[MUSIC]
Hi, thank you so much for having me here such a good way of getting people from
different disciplines especially I'm from social sciences.
To kind of talk about very important issues facing our
society and how we can use interdisciplinary methods to
kind of mitigate those issues.
So just a little background on me before I start my
presentation.
So as I said earlier, my name's Anamika Barman-Adhikari, and
I'm an assistant professor in the School of Social Work at
the University of Denver.
And in terms of my research interest,
I'm fascinated with the idea of social influence.
And for the last eight years, I have been trying to understand,
kind of why people cluster in social networks,
both face-to-face and online.
And kind of the psychological, sociological, cognitive,
and mathematical rules that govern how these
networks tend to operate.
In terms of my talk today, I have to add one caveat before I
even get into the presentation.
I am not an AI expert.
I'm a social scientist who has come to value and
appreciate how AI can be used in the public health domain.
And I'm gonna use one of our projects which is known as GUIDE
to kind of showcase how interdisciplinary research,
especially AI in public health and
social work had come together to solve a very important issue
in the domain of behavioral sciences and
specially substance abuse prevention.
So as I said earlier,
I'm gonna talk about GUIDE which is an acronym,
which stands for Group-based Intervention Decision Aid for
Substance Abuse Prevention among Homeless Youth.
But before I get into that,
I think it would be useful to get a sense of the scope of
substance abuse and homelessness among youth who have
experienced homelessness in the United States.
So in terms of the population and numbers, approximately 1.5
million youth are homeless in the US at any given point of
the year, which might be surprising to a lot of you here,
because the US is one of the richest countries in the world,
but we still have stark levels of inequality.
In terms of the rates of substance use among homeless
youth, there's so many things that plague you to experience
homelessness, but substance abuse is especially problematic.
So based on data collected in between 2011 to 2013 in
Los Angeles from a thousand homeless youth.
What we found is that about a quarter of youth engage in
either meth use or prescription drug misuse and about 15%,
13% to 15%, engage in heroin or injection drug use.
And to put these numbers into context why this is problematic,
approximately 1% of youth, equivalently aged youth,
in the US who are housed and
not homeless engage in the same kind of drug use rates.
So 1% versus 25%, that's pretty stark.
So one of the reasons why I'm so interested in substance abuse is
because it is a network based phenomenon.
So social influence is key to understanding why people
engage in drug abuse.
So this is a network map and what you see here is 160
homeless youth who are connected to each other in a social space.
And the blue dots are people, all of the dots are people and
the lines are basically the relationships among them.
And the blue nodes are people who engage,
youth who engage in drug use.
And the grey and
white dots are people who don't engage in drug use.
And what you'll see here is that,
really drug use is a public health phenomenon, right?
All the drug users tend to cluster together,
which is explained by classic social theories.
Homophily, the notion that
birds of the same feather flock together.
And also social learning theory,
if you're around people who engage in a behavior,
guess what, you're more likely to engage in that behavior.
So, however, the ironic thing is that in terms of preventing drug
use, even though networks are implicated in how,
why substance abuse happens.
Networks are also the key to preventing drug abuse.
So what research has found that prevention programs are most
effective when they have a peer component,
because then people get to engage in drug refusal skills or
drug resistance skills in their natural peer environment.
So the way these peer network interventions have been designed
is that people are usually assigned to groups based on
random assignment without any knowledge of how they
are connecting to their peers.
Or also, not an understanding of the social space
where they're engaging in drug use.
Or the other method that is used is that youth get to choose
their own peer groups.
And if you go back to the previous slide,
what we saw is that peers tend to hang out with other peers
who are like themselves.
So a lot of times what happens is that a lot of users come to
get together and that increases their drug use behavior which is
known by the term called Deviancy Training.
And Deviancy Training is a problem in prevention science.
It has been a sticky problem in prevention science ever since we
have been trying to prevent drug abuse among any groups of youth.
So in terms of the question we wanted to address or the problem
we wanted to address through our interdisciplinary collaboration
is, how to assign people to groups based on their existing
network relationships, and also their substance use behaviors.
And the way we saw how AI helped is that with AI,
we wanted to understand whether we could have a predictive
algorithm that could tell us whether or not if we engage in
that intervention, if Deviancy Training would be exacerbated.
Or if you could really use Deviancy Training.
So in the context of AI,
we really saw this as a graph partitioning problem.
So if you go back to the figure that I showed you earlier,
what if we could design an algorithm
that could reconfigure this network
based on drug use as well as preexisting relationships,
in a way where we could optimize positive influence.
So instead of cutting these networks into clusters,
we had to reconfigure people in a way that we could basically
reduce the problem of deviancy training.
So in order to do that, we thought we had to model three
important parameters, and I'm gonna talk about what parameters
we included and not really the simulation techniques,
and Phoebe's gonna address that in her doc.
But in terms of understanding why Deviancy Training happens
and how to mitigate it,
we thought there are three important modeling components
that we needed to include in our algorithm.
First, is Network Influence which I'm gonna talk about.
The second, is Network Dynamic Modeling and
third is Interventionist Modeling.
So in terms of Influence Modeling, as Amalia was saying
earlier, there are competing influences in that perks there.
People are associated with a lot of people and
because of the diversity of these influences they are both
positive and negative influences on people.
So the way we saw it is that, it is linear threshold model.
So basically people get both positive and
negative influences from their peers, and
if the influence signal crosses the threshold value,
then the person is more likely to change their behavior.
So for example, if a user is connected to more non-users
than users, they're more likely to change their behavior.
Also, because this intervention the way we design social work
intervention is that it happens over a period of time, and
over that period of time people change their behaviors and
also their relationships.
It was important for us to model how those relationships change.
And what we hypothesized is that if a user became a non-user,
they're probably more likely to strengthen their
relationships with other non-users and
weaken their relationships with other users.
And that's something we
also wanted to model in the algorithm.
Also there is the interventionist components in
social work.
Usually, interventions are delivered by professionals, and
they are also a key component of the intervention.
And we wanted to basically understand what kind of
relationships the participants have with the interventionist,
and how that changes the nature of their behaviors as well as
their networks.
In terms of the results, what we found is that GUIDE outperforms
the traditional status quo methods of random assignment,
and peer chosen groups by almost 40%.
That's pretty huge, and that's statistically and
practically meaningful.
And what this emphasis is that it's important to acknowledge
existing relationships and network tie strength and
behavior in understanding how we can minimize Deviancy Training.
How GUIDE helps?
In the real world, we are planning to deploy this decision
aid in a shelter in Denver, Colorado, and what we'll be able
to do is to predict scenarios where Deviancy Training is
increased thus, saving time, money, and effort.
Future strategies, real world deployment that I talked about.
We also need to make certain tweaks to our model to
kind of include uncertainties and
we also need to make our model a little more complex,
because right now, we are just modelling three things.
We hope to model other behavioral parameters, and
demographic variables that might also impact substance abuse.
So that concludes my presentation.
I'll be happy to talk to you later and
answer any questions you have right now.
Thank you. >> [APPLAUSE]
>> Yes.
>> In homelessness and drug abuse.
>> Uh-huh.
>> Which peers does?
>> So that's a great question.
So substance abuse happens prior to people becoming homeless, and
it could be a cause of homelessness as well.
So a lot of people become homeless,
because they're substance users and
also because they have to survive in an environment
where drug abuse is considered to be a survival skill.
It could get exacerbated because of the circumstances they face
after they experience homelessness.
So it could be both.
It's a chicken and egg thing.
For more infomation >> Introduction to the public policy process - Duration: 5:47. 
For more infomation >> Family Of Toddler Killed Pleads With Public For Info On His Killer - Duration: 2:45. 
For more infomation >> Amb. Danon: There is a public UN and a private UN - Duration: 4:50. 
For more infomation >> Tesla's Elon Musk in twitter fight over public transportation - Duration: 4:07. 
For more infomation >> McMaster Goes Public With Trump's Military Option Against Kim Jong-Un - Duration: 2:46. 
For more infomation >> Metro Public Works opens new operations center - Duration: 1:08. 
For more infomation >> Mumps prompts extended winter break for public charter school - Duration: 0:33. 
For more infomation >> 12/19/17 Rules, Confirmations & Public Election Committee - Duration: 55:06. 
Không có nhận xét nào:
Đăng nhận xét