– It is now my great pleasure to introduce Professor Ilias Bilionis. He joined Purdue also in 2014, same time when Marcial came and has quickly established a
predictive science laboratory, PSL, as he calls it. His research spends the
general interdisciplinary space of design under uncertainty, spending a range of
socio-technical systems. His research works is based
on exploiting physical models to inform statistical and
machine learning techniques in order to overcome inherent limitations of engineering systems
due to the high costs of information acquisition and limited number of observation.
His research work
establishes new directions at the intersection of machine learning and artificial intelligence
with engineering systems. His research has also
been funded significantly by governmental organizations
like NSF and NASA and DARPA but also by industry in particular, Ford and Facebook and some others. He's a natural collaborator, who's able to make diverse contribution and add value to a wide
range of research programs. I think that's really
very unique about him. He's truly an interdisciplinary
or cross disciplinary, researcher for us in the school with many different collaborative efforts. So in addition to all of his research, he has proven himself
as an excellent mentor of his graduate and
undergraduate students. He was presented with the
outstanding faculty mentor in mechanical engineering
graduate students award and has engaged in numerous undergraduate or has engaged numerous
undergraduate students in his research activities and giving them opportunities
through SURF, as well as, or Bottomley fellowships
that we have in ME and undergraduate research assistantships. So furthermore, his teaching efforts, highlight a significant commitment to developing state of the
art educational models, effectively integrating novel
educational technologies and tools with the fundamental tools.
And as a result, he has been recognized as an outstanding engineering
teacher, three times by the College of Engineering. So I would like to mention here on the teaching effort at the end also that he's instrumental in
developing our data science, big data, Data Analytics course for our ME students currently, which is a major undertaking. So with all that, I would like to hand it back to Ilias. Sometimes I'm moving back
to the German pronunciation, I'm sorry about that. But I really like to welcome him here and I'm really looking
forward to his remarks.
– Thank you, Eckert and very happy to be
here particularly with, during the same time as Marcial. Marcial was probably,
the first friend I made, when I kind of got into a new department. And so I'm really really
happy to be honored at the same time as him. Congratulations Marcial. All right, so when we get started, I'm gonna tell you a
few things about myself so that you can get a feeling of who I am and I also like to take this opportunity to honor a little bit
where I'm coming from.
So I'm coming from
Greece and in particular this little town, a little
bit outside of Athens. It's called Aspropyrgos, white tower, it's like 20 kilometers outside of Athens. And it's an industrial hub, has basically some, basic, what? How do you call it? So oil refineries mostly and steel manufacturing plants. And it's really, it's actually
a really bad part of Athens but it looks beautiful from away and you see it has nothing to do, it doesn't resemble Indiana at all. So there is a beach, there are
mountains on the background. It's always sunny but it's
also a little bit smelly because of the oil refineries.
My mother's coming from
the North of Greece, a town called Thessaloniki
as you can see on the top. My father is coming from the
middle of the Peloponnese from a little village called Lagadia. And this is where I go, when I go to Greece during the summer, I basically go to that little village and I do my work from there, completely undisturbed from anyone. The village has about 1,000
families living there. And this is my favorite Greek Island. I'm not gonna tell you which one it is because I don't really
want you to go there. But if you're really
interested in knowing, you can send me a personal message and I may tell you. Okay, so this is how I was educated, I started in Athens, at the National Technical
University of Athens and I started Applied Mathematics. And to be honest, I
started Applied Mathematics because in high school,
I didn't know what to do. And I didn't know what
Mechanical Engineering was, I didn't know what civil engineering was? I didn't know anything really. And Applied Mathematics seemed
like not making a decision.
So it was a little bit out
of luck that I picked it and because I didn't
wanna make a decision. And continuing not making a decision, I also did a PhD in Applied
Mathematics at Cornell. And initially I went there,
wanted to study Finance but my arrival to the U.S coincided with, Finance and Mathematics in particular. My arrival in the U.S coincided
with the crisis of 2008. And there wasn't a lot
of excitement back then for Finance Engineering. It was actually blamed quite a bit. So I started experimenting
with more engineering projects. I was good in probability and statistics and I happened to find my, talk with my later PhD advisor, Professor Zabaras who helped me understand about how I can apply what
I knew about profitability in engineering systems. So I liked that a lot, so I decided to do my PhD on that.
I then joined Argonne, again, I worked in the intersection of, Engineering plus Statistics
or Physics plus Statistics at the Mathematics and
Computer Science division. And I finally came to Purdue as part of a cluster higher
of Predictive Science and Engineering. And so I came to Purdue to
actually do collaborative work and that's exactly what
I've been doing so far. These are my intellectual heroes and so Von Neumann, for various reasons, mostly formulation of Game theory and the groundwork on Decision Making. Richard Feynman, one of the
best teachers of all times. A person that I listened to on my Walkman on my bike during my high school years. I listened to his lectures. E. T. Jaynes, one of the pioneers of the
Maximum Entropy principle. Alan Turing, one of the pioneers of AI. I. J. Good, my favorite statistician and Judea Pearl, whose a person who has
formulated causal inference. And these are some of my favorite books, I like to read a lot. I don't have a lot of time anymore, mostly because of my toddler but I like to read a lot of history and focusing on particularly pre-historic and I also like Biology.
So one of my favorite
books on biology is the, "Selfish Gene" by Richard
Dawkins and pretty much all of Richard Dawkins books. All right, so what is the mission of my lab as a so-called
Predictive Science Lab? In one sentence, it is to create, artificial intelligence technologies that accelerate the pace
of engineering innovation. So I wanna help engineers
do their job faster, without having to do the dirty work of programming stuff and basically accelerating what, the way they design things, okay. And the way they bring data into whatever it is they're doing. Now, this is my philosophy and this is the backbone
of whatever I'm doing. So I develop communication channels between physics and data. So yes, I'm doing data
science and mostly learning but I'm doing it in the
context of a physical problem. So I'm using the physical
equations, the PDs, Parson Division equations, ODS
or other physical equations. And this is all done under the following
communication protocols.
So there's probability theory, which I think of as an extension of logic. As the language of science with an additional layer of causality, expressed either implicitly
through the physical laws or through graphical models. And I use modern Marcialan techniques to basically represent
certain of the quantities that appear in whatever we're doing. All right, these are my
overarching research themes. You'll see later on many projects but these are the core
problems I'm working on. So there's quite a few things under the category of
theory-informed machine learning, high-dimensional
uncertainty quantification. So when you have a model that has parameters that are uncertain and these parameters are high
dimensional, think about, let's say, not knowing an entire function. An entire function, is an inherently
high-dimensional quantity. So how do you quantify your
uncertainty about functions? How do you appropriate it through the rest of your physical model? Filtering and calibration. You are observing part
of a dynamical system and you want to infer the entire state of the dynamical system. Perhaps there are parameters you don't know about in
the dynamical system, which you would like to calibrate.
This has applications in
control, digital twins. Sequential design of
experiments and simulations. So you have a fixed budget
to do a certain number of experiments or a fixed
computational budget and you want to design your
simulations or experiments in order to achieve a certain goal. Like maximize something, estimate the probability
that something happens and so on and so forth. I design algorithms that guide you into the selection of these experiments. Fault detection, diagnosis, and prognosis. You have a system that can
break down a certain way. How can you, by looking at central data, figure out when something has gone wrong and make predictions about
how much time you have until you really have to fix it. This is very similar to the
filtering and calibration but it has some nuances added to it.
So this is one block of things I do, the other thing, the other block has to do with modeling human behavior. And I'm really talking
about modeling the human as a disturbance in an engineering system. So in particular in the context of my
buildings applications, I develop models of humans interacting with the lighting system
or with a thermal system. Humans making decisions about
the thermostat set point, for example. And I'm also interested in
humans as decision makers, inside engineering systems. So once you increase the
complexity of your system at a certain level, you're
gonna have to introduce humans because the current state of the AI, does not allow for full autonomy. So you're gonna have to bring
humans to close the loop and make, have them make
the difficult decisions. So how can you deal with that? So this is an overview
of the projects I have, ordered from more physical to more human.
And we're not gonna go over all of these, I'm just gonna mention briefly at the very top, we
have basically Physics, design of materials and as we go down, we go to a little bit
of engineered systems, electric engines, combustion engines biomedical applications. And we go to even more complex systems, like acceleration hub that's project about which we're gonna talk about and smart building projects. Of course you may ask yourself, do you really know all that stuff? No, I do not know all that stuff, okay? So I'm not an expert in pretty
much any of these fields, where I'm an expert on
is on bridging the gaps, between Physics and theory
and data, all right? So I have developed the skill
to understand the Physics in a wide array of fields.
And I can help people connect with data and I can help them formulate
decision making problems and I can help them quantify
uncertainty in there modules. All right, I have 35 guard,
Purdue faculty collaborators, which says a lot about the
way I like to do my job. I have, I'm collaborating,
I have at least 14 from the mechanical
engineering department. I'm working with people from
electrical, civil, aerospace. I have written proposals with many more. The good thing is that we
haven't won all the proposals, otherwise we'd be in trouble in terms of the amount of time, we have to carry out the projects.
And the two projects that
I would like to mention to give you a more concrete
idea about how I'm involved. So the first thing is a Smart
and Connected communities, project funded by NSF, where I lead the data science
and mechanics design efforts. So the goal of this project
is to go to communities, low-income communities that are, some of them are subsidized and to design a thermostat, a smart thermostat that
gives them information, gives them feedback that
incentivizes these people to reduce their energy consumption. And the idea here is that, the amount of income these
guys spent on energy, is so significant that even
a little bit of savings, will have an impact on
their quality of life. So what I do is, I work
on the part that designs, what sort of feedback we
should give them back. And this is a mechanism design problem, mechanism in the sense of game theory. We try to find which incentives
maximize a community goal while at the same time the
individuals are acting, sort of selflessly and
maximizing their own utilities.
And the other project I
would like to mention, is the resilient extraterrestrial
habitats project, where I lead one of the three thrusts, the Awareness Thrust,
which is responsible for, using the censored data
to develop an awareness about the state of the habitat. Where is it right now? How likely is it that
things have been broken? And what are the actions
we should take next to mitigate any issues? Now, let me motivate a
little bit this latter part and what exactly, I mean
by developing an awareness but I'm gonna touch
upon what are the issues that we're trying to address
in the next five or 10 years? So I'm gonna use an example for that. Am I running out of time? No, okay. 'Cause I saw you turn on your camera, I was a little bit… – We have a little bit more time, go ahead but, you know… – Okay, sorry, let's say
I have one minute, okay? How about that? – No problem, no problem.
– All right, so let's skip
this completely, okay? If you're interested about learning, more specifically about this project, please reach out to me. I'm gonna read this more slowly than I originally anticipated. I'm on slide nine if you see, I just went very slowly. Okay, so just wanna
mention my graduate class, ME 539, it's called, Introduction to Scientific
Machine Learning. So far I had 350 graduate
students taking this class. So this is data science for engineers. It's basically, specifically
using physical problems to teach data science. And that's the difference
between my class, for example, with Stanley's class.
I don't go as deep as Stanley. I try to connect to the level of my, I'm assuming my engineers know about, differential equations and
that they don't know about, probability so much for optimization. I'm also developing the
undergraduate data science class. All right, I wanna thank my students, these are the people who did all the work, past and current students. I wanna thank my mentors and these are not all my mentors. The very top are the Greeks, Professor Papanicolaou
taught me probability. Koutsourelakis, taught me
about vision statistics Zabaras taught me how to
do basically a faculty job. Jitesh is the first person
I wrote proposals with. Karava is the first person I
had successful proposals with and we're continuing our
collaboration in a full-blown way. Professor Dyke, my main mentor, she taught me a lot
about how to do the job and also about how to mentor students and by watching, Rhoads is also my other
mentor by watching him teach, I improve myself considerably.
And finally, I wanna thank
my family, my grandparents, my father and my mother on the right here. My little brother, he's
three years younger than me, you see us right down on the bottom, who taught me how to tolerate
people that are different. My brother is gay and he was, he helped me a lot to understand
a different perspective. So I knew that he was
different ever since, we were in this picture together, ever since I was six or seven years old.
And it was great growing up with him and watching him develop
into the man he is right now. And of course my family,
my wife and my son. Without my wife, I don't know
if I could have done anything. She and I, we managed to be together from a distance for more than six years, her in Greece, me here in the U.S. And it's been a great journey and really, without the stability that she had provided to me, I wouldn't be able to accomplish anything.
All right, that's all I have. – Great, thank you very much. Ilias, wonderful presentation,
wonderful remarks. I like, I love the personal touch. Any questions from the audience for Ilias, you can either unmute yourself or write something on the chat room. I can start out, right? I would be interested to better understand how you model humans. I'm kind of a thermal systems engineer and I model thermal systems, right? We have some basic characterization in form of first principle
to model the equipment, then some environmental inputs and we get a performance of the system.
But how do you model humans? That seems entertaining to me. – There is a first principle
formulated by Fort Norman in the '50s. So the principle is, people
maximize their expected utility. All right so they have some sort of, they have some goals and some preferences and these preferences are
expressed as a function over earlier choices and they try to maximize that objective.
Now, the problem is that you
need to go a level above that because, you know, do people really know, what are their preferences, question mark. Even if they do, do they really, can they really maximize the objective and you relax this a little bit and you go into Simon's approach, which is that they
don't really maximize it but they're satisficing in the sense that, when they find something
that is good enough, they just make the decision, okay? So they have an objective, they have some preferences, they don't try to maximize it perfectly but they find something good enough, they make a decision.
And that can be expressed mathematically in the language of probability theory and that's how we do it. And at the end of the day, really I'm not, it's a matter of, whether or not it matches
the experimental data. It's a model just like
your thermo science models. And you could say that it
does match the data sometimes and there are lots of
examples where people deviate from this behavior. – Great, thank you. Any other questions or comments? Who would like to chime in here? Yes, Ervin go ahead. – Non-technical question. But I did wanna comment that, you know, Ilias put me on to a very nice wine and I've been going back to it. So, you know, he's got a very nice but also effective taste in wine.
So I really appreciate that,
great colleague as well. – I have some excellent
suggestions for Greek wine that I have discovered. – Not Retsina I hope. – [Ilias] Not Retsina, no. – Okay, good. – Top of the line, affordable, I'm not gonna mention them here because there are in limited supply so… – So the thing is, you've
got this favorite island that no one knows about. You've got the favorite
wine nobody knows about, what other secrets do you have on? (man laughs) – I can not tell you.
You should be careful
with things like that because islands can be
crowded very very easily. So you can go to Santorini
if you want, okay? – No, no, no, too many
people, too many people there. No, no, no. – I have one more question for you that may be helpful to others
who are on the call, right? How do you manage collaboration with 35 different
collaborators across Purdue? And I mean that in a serious way, actually, I'm gonna
talk to Ervin about it. We might have a, should start an award for like the most type
of collaborative person in the college or so.
You probably are on the top of that list but I think, right? We have all kinds of different characters, mentalities, and you know,
you do need to manage that. And I think there's some
logic to that as well, right? – There is. So you need to do spend the time to understand their application
and their perspective. So I do that, I will, when I
work with Karava for example, on buildings, I'll sit down
and learn about buildings. I'll learn about the HVAC system, heating and cooling, the equations they use, I'll learn about how they
design their experiments. I spend the time to do that. At the same time, I'm interested in, instead of just doing my own stuff and write my theory papers, I'm interested in solving their problems. So I'm like, I don't come to collaboration with a technique that I want to apply. I want you to tell me, what is your problem? What is your problem? And then we solve your problem.
That's my attitude. Now this has put me a
lot, a little bit back on my, let's say theory endeavors but at the same time I feel really happy, solving actual problems, which is something I didn't do before. Because in all my theory papers, the examples are rhetoric examples..