent off of usage and closer to
here in Norway. They also had some experiments where they also used applications to track contacts. In that case, the contacts would basically be so with G.P.S. collection at the same time. And even if authorities claimed that there was a very good accuracy in tracing, it only covered about 20 percent of the population in the test zones. So the thing is, that's what this application is clearly something that is that is going to be important.
It's going to be very useful, but it probably won't be useful on its own. It won't be a silver bullet that is going to kill the disease by itself is going to be once on a solution about medicine of many of the solutions that they're going to be bringing down this production number below below its value of one. So actually, when this kind of application was proposed in Western countries and in the U.K. Well, one word here and this sentence was very controversial,
was this notion of central sovereignty. And actually it means that's part of the information would be sent in central servers. Fransen It would be used by the NHS in order to dispatch the information of contacts based on a contact with Person B with Bluetooth. Well, isn't this mapping from person to person B would be basically organised and decided in a centralised fashion.
And this question of centralisation was something that became very controversial because that these sorts of other ways to do the same kind of task by means of decentralised. So in one way would have a centralised server with the some of the information stored even anonymously in another one. Everything is done with the information, stays on the phones and creating this controversy. There's been many papers in different countries, as you can see, where people questions the level of basically the
intimacy that was revealed by such applications. So first of all, it was questions. Question the guarantee for the precise proximity and as we say, route, which is not ideal to measure the proximity between people. For instance, the identity of the book, the signal is different for different phones. So people with a recent iPhone will be more often detected than people with an alderman, for instance. Also, it excludes part of the population. About 30 million people in the U.K. don't
have a smartphone. And that could provide some for sentiment of safety if the application. Nothing happens, it means I'm safe and I can behave normally. But the most important was that there was a key concern that this epidemic could create and let you legitimise a Soviet surveillance tool that would then be continued to be used afterwards. And so that's the reason why I think that's that seems to be like a consensus nowadays. That's. The data should be kept decentralised.
The mobility that at this very, very. The contact data. This is very, very personal. Should remains remain decentralised. Instead, instead of going to a centralised option. But what is very important here is that there is clearly a need for clarity and pedagogy, because at the
end of the day, the people are going to be using. Yeah. And so you want to foster the trust and to justify the trust that the broader public is going to give to researchers and to the app developers if you want them to adopt. It's a very, very high level in such applications. There is also a poem that is also quite important, you should remember, about this network effect of privacy.
So in this network effect of privacy, we we basically say that's it could happen that you don't want to reveal your personal information or Facebook, for instance, but your friends reveal it. Well, let's assume that you have such a contact tracing up where you want to reveal your contacts, but not your location, but your friend, your contacts reveal their locations
where clearly because your friends reveal allegations there regularly when they met you. And so they reveal Passey, the places where you've been and there've been some very nice results were actually it was possible to show that just a tiny fraction of phones revealing locations could reveal the locations of a large number of phones if contact tracing was implemented with partial G.P.S. locations. So
not the technology that people are also using are these called data records. And for these quality records actually, well, for instance, a few months ago, there's been an agreement by European mobile operators to provide data to researchers to help monitor the way the people are moving and adapting their behaviour to social distancing, distancing measures. And there have been all sorts of monitors that have been made available to the broader public with also to decision makers
to properly quantify the way people change their behaviour. This is an example from Italy, for instance, where you can see the way the mobility is decreasing depending a different phases of the disease, but also the decisions that I think are taken by the government. There is a similar project taking place in
Germany where actually when you look at the curve, you see that there's been a decrease in the mobility of the people. And then actually these decreases is relaxing, which is usually cool because this quarantine fatigue where people start to be tired of following these very strict and boring social distancing measures and slowly increase their mobility
even if the government recommends not. And actually, there is also there is also a project here in the U.K. that is run here at the University of Oxford, Oxford, by some colleagues, were again using coded recalls, actually. They also looked at these changes in mobility and could also distinguish four different ages, but also for
different Janda, for instance. So people have been doing these kind of analogies with quality recalls, but they also did it with G.P.S. traces using this Kubic data that I was mentioning
before. For instance, a very nice article in The New York Times where they showed that actually the way that people change their mobility was very, very different in counties with stay-at-home orders, where there was a very, very drastic drop in mobility versus counties without any kind of stay at home order where there was
a G.K mobility, but clearly not the same. So you clearly see here that indeed decisions, political decisions can have a huge impact on the changes of mobility that the people may have. So, again, for these dubious data that are provided by Kubic, well, here in the University of Oxford, in the same project, they also have access to to to some of these Kubic data and also carry on. And then it is about whether people are moving and adapting their behaviour
by means of these deeply G.P.S. information. So far, for G.P.S. information, it's not only Kubic that provides data. There are also other sources of data.
And there's, for instance, this very nice project from Google where they also provide information about changes of behaviour, focussing that case on the way the people change the type of place where they go, for instance, which is a decrease in the visits to retail and recreation places or grocery and pharmacy or parks, which is also some very important information for governments to have if they want to understand the places
that are important, but also to try to track down the way the people are adapting their behaviour and the places that might become places at risk for people to be meeting with each other. Another type of data that people are also using would be Facebook did so in the case of Facebook data. Facebook is also also has a data for good initiative where they provide to researchers and, of course, information about the number of users located in a certain location,
in space, in a different types of attack times of the day. And there's a very nice project from Copenhagen, Denmark, where they also look at these Facebook death and try to understand better, though, the way the population has been shifting from certain
places to other ones. And in this verdict from Copenhagen, they could see that people were going out of the cities, for instance, an observation that was all sorts of New York, for instance, where many people from New York were shown to leave the city and to go more to the countryside, for instance, to the seaside in order to escape the city before the London actually took place. So one last type of data that is also available would be participatory
surveillance. So in participatory surveillance, this is exactly the kind of thing for a system where you're standing up and on a day to day basis, you report the symptoms that you may have. And all of these symptoms are also with certain information about your age or your gender, but also your postcode collected together.
And this is a project from King's College, London, where they collect this information and they have information about the symptoms, possibly before the people actually develop the disease, are reported as having the disease, which is a way to get access to. Potential number of disease of people with the disease even before they are tested in well before they go to the hospitals and could also be something implement them and using mothers.
So talking about mothers, let's go back to these sorts or so. As I said, if you want calibrates mothers for epidemic spreading. Well, these mothers are very rich. These mothers have very, very many parameters about the way the contacts take place, about the way the people are moving around between different places.
And actually more and more groups are using some of the data that I've been presenting based on Kubic G.P.S., based on quality records, for instance, in order to try to feed these mothers and to improve the modelling and the predictions that the mothers are going to make. So just to give you an idea of all of the ways by which this can be done. Well, this is a new situation where we are. We can see like some imaginary trajectory of two users
going from home to different places during during during the day. We would have basically all of these Verano cells are associated to certain cell towers. So that would be something for quality records. And he up and then you have a certain number of bones that somehow represent the trajectory that a person has during the day. Of course, there are many missing bones, but because you only have information
when the person uses his phone. But this is better than nothing. Well, actually, from data like those with more than two customers, what you could clearly identify hotspots that would be
cell towers were there would be a huge increase of the population. That's a certain time. For instance, Krost barks when it is a sunny day, for instance, you could also identify the origin destination matrices, how the people are moving from one place to another, wondering the day and do it in on a daily fashion and tracking down how these origin destination matrices are other in the course of time, which would be something very important for this metabolic population.
Mothers, for instance. You could also try to estimate contact matrices. So for the contact measures, this is something that is much more difficult because, for instance, in the case of Colditz Records, you just have resolution of the order of 100 metres. But still you can try to estimate potential contacts when you have people being in the same cell tower
for a sufficiently long time and potentially having interactions with each other. And this is an approach that people are exploring the how to use these evolving densities in space in order to try to estimate the evolution of the contacts between the people and find that you could also estimate what is the person of the time that people spend at home or at work, because clearly each spend 50 percent of your time at home or 99 percent of your time
at home. This would have a very, very different effect on the potential contacts that you had and the potential ways in which you will be propagating the disease. So here are all sorts of ways where you can use this call data reporting that to basically provide guidance and help fits parameters of these Methow population mothers, but also from these contact niches that I mentioned during the first part of the talk
and actually about the contacts themselves. Well, actually, for the contacts, some people are actually using some of these G.P.S. data in order to try to estimate contacts. So, as I said, when you define a contact, you need to define that two people are within a certain distance during a certain amount of time when that case, they don't define the distance based on biological parameters or physical parameters, but they use the best that they can with the data they have.
In their case, that would be 25 metre. This is clearly large, but this is the best
you can do with this kind of methodology. In that case, within New York, this group used to be data Gibbings data in order to try to estimate how contacts estimated as people that are close to each other for a certain amount of time within twenty five metres hold, this number of contacts would be evolving and you see that there's a drastic drop, which is again, something that can be fed into mothers in order to take into
account the way the contact between the people are evolving in time, but also as a way to track changes in contact patterns when some of these social distancing measures can be relaxed. So given or all the above. So let's try to find a conclusion here. So. Clearly, we're at a time now where people are trying countries are trying to find the best way to relax social distancing measures. So that's the economic and social life. And countries can start again. And yet keeping
the keeping the disease under control. And here I just wanted to show you two very different types of solutions that are implemented in two very different countries in terms of culture and also legal systems in France and in China, too. In France, you have a system right now where depending on the department we live through the place where you live, you will have different rights in terms of mobility. Lives is different.
Yeah. Right. In terms of the shops you make visits or the amount of time you can spend outside and you have this day to day updating of maps of the country. If you live in a red region, you have to stay home, basically, if you're living in a green region. You are much more free to move and to and to and to moving freely fashion. So on the other side of the spectrum. So in China, there is a system called the IP Health Code,
where here everything is based on the use of mobile phones. And there is there is no GDP are in China. And actually an application is collecting information about your mobility, about your contacts, about many aspects of your life. And then this information is fed in to some machine learning algorithm. And depending on your profile, either gives you a green coat, a yellow coat or a red coat. So clearly, you have two different approaches, one that is more geographic in terms
of groups. One that is very much personal and using extremely personal and sensitive data. So the action to the right would be unthinkable in European countries for legal and also cultural reasons, wouldn't be willing to provide such datasets to governments and to and to have these data decide whether or not in a machine learning black smoke, white, black, black box way, we can move on that. But on the other hand, the left's solution is or is not is also not ideal
either. It is very cold drains. It blocks huge parts of the population, while actually maybe only some very small, smallest subpopulations are at risk of spreading the disease. And so clearly, there is this there there's a question of trying to find the right intermediate solution between these calls green and these very fine solution, a solution that would be sufficiently efficient, but
yet respecting the privacy of all of us. And as I've been trying to argue that mobile phones and all of the data that can be collected by mobile phones could be a way to try to go in this direction. So the collection of these data need to be done in the privacy aware setting, respecting regulations and look and also respecting the trust that the people may give in researchers and
public health workers. And I think that if you want to go this direction and this is a direction that people are moving on to right now, you need to have trust and communication between different actors. The government clearly researchers, mobile phone companies would be willing to take their social responsibility to make these data available for for the common good, but also human
rights organisation. We need to have a very transparent description of what these data sets could be used for and how it would be used. And I think that it is only by having this conversation and this transparent opposition between all of these actors that we're going to be reaching an inefficient solution that is acceptable for for the whole community. So so to conclude, I want to thank you, first of all, for listening to me. Thank you very much for spending
this time with me. I also wanted to thank my collaborators for these Oxford Impact Melito, the one that I mentioned where that is run by much Yassky Yan and Adam Sanders from the University of Oxford, as I am. And I also wanted to thank many of my collaborators with what I've been writing, an article which is freely accessible. These go and have a look at it where we describe some of these ideas that I've been discussing today and and what we do.
It's by writing and I clearly hope that you will be interesting in reading it. So thank you very much. And specif.