Hey everybody welcome today's brown
bag hopefully you can hear and

see me my name is Keith Edwards I'm
the director of the g.v.

center today we're very fortunate to have
our own Gregory a bout as our speaker

before I prefer introduce Gregory let me
just give a couple reminders 1st for q.

and a if you have any questions for

Gregory about his talk please
just post those in the chats and

the moderators will try to relay those to
Gregory near the end of the talk if you

prefer you can also ask us and
we can promote you so you can present so

you can actually ask over video and
audio Just be sure that you yourself

reminder there will be no brown bag
next week so we have nothing schedule so

you all get a week off today it's my great
pleasure to introduce Greg Dowd I think

most all of you certainly already
know Gregory he's been a professor in

the school of interactive computing sense
since for a while I just read his Regis

professor and the Jay z. Liang chair has
been very active in a lot of aspects of

the Krista puting is going to talk to
us today about his very recent very

urgent work on using y.
5 for contact traces of the g.t.

campus of Gregory I'll let you take it
away Ok thank you very much Keith I'm let

me quickly get to the point here of
what I'm going to talk about today so

I'm going to give a little bit of
a historical background the project

basically that was an umbrella
effort that led to the specific work

I'm going to talk about later on this
that this is the Campus Life Project and

how we're trying to use passive
sensing to understand individual and

collective behaviors then I'm going to
talk about what happened in March and

how we so
the challenge that we had on campus

as well as many others on campus to try to
see how we might as technology researchers

is something to support Georgia Tech's
response to the pandemic and I'm going to.

Right a case for using the authentication
logs that happened in our wireless manage

wireless network on campus describe more
about that but the interesting thing here

is that this is a passive sensor stream
that's fairly easy to collect for

a large number of people over
a long period of time so

it has some value but
it clearly has some privacy concerns and

I think I'm going to try to highlight some
of this is the balance between the value

added of using the stream stream of data
against the privacy concerns you have and

I hope you can engage in a conversation
at the end or after this talk offline on

this particular challenge I'm going
to talk about it with respect

to the value added of this stream of data
how it helps with contact Tracy that's I

think what was advertised in this op but I
also want to talk about some other things

that this data can do how can help us with
occupancy tracking at the building and

the floor level and why that may be of
value and how it can help in conjunction

with the surveillance testing
activity going on on campus and

the context racing efforts out
can we give a database assessment

of the exposure risk that
the entire campus Yes So

the discussion point as I said is
going to be privacy public health and

the responsible use of this kind of data
and I want to point out to you that this

is very much a work in progress I love
giving these give you Brownback as

it forces me to collect together a lot
of work that I've been doing with many

other people gives us a forcing function
to get some of that work done and

I'm sure there are some people
listening to here that haven't had that

a lot of sleep or less few days but we're
still plenty of things you can do to help

influence the direction
of this particular.

Ok so I'll come back to this light at
the end I just want you to know that

there are a lot of people involved
in this work faculty members and

I see and in other schools ph d.
students and a large number

of people who over the summer for
the most part volunteered their time or

paid a little bit to help with this
particular effort as we ramped up.

So also a little bit of history
here of the campus life project so

in the 20132014 at Dartmouth
there was a very seminal project

called Student life that they gave
didn't instrumented answer a phone and

through the course of a 10 week semester
they collected passive sensor data on that

smartphone and
also had individuals fill out surveys.

Ecological momentary assessments
throughout the semester

I will give a lot of the details here
except this show that they say that

they were able to show that this
passive sensor stream could predict or

model appropriately academic and
wellness outcomes for the students and

this very much motivated
the larger community.

To explore and expand upon this idea
of how can we use passive sensor

streams in support of understanding
of undergraduate campus.

So Georgia Tech started this campus
life effort with a number of other

partners around the world but

our specific contributions were in doing
responsive real time eating detection and

how that relates to how eating
behaviors relates to mood and well

other wellness factors this is work that
Thomas plots has led with Ph d. student.

American Marsha we also look at social
media as a passive sensor stream

this is work that our faculty member
movement a child who has been reading and

what a lot of her work is about with Ph d.
students.

And the doc that's way but I'm going to
focus on here is the other contribution

that Georgia Tech has made in this area
and that's looking at why 5 as a passive

sensors so instead of looking at sensor
data that's on client devices you know

that you have in your hand or on your
body are mounted somehow we look to

data that's in our digital infrastructure
and wife eyes the specific we're going to

look at but there are plenty of other of
these infrastructure of a status troops.

So let's jump right into talking
about why fi as a way of

being a provider of course location
information for individuals so

as you may or may not know your
wireless devices regularly off and

authenticate with the campus network
if you are connecting to edge Rome or

the Georgia Tech wireless you're doing
that through authentication using your I

see issue id and
password to connect there the wireless

systems heats a lot of this authentication
records and they use that to help

maintain the network if you ever have
any problems not getting being able to

connect You can give your i.d. it's u.i.t.
and they'll try to sort out where and

when the problem happened or they want
to figure out how to balance loads so

everyone gets the quality of
service they use this kind of data

along with other things to help you that
the logs look a little bit like this.

Each line shows an example of some kind
of device trying to authenticate So

this is a amassed walk that we've gotten
and you see on the left hand side

we have a timestamp we have the mast
user id so we're not showing any.

Real Georgia Tech I.D.'s here
on the far right is a symbolic

note of the name assigned to the access
point that you're connecting to and

that's put in a particular format where
it's a building idea number that and

then some region within that building
that's oftentimes a room like

a lecture room but can also be a region
if you have an open plan office area

there we get a tie stamp list of where
particular individuals are with respect to

this Access Point Well it's interesting
about this is that Georgia Tech Like most

campus universities has a pretty dense
network of these access points so

there are over 7000 access
points across the 250 or so

buildings on campus they can get
a pretty good idea of where.

Resolution here of this day 2
of the questions we asked is as

a passive sensor stream meaning it's data
that's collected without users actively

doing anything how useful is that passive
sensor stream for analyzing social

interaction that's what we asked we're not
just looking here at individual behaviors

we're looking at collecting so we started
doing this to have 3 years ago and

we took for example that raw log
that was on the previous life and

here's an example the rows
here are students there

the ids here are masked id and
they gave us kids consent for

this but we're not showing the identity
in each of the dots on here

are just an example of some access point
reporting and affiliation from this

user id whether it's one or
more devices it doesn't matter but we.

Work that out so we have the time
of day and we have for each.

Row here born of the y.
axis we can see that the history

of authenticating to access points
well with a little bit of effort.

We can analyze that data to make
a little bit more sense out of it so

in this particular view of that
data we've tried to connect

regions on these rows to when
individuals are moving so

that's in red or if they're stationary
that's here showing green and orange.

This particular slot and
time is the beginning and

the end of a lecture of a number of people
that we can sense it into our study and so

we can see when people show up for
class roughly and

when they leave after class
those are the green lines or

Zona lines being shown here also what we
can do is we can take any slice of time.

With this kind of data and
we can determine the individuals who

are colocated because in terms of this
figure they all have the same color

of line so all the students in the class
at that time are co-located So

we have a way of being able to determine
when people are near each other and

here is basically defined
by the access point but

a lot of times these access points
are specific wombs are around campus

So what do you do with
this kind of data so

we were very interested in looking at the
relationship between social interaction

that those co-location information I was
just showing and academic performance

by students does specifically we
asked this question the project teams

who meet outside of class perform
better than those that don't

that is the amount of time that it's seen
collects together outside of schedule

class that doesn't have any relationship
to the performance of that project so

we did was we worked with the junior
design class here in our computer science

undergraduate major We recruited
186 students from a number of

different sections of that particular
class in the spring 2100 the mester.

And you can look at the details in this
archive publication that's currently under

review but the result we saw was that yes
the amount of time the project group spend

outside of scheduled class time actually
is a better predictor of how that project

team does versus other forms of predicting
like looking at your evaluations for

example but this is very promising but
it was never the end goal

of this work to just do that kind of
study with you know 200 or so individuals

we want to think a larger scale so we want
to take the results from this study and

say Now what can we do if we look at
an entire semester across the entire

campus population that was around
February March of this year and

we all know what happened around that time
a lot of things basically came to a halt

and as researchers they say that you
wanted to do You couldn't do any more.

But a lot of people at Georgia Tech and
you should be as a as a yellowjacket

you should be very proud of how our
community has stepped up to try to

address making things understanding
the impact of this pandemic and

how we might do we make a safer
return to campus though for

example a line of effort and money has
gone into setting up a surveillance

testing system on campus here that has
a capacity of about 1500 samples per day

I urge all of you if you spend any time
at all this semester on campus you should

register and get yourself regularly tested
by which I mean once a week you should go

to one of the testing sites give us a live
a sample and find out the results but

we also have a website now so that some
of the results of this later that is

basically reported to the community a very
transparently how many people are testing

positive how many people are being tested
right there is that you are you are all

for that if you go to my test I got
checked out even you there's a banner at

the very top of the screen that say
should say that those results as well.

That's best done by a lot of people and
I recognize those folks later but

I'm highlighting here Greg Gibson
the Joshua whites were given a couple of

outstanding public lectures on
the role of testing on campus and

what impact it's been having We have
a number of colleagues at 0 who stepped up

to provide some services to individuals
on campus there's a website covert

Central that got checked out edu that's a
kind of one stop shop for finding out all

the information you can about what Georgia
Tech is doing with respect to helping

the community they've also helped
us to adopt the CMU a defense

app called no hit which has proximity
alerts alerting outfox about that

later I urge you if you don't want to pay
attention to this lecture go and download

the no hit app and in settings set
the community do the easy easy and you'll

be part of the Georgia Tech community
not splay later why that matters.

Now let's talk about context for us so

our health service called stamps here Ben
Holmes the director they've been leading

the effort in coordinating
surveillance testing along with.

Contact racing and they report information
to the partment of public health so we

should get the point here that these are
all pieces that have to work together and

it's it's a lot of effort to do this but
let's talk a little bit about what

the goals are contact tracing
are the 1st you're trying to be to and

counsel people who test positive to
give them help in determining when and

how to isolate until they are no longer
infectious and no longer a risk to those

around them that's probably the main goal
of doing contact or so but the other thing

you do and where contact tracing comes
from is in speaking to someone who is

positive you find out where they have
been during their infectious period and

who else has been potentially
exposed to that who has then 8 times

with a positive case the point is that
you limit the spread by identifying those

contacts as quickly as you can and
getting those people the poor insee and

also to get tested as though that's
what contacts racing is about and

there's a whole system that's
set up here on campus but

I can tell you that we need more
volunteers to help with this context

racing nice a specific plea think we need
student volunteers I think they can play

a very important role in helping
communicate in a compassionate way with

other students to talk about what services
are available and so try to identify.

The police if you feel that the urge
to help the community volunteer

to be a context you can can contact me or
you can contact them hold.

The what is a contact Well there
are 3 important categories of contact

in this particular case one is if you have
physical contact with another individual

touching someone else none of the
technologies I'm going to talk about here

to determine if you touch someone else but
if you live with someone it's highly

likely that you touch someone else of
often want to find out who you live with

or if you have a relationship with
anyone else that you may not live with

those are people that you likely become in
physical contact but the technology that

I'm going to talk about today doesn't
address that the next category which you

probably heard a lot about is if you come
in close contact which is defined as being

within 6 feet of someone for
15 minutes at a time and

there are technological solutions that are
trying to directly address that category

of close contact the next
category is extended contact So

this is if you're in an enclosed room for
up to an hour or so

a lecture with in a in a room a contained
room can be a risk of exposure

if it's over a long enough period of time
right so that's why we have you do things

like wear masks and socially distance to
try to limit the risk in those particular

situations so we've got physical contact
both contact and extended contact.

Question is can we determine these lives
levels of contact with digital technology

already told you know for the physical
contact at least in what I'm going to show

you know for that but we do have
technologies to address the other 2

categories the 1st there are 2 approaches
to this kind of contact tracing or

proximity determination the 1st I'll call
phone based cross imagery alerting and

this is probably what we've heard a lot
about probably since about April or so

where the smartphones essentially dents
the Nereus of other smartphones and

if we can assume that a smartphone is a
good proxy for a person's location then we

can determine that people are close to
each other the other example that I'll

poke is that it's been the focus of our
work has been using infrastructure based

determination of social interactions
specifically knowing when and where people

are connecting to the wireless network on
campus so let's take a little more look

at both of these examples the 1st
approach bone based proximity alert and

the no hit app that I talked about
earlier that is an example of this.

Problem based proximity alerting it is the
one that has been recommended an adopted

by Georgia Tech it was developed very good
Melhem please download it in settings

that the community as lowercase be used
is easy and you will become part of

the Georgia Tech community participating
in this but the idea here is that 2 phones

can sense when they're nearby and exchange
messages about their co-location So

how can they sense that they're nearby
there are idea of ways but these platforms

have a number of different kinds of radios
or communication technologies where you

can detect that from another individual
and detect the strength of that signal and

use that to get an estimate on the range
the distance between the 2 devices

where the using Bluetooth or in the case
of No It moves 2 plus ultrasonic.

Transmissions.

Each key point here is not just that you
can measure the distance between other

phones you can send informations of those
phones to say you were near this but

this particular phone at this particular
time but you do not reveal the identity

of the owner of the phone or anything that
can be traced back to that particular form

in doing this so it doesn't reveal any
information about individuals or so

from the very beginning this
whooshing is privacy preserving and

that's really one of the main arguments
for wanting to use it widespread then when

an individual tests positive they simply
tell the app that they are positive and

it will you know miraculously
traverse through the network and

inform all individuals who were seem to
be in contact over some period of time

up to 14 days in the past say you
have been in contact with someone who

has a possible again no identities are
being really revealed here you were just

told that's why they call proximity
alerting you were near within 6 feet or

close contact with someone who
ultimately tested positive and

then you can take action based on that so

let's assess the pros and
cons of this kind of approach 1st of all

privacy preserving from the very beginning
it was designed so that you cannot trace

the identity of a phone and its owner
in any part of the system is very

good at determining this close contact
category that raising of 6 feet I'm

actually surprised how good it is because
if you it's actually probably more common

to how bad a researcher I am trying to do
something like this a few years ago for

another case and didn't have
that much success of being able

to do the ranging that particular level.

It also if people day when they
test positive it's very fast and

alerting doing the proximity alerts was
a quick way to get to the contacts.

Let's look on the flip side that.

A system like this only works if the
majority of the community is participate

in this case you have to download the app
and you have to say you're in the buzz

community well most popular apps
have if they're really popular they

have an adoption rate around 40 percent
experts believe that you need at least

70 percent adoption rate or something
like this for proximity to learning or

to have an effect on infectious spread
Well currently as of this morning

there were 3680 people in the Georgia Tech
buzz community and there are about

12000 people who are regularly on campus
right now so we're not even that's

close to the 30 percent who don't have
a great adoption rate that's a problem.

You also have to conform as a participant
you have to run the app in an appropriate

way so that's always doing the ranging
estimation turning the phone off on

Android and i o s often puts these kinds
of applications in the background and they

don't work a lot no it has a workaround
for that which is basically a Locksley so

it can still work even though you think
your phone is turned off but you also have

to if you test positive you have to tell
the app that you're positive otherwise

it's no proximity alerting happens
this is what I said before it's not

good these things are not good at running
the background because you know Apple and

Android o.-s. controllers don't want
battery life of being sucked away so

they prevent 3rd party applications
from having the privilege

of running these kinds of
networking solutions when

their church when the app is in the
background of the phone is turned up so

there you see the pros and
cons of this particular approach.

Also add here is your phone always
with you there was a recent

study that was just published by one of
my students just part of just that just

came out this month in fact that is an
extension of studies about how often and

our people close to their
particular phone and

you'll be surprised to know that despite
you might be thinking that your phone is

always near you there's a lot of times
when your phone is not near you and

all this phone based proximity learning is
based on near miss of phones to each other

and the phone being a proxy so it doesn't
always mean that you are near someone is

that your phone was near some other
phone Ok so now let's take a look

at the infrastructure based approach
which are many social interactions so

here remember I showed you before that we
could determine where people are colocated

in some timeframe Let's call that
timeframe of dwelling period and

will say we want to know if people are in
the same place which in this case for

the wife eye is the same access point
roughly speaking the same room or

region of the building.

Are they in that same space for 15
minutes 30 minutes 60 minutes whatever so

given an individual who's tested positive
the parlance for that is a case we can

generate a list of other individuals
who during the infectious period for

that individual sos for some 10 days
we can say who is co-located with that

individual for some period of time and
the question we were asking is how do you

best present that information to support
the people who are doing the manual

time concept tracing investigation
that was one of our primary goals.

No I got to point out
that we are not alone

in thinking about this idea
of executing on prior to

the are working on this area we new folks
that Singapore Management University in

collaboration people University
Massachusetts Amherst that and Hertz and

Hearst sorry about that we're looking at
exactly the same problem we talked to them

about what they were doing what challenges
they were they were doing it's also

the case that this is an opportunity for

some commercial individuals to vie these
kind of contact tracing services so

you can find various companies or at least
advertise that they have this capability

we've worked directly with degree
analytics and I can attest they're one of

the few companies that is providing
services to education establishment

to analyze life I data to be able to
help with educational analytics and

they did a quick pivot over the last
6 months and have ways of being able

to show things like what I'm going to show
you it's a day all the big players in

5 you know the Cisco's NIST Ruba
they are adding this

capability to the management structure for
their software systems and

there are other companies like Vine
technologies that are looking at doing

this in retail spaces there
are approach is to add additional y.

5 hotspots outdoors and it malls so

that you can provide information to people
in the spaces of where is are there dense

reserves in this large indoor mall or
large.

Places like a Wal-Mart they can inform
the company where there is too dense

a population of individuals so this is
clearly when you've got something that's

of value and important lots of people
are going to be going after that's good.

Let's talk a little bit about our records
here the 1st thing we want to ask is this

wife I authentication data doesn't
really provide a good information so

what are ways we can test what we can
predict with this the wife I lost that we

know are true so
we happen to know what the on campus

in role students are those who
are living in campus dorms or

residential housing and
those who are living in Greek housing

right the housing authorities said
Georgia Tech can provide is that they said

they tell us that basically there's
about 6200 units in residences and

about 123-0100 Greek housing and
these numbers here which are just showing

the gold is the Greek housing students
the red is the residential and

the blue is simply just the combination
of those we're getting numbers that

are similar to what is being reported
in terms of people come to campus and

this is just showing from the 1st day that
students started arriving on campus August

30 up until Monday of this week
which is the 24th Ok so we

can have some confidence in this data that
it's telling us something meaningful and

so other kinds of analyses we may also
be able to trust that that's important

to do at 1st.

So here I'm sorry this this is a pretty
busy screen shot and I couldn't show you

the actual working wire wire frame for
a variety of reasons but

this is an idea a mock up of what
you might have an investigator doing

Contact Us And so remember they have
got a case who they need to talk to and

so we're going to provide some details
about that case of the name of this person

when they were infectious what role
they were and other things and

then we're going to provide some
details about where we think

from the why 5 data that case has
been over the infectious period so

this colored ribbon efforts
whop here is just showing times

when according to the wife I data this
individual was dwelling somewhere.

So we can see basically when they stop
someplace on campus and spend some time so

you might be able to use this to refresh
the memory of a case when if it's

particularly it's over a week or so for
5 days which is what's happening now it's

45 day period there tracing back you may
not remember all the places you bend or

and this kind of information can help Quad
the memory of a particular individual

we can also try to identify who the
contacts were people who spent some amount

of time in the same location as the case
and underneath this dwelling timeline

we see a visualization that's
showing ones that are directly

vertically aligned with the a place where
a chase was dwelling those are the people

who spent some amount of minutes and
that's a threshold you can say you can set

say Ok anything over 15 minutes highlights
of me so I get highlighted here all

the people who are in that same location
at that time with this individual and

you could order those are the ones at the
top are the ones that spent more time and

you could do that for all the different
places that someone was dwelling and

the different colors here in this time are
just indicating different kinds of spaces

on campus residential research dining or

a classroom you can also provide
a textual listing of these contests and

order them by various parameters
like overall dwellings though we've

been working with staffs showing them this
prototype and I want to highlight here

if there's anyone here from the n.s.a.
CIA program at Georgia Tech kudos to the 3

former students from n.s.a. CIA that
rats have worked on building this and

they're following the driving through
human centered design process to help us

try to figure out in a kind of her to
sort of her fashion how we might produce

something that's useful for stamps given
what they're trained to do and the tools

that they already use how do we integrate
something that works best with that.

We also what you see back here there's
a visualization a little bit that's not

functional but we've actually been working
with our visit that we you know we've got

a world class visit here Alexander
Johnston asco also polo Chow in c.s.c.

but he's not worked specifically on this
particular one this is a screen shot

I think I might have time to let me try.

To share a different view so
I can show you this real shot.

So here is a this is working on live
data the interesting thing here this

is a fairly sparse visualization because
it goes back to before classes started

this is someone who is likely a student I
don't have the idea here I can it's well

over if it's if you have the time line is
that the line at the top here you probably

can't see this but this tells me that it's
a residence hall I can also tell that

by the fact that it's yellow here I can
move over to when classes started and

I can select a particular well in time and

automatic there were photos of together
all the people that were at that same

academic location as that particular
individual I can do that for any of these

particular spaces what's interesting
here is if you look at this now and

we have mostly online classes and
a relatively small number of students you

don't see that many co-location or social
interaction events if you compared this

to this time last year completely
different situation so that's

a live visualisations will that we've
been working on to help us figure out

how we might hone down to
identify the contacts and

see other patterns like this these
alternative blue patterns that some

epidemic space but they're not happening
at the same time they're at it in

with a different set of people so there's
no overlapping where these kind of trends

pop out in this kind of visualization
you go back to my presentation

so.

What I shown here are 2 ways we
can help with contactors and

one is a kind of case management system
that helps you identify some shows

a dwelling a line in this visualisation
that is mocked up on the left hand side or

on the right hand side we actually have
a working version that you might use to

a place inside here let's talk about
the pros and cons of this approach

though some of the pros Well I said
the problem with the smartphone proximity

learning was the adoption rate and you're
lucky to get 40 percent adoption rate but

people who use campus life eye on
their mobile devices as a very high

adoption rate well over 90 percent
that data sparely easy to obtain So

we work with i Pad researchers met
Sanders and Nathan angle from.

I t to get access to this
data anonymize it and

secure it somewhere under a data
governance agreement with the campus so

that people can't tamper with the data or
get access to it without permission.

We show that this can be used to help with
a case interview you're walking through

a timeline trying to identify who
are potential contacts it's also good for

extended contacts that situation
where you're in the same room for

maybe an hour at a time it's
particularly good at that and

it provides location specific
informational sure a little bit about this

later that can be given to facilities
kind of automatically this album

where have people who tested positive
been spending time on campus so

they can determine what kind of
disinfection or clearly has to be done in

various spaces but let's talk
about the flip side here privacy.

This is a lot of data about
the movement of all people on campus

will students trust Georgia Tech
to use this kind of data.

There's a lot of discussion we can have
about this about today even have the right

to use the state or do they have
the right to use this data if it's for

the important use as deemed
by the administration to

help the campus this is an interesting
discussion for us to have and and

depending on where you fall there what
level of consensus do individuals need

to give so that a case investigator can
actually look at their while on time so

there is an end user agreement for George
Tech life I bet if you look at the details

does give Georgia Tech the ability to
use this data both do monitor the health

of the communication network but also for
other purposes that it deems important.

This is not a good technology for

close contact it will not tell you
someone who is within 6 feet so

it's going to overestimate
the number of potential contacts and

the so the question is if you use a tool
like this as a menu of time sectors or

would it really save you time is all
of this technology worth the effort and

that's an open question still.

So the question I have for you that we
can talk about at the end is this context

if it were the privacy risk to use this
data for contact tracing before you answer

I want to talk about the other uses of
this same kind of data that we can look at

building occupancy patterns and we look
to look at the relationship between

people who are positive on campus and
the risk that they have for

exposing other people on campus now
go through these a little quickly so

we have some time for discussion we can
show the number of unique visitors to

various campus buildings so we look at a
number of buildings on campus on Aug 13th

we look at those same buildings and August
28th and we can see changes so we can see

trends in how many people are spending
time in these buildings over time.

We can look at specific buildings like
c.r.c. and we can look at how many people

are there every hour and Circe now has 3
times that they're open during the day and

you can see the piece but these 3
different types Ok the Circe is not really

that interesting a space because they can
kind of control the flow of people coming

in there are spaces on campus that don't
have that kind of control the class

Commons being one of them so here we're
showing all this past Monday how many

people were in the class Commons building
for at least and minutes every hour

throughout the day and you see it peaks at
around 150 or so in the middle of the day

but we can go further we can actually
look at various floors of the club

because we know different things happen
on different parts that the 1st floor of

the 2nd floor are where a lot of kind
of students hanging out studying and

whatever happened to be so
you might be more concerned about this so

this crash shows in yellow the number of
visitors on each of the 4 of the 5 floors

plus the outdoor and
then in blue how much average time during

the during that visit are they spending
their ranges from 30 minutes on the 1st or

2nd floor up to 40 minutes the time of
the duration as well if you're so but

that means is that they move they.

Stay in place or
connect at the same access point and

then do some kind of movement that
could be movement out of the building

it could be someone who is on
the 3rd floor in an office and

goes to the restroom so they disconnected
come back so don't put too much.

Meaning there except to say that we
can look at this at a finer level of

granularity we can actually look at
classrooms in specific I now want to

talk about a stop here with what I think
is the most exciting piece of information

I want to share their already pointed out
to you that the campus publishes a daily

the number of people who have tested
in the in the previous 24 hours and

the number that were positive so
you can see in the bottom here 2025 people

were tested on August 25th and 82 of those
people were confirmed were confirmed

positive it's usually a 48 hour turnaround
have been for positive in total we can see

here 13050 people have been tested since
they started I think on August 14th and

277000 people have tested positive so

I'll call that the positivity rate
which is 270 divided by 13050

it's a 1.7 percent positivity rate
now this is a little misleading and

this is why I was telling Please go get
yourself tested if you spend any time

on campus because this is an important
number for us to know and I'll

tell you why it's an important number
if we look back to the fall of

2019 on the left and the on the x.
axis here we have that positivity rate and

it ranges from point one
percent to 2 percent.

And using the wife I data based on
Fall 2900 behaviors people going to

classes moving around interacting with
others we can say well if we have

a say a 1.7 percent positivity
rate this bra line is showing for

the entire campus that means on the y.
axis is the percentage of the people who

possibly who put who came in contact
with a positive case over the span

of a week and it says here that over the
span of a week with 1.7 positivity rate

60 percent of the campus was
it suppose now I told you this

is an overestimate because this is not 60
percent of people came within 6 feet for

50 minutes but 60 percent of people I
think in this particular one it's at poor

at least 60 minutes they were in the same
room location which is the extended

contact condition that's 18000 people
on campus we couldn't hope to contact

80000 people and even test those 18000
people to see if they were positive and

whether we should isolate or
porn people but let's look at the data for

this law though since this
is from I believe this is

I don't have the dates of but I think
this is the last week of the 1st week of

classes up until I think it's just the 1st
week and we have about $12000.00 people

that are on campus right now if we
look at that 1.7 percent positivity

rate that we have we see that around
5 percent of the campus is exposed

again it's the brown curve here the blue
curve are just residential students but

it's not important for the discussion
I have here that 600 people.

That's a lot lower what this is saying
is that I have this question are we

actually mitigating risk on
campus by having only online

classes given the mobility patterns
has it made a difference and

absolutely it's made a difference now
there are lots of people are not coming to

campus partly because we have online so
it is what it is right now but right but

as our data shows now that about $600.00
potential contacts on campus and

remember that our capacity for
doing surveillance tests is 1500 so

we easily have the capacity to
take these potential contacts and

have them tested and
that all helps us contain the spread so

targeted surveillance can handle what
we've got going on right now but

again we have to have everyone doing the
surveillance testing and then we can use

this kind of data to actually determine
whether we are doing ourselves right

by the policies that we've implemented
you know having all the classes be

online the 1st we did it make
a difference I would argue this as yes.

As I want to end there and just highlight
some things that the wife I gave it has

some advantages you can identify potential
contacts you can indicate contaminated

locations on campus you can alert building
managers of occupants obviously trends

we can provide as I just
showed data based evidence for

the effect of our teaching policies and
our surveillance testing capacity and

the message I'm giving you here is quite
a positive one that we're doing a good job

at being able to contain things and
we know we have data to back that

up you know we're engineer scientists we
argue with data so the question I have for

the community here is this kind
of capability worth the privacy

surveillance threat of some entities
on campus holding that information and

using it to track mobility patterns to how
do we ensure that only the right people

see this personally identifiable
information the g.t.a. i.v.

how much of the results that we do
are actually aggregate all the occupancy

numbers I showed you don't identify
anywhere and are those useful those

aggregates at the room level at the at
the floor level are those of value I

think they are but for context racing the
identity of the case and the identity of

the contacts is all part of doing concept
tracing will individuals will a case

consent to an investigator being able to
see where the case has been over the last

4 years for the purpose of talking with
the case to see where they might have

exposed that's the question that I have
that I'm interested in finding out.

And there by saying again here all the
people out of work I specifically want to

call out the jackets protect jackets
effort on campus it takes a village view

these things all the surveillance at
testing at the covert central and

this wife I work in
a moderate on an ice way and

you know I'm showing a lot of the
organizations in addition to the people

that I've been working specifically with
and I see we have about 5 minutes left so

if you want to moderate any questions
I mean I'm very happy to take that.

First off thank you Gregory this
was a fantastic fantastic talk

though we have a few minutes left all
asked the attendees please think of

the questions you'd like for Gregory
we've already got one in the chat here.

Becky cruncher and
I'm going to I can work the.

Lottery rater interface here I'm going
to invite her to become a presenter but

she can ask her question so

Becky hopefully you see that I believe
I stop sharing this with you as well

well let me let me just go ahead and ask
her questions and she said in the chat so

Becky says when you do human subjects
research campuses are treated

somewhat differently than other spaces so
just by being on campus or

giving consent to be involved in
research so do you think Gregory that

this might pertain to consent to be in the
works since it is since it is research.

Well you know Becky our vow to you
because of your experience an i.r.b..

We've looked at the end user
agreement that individuals have

in connecting to the y.
5 and you are actually giving consent for

Georgia Tech to store information but
there's this difference between

whether it's allowed to do and whether you
should do it and you know better than me

of actually that kind of balance so really
I think for something like this to work

I need to communicate
transparently what we're doing and

we need to get informed consent whether or
not we have to get that informed consent

I think we need to get it we've talked
to the context racing investigating and

they said that they're concerned that
they're already concerned that when

they talk to a student the student
will not share with them where they.

Don't know anything that makes
that even harder to do and

it's the reason why I want students to
volunteer to be these concepts race

investigators because I think doing to
students is better than administrator or

faculty to students or something like that
so i'd rather than answer question except

to say that you know the answer better to
me but it's more than just whether I or

he would allow us to do it of whether
people will tolerate it happens.

I'm here NASA can I just add one more
thing do you hear me I'm just I guess

yes that Ok so I was just going to add
that you know I think the challenge

the challenge that you're facing
actually reminds me of something that

you know we talk a lot about in r.c.s.
1001.

Which is the sort of the relationship
between privacy in the nation state and

privacy is constantly in dialogue with
questions of say national security and

I think what makes this interesting is
that we may not immediately think of

a pandemic as a national security
issue but of course it is but

you remember how you know concern
people are off the 911 event

that state support in particular I
think often people will feel that.

It's Ok for the nation state have more
access to their materials because they

need that or a higher degree of
protection they don't feel safe so

I think one of the inch one of the ways to
potentially frame this is to think about

how how does this pandemics that in that
relationship that has been in dialogue in

the United States I mean you just all
the way that say well yeah offering

would leave you with is a great point
in fact the exchange of information

from Georgia Tech a positively identified
individual to the Georgia heart department

public health is not a violation of HIPAA
because it's used in the in the case of

a national emergency a national public
health emergency so I think what matters

here is when and if anyone pronounces
that we no longer have a national

emergency with respect right spend them
right that's an important decision.

You know we were reluctant to declare
it in the beginning now that we have it

there are certain things we can
do on the public health side of

things to essentially support it but
you know the part of public health

is already overwhelmed with this context
rasing Georgia Tech has to look out for

themselves here we need to empower
ourselves let's do this and

the staff folks are able to look at
a person the person identified or

health information the volunteers
they have not as clear

so you know we have to have these open
discussions about whether we allow this

information to be revealed because
we're trying to help ourselves and

you know it's not just a political or
a safety issue it's economic as well

you know if Georgia Tech has to shut
down we are going to lose even more

money than we have already lost so
it is in our own selfish interest to do

things in a way where we can maintain
operation and our educational mission.

And it's also an important
social experience right the hope

I should start now that other people
have questions but I see that so

we have the obvious question asked by
an associate dean if anyone chooses to use

this technology term who spends which
night in who's got her absolutely.

We know sweet level information for

most of the residential spaces we don't
know that about the Greek houses.

So there is there are those kinds of
challenges of what and this is precisely

the kind of information that a student or
any individual may not want to divulge

is someone over the phone or in person and
certainly would not feel comfortable

with people being able to determine that
already so this issue of consent here.

But didn't think about it if I did bite
someone to spend the night with me

probably I had close contact
might have had physical contact

if I'm responsible individual I will
inform those people so maybe for

those kinds of situations we don't need
to worry about it because there's another

means by which someone is going to
be informed maybe I hope again or

usually go I'm going to be a psychic so
we're almost out of time when there's

been a request to share the slides
if you want to show the slides and

shooters over to us will will get
them up on the website if you're not

willing to just let us know but
there is one more I missions Are you Ok.

To let people see out there you're likely
going to be a town hall a real kind of

town hall or something in the coming weeks
where I'll be asked to give a smaller

version of this if it's a fairly
accurate descriptor and Lucy you had

a question in the q. and a and I've since
it invites you to ask your question so

if you'd like to do that
you'll have to a new toy.

Don't the interest of time maybe I'll just
ask her question if she's if she's not

able to do that do you think we can better
leverage group surveillance testing by

basing them on people's occupancy patterns
Well I think that's the next step

is to try to be more strategic in our
surveillance asking right now we're in

a mode of encouraging everyone if you
spend time on campus go once a week make

it part of your regular schedule the next
thing we can do is if we don't have

the capacity to do that we can target
places where outbreaks are happening so

absolutely I think this kind of data
can be used to try to like that that

the slide I was showing at the end there
is that we can identify who are the people

that were the potential contacts and that
set of people in the interesting thing is

that under the threshold for daily
capacity so we could be strategic about

that so I think we're just about out
of time rush a little bit over time.

I suppose if people are actually getting
involved should contact you chair and

task like Well thanks again Gregory
this has been a great talk we appreciate

you hear thank your banks that everyone
who helped me put this together but

over in the Bears.

This is fantastic Ok stopping there.