Post on 29-Aug-2019
Kirill: This is session number two with machine learning expert
and entrepreneur, Hadelin de Ponteves.
(background music plays)
Welcome to the SuperDataScience podcast. My name is Kirill
Eremenko. Data science coach and lifestyle entrepreneur.
And each week, we bring you inspiring people and ideas to
help you build your successful career in data science.
Thanks for being here today and now let’s make the complex
simple.
(background music plays)
Welcome everyone to a very special edition of the
SuperDataScience podcast. And you might say this is only
the second episode Kirill why is it so special? Well, that is
because today I’m bringing you a very good friend of mine,
Hadelin de Ponteves.
And in fact, we only met a couple of months ago over a work-
related project but we instantly bonded. You know how you
get that feeling sometimes when you meet somebody whose
values are very similar to yours..their aspirations and
ambitions are very similar to yours and you instantly
develop this strong connection and understanding of each
other and that’s exactly what happened and I’m super
excited to invite Hadelin to this podcast and we had a great
chat. Hadelin - you should know about it. This guy is a
machine. So for the past three years, Hadelin has been
sleeping only three hours a night. And then once a week he
sleeps for eight hours to catch up.
For us normally like myself I sleep eight hours everyday
pretty much. This guy is so driven, so passionate about
achieving his goals, about working, about learning about
data science, about improving his skill set and also he’s an
entrepreneur so about building businesses and about giving
back to communities and creating value for the world. He
doesn’t even have enough time to sleep.
How cool is that? How interesting of a person do you have to
be to sacrifice your sleep to build things and create things
and improve. He’s also a very very inspirational guy.
So, Hadelin got two masters degrees because that’s the only
way he could actually achieve them, by sleeping three hours
a day. I was about to say a week but that would be an
overkill.
So, three hours a day sleeping and therefore you got two
masters degrees. One of it is Machine Learning so that’s a
great and interesting field of data science and we talked a lot
about machine learning in fact we go deep into machine
learning. In this podcast you will learn about regressions,
classifications, clustering, association, rule learning,
reinforcement learning, deep learning, national language
processing and much more. So, lots of stuff.
Hadelin it’s not a surprise that he’s worked at companies
like Canal Plus which is a Canadian competitor of Netflix.
So, you’ll see how Hadelin went about developing some or
recommend models – models that recommend customers
what movie or what show to watch next based on their
previous experiences.
Also, after hearing all these you won’t be surprised to learn
that Hadelin of course worked on Google and there he
worked with terabytes and terabytes of data.
Also performing sophisticated machine learning and creating
algorithms and resolving challenges so maybe some of the
stuffs that you do. With Google, some of the products you
use with Google actually were created and worked on by
him.
In this episode with super exciting. I just can’t hold back my
excitement about this. For the first time ever, we reveal and
exciting and brand new machine learning project that we are
working on together with Hadelin.
So, it’s really cool because I am here in Brisbane, Hadelin is
in Paris but over the internet, over the miles and miles and
miles of space, we’ve been working on this massive project
we’re going to bring to the world very, very soon.
So, you’ll be excited to learn about it, you’ll be one of the
first people to hear about it in this podcast. Definitely check
it out. Something that we hold very dear to our hearts.
Also in this podcast, you will hear some very interesting and
philosophical discussions about the future of machine
learning. Because Hadelin had so much exposure to do
machine learning and AI and things like that, he’s got his
own views about what’s going to happen to the world. In
fact, he’s actually writing a book. It’s going to be a fiction
book but about how he sees the future of the world in the
next twenty to fifty years and I’m pretty excited to see that
come out. Hopefully that’ll be published sometime in the
near future but otherwise we talk about how robots and AI
are on one hand very positive but on the other hand can be
dangerous and something that I’ve never discussed before –
what kind of career implications that can have for data
scientist? What kind of new jobs that can create in the
future? So, controlling AI, AI security, AI oversight and
things like that. So, very interesting discussions and that
space as well.
And at the of this podcast, we talked about some
aspirations, what inspires Hadelin to move forward, who he
follows, some movies that were both very interesting and
exciting about so definitely check out. I think some of the
movies that we’ve talked about were very interesting and
actually life-changing for a data scientist and finally, Hadelin
recommend his favorite book.
Call away to get started and without further ado, I bring to
you Hadelin de Ponteves.
(background music plays)
Kirill: Hey, everybody! Welcome to the Super Data Science
Podcast and today we’ve got an incredible guest, Hadelin De
Ponteves.
Welcome Hadelin! Thank you for joining me on the show.
Hadelin: Thank you very much. I’m very happy to be here.
Kirill: Awesome! And right off the bat Hadelin, tell us what do you
do or where have you worked before?
Hadelin: Today I would characterize myself as both a data scientist
and an entrepreneur and in the past, I’ve tried many things
before finding my own way, before finding my passions.
I tried with finance in the past and I didn’t like it because it
was all about analysis and I needed to create stuff and I
found out about data science, and I found out that there was
a huge potential for creation. You can let your imagination
go very far and we can use data science to create these
things. So I started to be passionate about data science and I
became a data scientist.
So, I first had an experience at Canal Plus which is a French
company, competitor of Netflix. I actually had a big chance
there which was to build the recommender system based on
data science and machine learning. So that was my first
experience as a data scientist.
And then I had a second work experience at Google where I
worked as a data scientist as well but rather on the business
side. I was implementing some machine learning algorithms
or data mining models for business to create added values for
business.
Kirill: Awesome!
Hadelin: That was my second experience. And then, I became an
entrepreneur.
Kirill: Awesome!
Hadelin: I had this thing to become an entrepreneur and that’s what I
am today.
Kirill: Beautiful! So, everybody who’s listening, you can tell why I’m
so excited about this interview. Hadelin actually has so much
experience in data science and we’ll get to his background in
just a bit. And also, this is a person that worked with Google
so this is a person that has actually seen the frontier of data
science, what is exactly happening on the brand new front of
data science and the most exciting stuff that is happening in
the world right now. So, that’s what we’re going to be digging
in today.
Before we continue, I’m super excited to talk about your
entrepreneurial quest because we’re actually doing some
work together in that space right now. So that would be
really fun to share with the listeners.
Can you tell us a bit more about your background before we
get more into exactly what you did at Canal Plus and Google?
Just tell us a bit more about your background. Where did
you come from? Where were you raised? What kind of
education did you participate in?
Hadelin: Okay. So let’s start from scratch. Let’s start from the
beginning.
I was born in Paris. I was raised in Paris as well. I’ve been
living nearly all my life in Paris.
I found out that I was good in Mathematics in high school so
I decided to pursue my education with a Bachelor in
Mathematics. Then once I got the Bachelor, I started in
engineering school in France.
So, French engineering school are quite particular because
it’s very general. You study everything. I’ve studied
everything from Mathematics to Philosophy, passing by
Economy, Finance.
Kirill: Wow! That’s a lot.
Hadelin: Yeah. It’s quite particular. It’s not like you specialize at the
beginning like in other countries. But, in the final year you
choose a specialization.
As I said, first I was interested in Finance so I’ve tried
Finance but then I quickly realized that I needed more than
that, more than just analysis. I found out about data science.
Then, I finally chose to specialize in data science. That’s my
final year of Engineering, specialize in data science.
Kirill: Sorry, if I’m going to interrupt you quickly. Because data
science is such a new field did you actually have the data
science faculty or like a course on data science or did you
put courses together yourself to create this?
Hadelin: No, actually I was very lucky on this one because my school
just started a program in data science for the first year. I was
the first student of this program. I was very lucky on this
one.
Kirill: Awesome! How did it feel like being the first student? Was
there a lot of things that they could have done better or was
it a good program right away?
Hadelin: It was a good program but I think that some things could
have been done better but that’s totally normal. It’s the first
year. it’s like a try. We’re part of this program as a person
who help to build this program. So yeah, I was very happy to
have this year.
Besides, I did this year in parallel of another program in
business. It was quite a challenge as well.
I have to say I’m an autodidact. So I learned a lot by myself. I
self-teach a lot.
Kirill: What is that called? Autodidact.
Hadelin: Yeah, autodidact. Is that what we say, right?
Kirill: That’s awesome! I haven’t heard of that ever before.
Hadelin: Yeah. I do a lot of learning by myself. So, in parallel of those
two programs, I also did a lot of books, read a lot of books
about data science because it was still about discovery.
The program was good but I found a lot of different stuff
outside that was very interesting as well. It was very new.
There was like this new wave that it was awesome to take it.
Kirill: Wonderful! After completing that degree, did you right away
were you able to find a job or did it take some time?
Hadelin: Yes. Right away I was able to find a job because there was
this new wave of data science in France. And in France,
everybody is looking for data scientists.
Every company is starting their department of data science.
Even consulting firms, for example, I was called by the
Boston Consulting Group. They were starting a new
department in their company about data science.
Kirill: Yeah, that’s lovely. I actually have two friends working for
BCG. One of them was in the data science division. I don’t
know what the other one is doing, actually.
Hadelin: Okay. Every company is starting a data science department
or data science team. So, there’s a high demand in data
scientist so I had no trouble finding a job.
My first job was in Canal Plus, the competitor of Netflix. And
it was very machine learning oriented because I have to build
a recommender system. It was a great experience. And then,
Google happened.
Kirill: Google happened.
Hadelin: Yes. Google happened. It was a great experience as well
because it was very challenging. The team was great. The
mission was great.
Kirill: What did you do at Google?
Hadelin: I was a data scientist in the business team. I was using the
data to help make good decisions. I was analyzing all
Google’s data and there was a huge amount of data. We can
really call big data. We have a lots and lots of data. We were
working on Hadoop, big data systems. And I was trying to
find some patterns in the data. So I have to use machine
learning and data mining models to find some insights, to
draw some insights and help the business teams and
managers to make the good decisions about Google strategy
and their product, that sort of things.
Kirill: When you say Google because Google is so broad – there’s
Gmail, there’s Google Drive, there’s Google Search. Are you
talking about a specific service of Google or something more
broad?
Hadelin: Yeah, that’s a good question. I was working for anything
related to mobile.
Kirill: Mobile.
Hadelin: For example, I was working on the click through rate on
mobile. We had all these metrics comparing desktop to
mobile. I was actually implementing a machine learning
system to optimize the click through rates on mobile. That
was my mission most specifically.
Kirill: Okay. This is where we get to the fun stuff because the
project that we’re working on right now, you and I together is
really involved around machine learning and personally I’m
learning a lot from you about machine learning.
So, can you tell us a bit more? What is machine learning and
what is this science all about?
Hadelin: Machine learning is very broad. You can use machine
learning to do a lot of stuff.
First, you can use it to make some predictions. You can
predict the future. You can predict some behaviors. You can
even predict some future that is going to happen based on
the past, based on the information you have on the past.
And you can also find some unknown information. Like,
you’re looking for something and thanks to machine learning,
you can discover some logic into a field that you’re studying.
So that’s called clustering. When you don’t know what you’re
looking for, you don’t know the answers, you don’t know the
categories of the segment that you’re looking for.
There’s also interactive machine learning. That’s what I did in
Google. I was actually trying to find an algorithm that
chooses the best ads to place at the right moment so that the
customers are more likely to click on the ads. That’s a
reinforcement learning.
And there’s this huge field of machinery called deep learning
that’s very powerful and that you can use it to do facial
recognition and you can use it to build the lying detectors for
example.
Machine learning is very broad. You can use it to many
things and many applications.
Kirill: In that case, like somebody just off the street who is not into
data science, who is hearing about these things for the first
time, machine learning sounds very sophisticated. Sounds
like machines are learning something.
How is that different to just the normal trend line or just a
normal like a pattern recognition system or is that a sub-
clause of machine learning?
Hadelin: Actually, machine learning englobes everything. Machine
learning englobes pattern recognition. It englobes deep
learning. Well, to my definition.
According to my definition, machine learning is like a big
science that englobes many sub fields of machine learning
and pattern recognition is one of them. Clustering is one of
them. Deep learning is one of them.
But then, I heard a lot of many different definitions about
machine learning and some even say, “artificial intelligence”
instead of machine learning.
So, I guess there is not a unique definition. I guess that for
some person in the street, I would tell him about artificial
intelligence because it’s more popular term. The machine
learning is more scientific term.
Kirill: Okay, I see. But in your view, in your more knowledgeable
view from the things that you have studied and learned, how
would you distinguish between artificial intelligence and
machine learning?
Hadelin: For me it’s the same. For me artificial intelligence and
machine learning is the same because artificial intelligence is
a machine that is intelligent enough to do things and by
doing things, it has to learn how to do it and that’s exactly
machine learning. Machine learning, it’s a machine learning
to do stuff by itself.
For me it’s the same. But if I had to precise the difference, I
would say that artificial intelligence is the cool term.
Learning is the scientific term.
Kirill: Yeah, totally. And so just hearing about your definition of
machine learning and your experience with it, it sounds like
data science is going in the direction of machine learning or
that for somebody to be successful in data science, inevitably
they’re going to encounter machine learning. Would you say
that’s…
Hadelin: Oh yes. Oh yeah, absolutely yeah.
For me a data scientist is also machine learning scientist,
Because anyway, machine learning includes linear
aggression, logistic regression; which are the basic models in
data science.
So yeah. From the moment you know about linear
regression, logistic regression and clustering, well you know
about machine learning. So, yeah. There is a lot of similarity.
And a data scientist for me is a machine learning scientist
and vice versa.
Kirill: Yeah, totally makes sense. And on that project that we’re
working with together, actually for those our listeners who
haven’t yet learned about this project; what we’re doing
together Hadelin is we’re creating a massive course about
machine learning. So everything, everything you can possibly
imagine of machine learning that exist today is going to be
covered. Or if you’re listening to this a few weeks from now,
he’s already covered in that machine learning course. So
that’s a very exciting project that we’re working on.
And just going through it because a lot of the input actually
came from Hadelin. I’m just there putting in like my spin on
things, but I was surprised to see at how many different
areas of machine learning exist.
If you don’t mind Hadelin, could you please walk us through
maybe just give us a list of the ones that on the top of your
head or are the most important areas of machine learning.
Hadelin: So, we started by the most simple algorithm like most simple
subfields of machine learning which are: regression, so
anything to predict a real value; then, classification to predict
the category, like doing some customer segmentations; and,
clustering. So that’s the three basic models of machine
learning: regression, classification and clustering.
And the, once we master this, we start with more
sophisticated machine learning models. So, then there is
association rule learning and actually this is going to be a
very popular subject very soon because a lot of companies
are already working on this subject to find some
associations, some logic in their business to create value. So,
that’s a very important subject.
Then we have reinforcement learning which is what I did in
Google; so to solve interactive problems. So, that will be a
very interesting subject. But actually it’s not very recent, it
exist for quite a long time. Actually there was some books
covering reinforcement learning already in the beginning of
the 2000.
Then we finished with some very, very, very advanced subject
that will make sure to explain very clearly so that everybody
can understand. It’s natural language processing that is
applying machine learning to text. For example; analyzing a
sentiments and feelings and their corpus of text; and deep
learning. Deep learning is one of the most powerful field of
machine learning because you can do amazing stuff.
With deep learning you can do a facial recognition, voice
recognition. You can spot some disease in a picture. You can
recognize anything in a picture. We will do this and this will
be a lot of fun.
And then we will even study some advanced subject of
machine learning like dimensional reduction; that will allow
people to kept the important variants in their data and
actually be able to visualize their data if there’s too many
dimensions.
So basically, we will study all the field of machine learning.
We will make it simple for everybody. Everybody will be able
to understand and use it for both personal life and their
businesses. It’s going to be a lot of fun.
Kirill: Yeah. It’s definitely a massive undertaking.
Hadelin: It’s part of challenge.
Kirill: Yeah. I was a bit concerned if we’ll be able to lift this off the
ground. But, actually this brings me to a very interesting
topic that I want to discuss and I hope you don’t mind
Hadelin.
I was surprised to learn just recently that the way Hadelin
structures his day - his day to day activities, what sleeping
patterns he involves because I think that’s a very important
part of our lives today. We live in such hectic world and
especially if you’re working in data science, you have to find
the time to both learn and also perform the analysis and also
present your findings and also learn again and there’s lot of
stuff going on.
So, I found that I think part of Hadelin’s success is the way
he approaches his daily routine. Hadelin, if you don’t mind,
can you share a little bit about that with us?
Hadelin: Sure, absolutely. Okay. This is quite particular and I hope I
won’t sound too weird.
But, I think a lot outside of the box. That means that I don’t
do things like everybody else. I’ll give an example.
I knew I was going to be an entrepreneur. My first investment
was in a bed, in a bed that would allow me to sleep less
because the bed is highly sophisticated. It’s an amazing bed.
I go on to my bed and I sleep immediately. So that’s amazing.
I fall asleep immediately. And I actually need three hours of
sleep everyday. And why did I choose to do that? Because, I
have a lot of projects in my life. I want to do a lot of things
and I want to be able to complete all these projects. And so I
thought about trying to sleep less to work more and be able
to complete all my projects.
So, yeah. That’s true. I only sleep three hours a day and I
take some naps from time to time to hold on. It’s actually
going very well for me because I’ve been doing this for quite
awhile and I feel I’m in a very good shape. I do some sports
regularly to have a balanced life.
Yes, I never worked so efficiently before in my life.
I complete all my project efficiently and I’m on a good way
about that.
Kirill: Yeah. So there you go folks. Hadelin, it was of course an
extreme example of sacrificing sleep for education and for
working on a person’s projects. But it’s also a great example
of how important it is to remember about these things. And if
you want to progress because as a data scientist, a lot of the
time, others are catching up to others, they’re stepping on
your toes and there’s lots that you need to grasp and taking
into account. Every minute of your life count.
Hadelin: Yes
Kirill: Again, you don’t have to go into the extreme like that’s
sleeping three hours a day. Like, I personally can’t imagine
doing that. But, for me it inspired me to kind of think about
the times in my day where I waste time. Always kind of like
now, taking into account that I can be doing more.
So, that’s a great inspiration. Thank you for that, even just a
personal thank you to you for that.
Hadelin: Well, you’re welcome.
Kirill: Alright. And now, moving on. We’ve talked about your
background, we’ve talked about some things you’ve done at
Google which is very interesting and we’ve talked more in
depth about machine learning and the different fields that
exist there.
What are some of the tools that you use on a daily basis? You
used to use at Google or maybe even now in some of the
analytics that you keep doing. What are the tools that you
use on a daily basis?
Hadelin: Okay. So, I use many tools. But the two tools that I used the
most are R and Python because they have amazing libraries
and amazing packages to do great stuff efficiently and in a
simple way and in a fast way. So that’s the two tools I used.
Besides these, I use Julia, I use Docker, I use that sort of
things. But, really I would highly recommend R or Python to
do some machine learning because for me it’s the best.
Kirill: That’s really great. And we always have this interesting topic
of discussion; R versus Python. What are your thoughts? And
I know that you know both but if you were to choose one,
which one would you choose and also why did you choose to
learn both?
Hadelin: So, that’s a very tough question. I think that there is no
answer to this question. I think it depends on what you want
to do. Because, Python is going be better for some subfields
of machine learning and better than R and R is going to be
better for some other subfields.
Perhaps, then I would count the number of subfields and the
one where I prefer R or Python. But it really depends on what
I have to do. It really depends on my mission, the business
problem.
I know that for example, for deep learning, I will use more
Python than R. But then, for visualization, for simple
machinery, I would rather use R. But maybe, I have a slight
preference for Python.
Kirill: Why would you say you have slight preference for python?
Hadelin: Perhaps it’s because that’s what I’ve been mainly using in
Google. In Google, I work most of the time on Python. Maybe
I have a better level at Python which make it more pleasant
to work on Python because I know more shortcuts. I know
more the libraries. Maybe it’s for this.
But, I really like them both. I use them both equally. But if I
have to say an answer, I would say Python.
Kirill: Okay, Awesome. You mentioned as well Julia, that you
worked for Julia. Actually this is funny because literally 15
hours ago, I’m just checking my messages right now. Literally
15 hours ago, I got a message from one of my students,
Ravender Ram. Ravender, if you’re listening to this, great big
shout out to you.
And he asked if learning Julia is worth it. He’s completed my
R programming course and my R advanced course. And he
was asking about Julia. So, can you tell us a bit more about
Julia and if it’s worth learning Julia.
Hadelin: Well, I don’t think it’s worth learning Julia because Julia is
all about libraries. I use Julia because it has very good
libraries.
I wouldn’t say that I’m a master at Julia because using
libraries is quite simple. There are some tutorials on the net
or there are some frameworks on the net. You just have to
know what inputs to input and then you have the output by
using the libraries.
It’s true that Julia has good machine learning libraries and
that’s why I use it. But, I wouldn’t start a new course all
about Julia to learn about the syntax, the complicated stuff
about Julia. Maybe we can introduce that a little bit in our
course, mention the good libraries to use with Julia. Once
you know what to input, it’s quite simple.
Kirill: Okay, yeah. Definitely that’s something that we could
consider for our project as well. What about some of the
other tools like apart from R, Python, is there anything else?
For instance, you mentioned you worked with Hadoop at
Google.
Hadelin: Yes.
Kirill: Did you worked with PostgreSQL or did you work with
certain types of Hadoop or any other database related tools
that you can mention?
Hadelin: I said Hadoop but actually Google is so protective of its data
that Google has its own database system. It was actually
called F1 and Dremel but it’s kind of like Hadoop. It’s based
on My Production which is actually introduced by Google as
well.
It’s like SQL. It’s like working on big data system and
managing the data inside of it. I was also building some
workflow. For example, in your data science course, Data
Science AZ, I was doing a lot of stuff that you explained in
Part 3 Preparation by making some workflow, by making
some templates, that sort of thing but it wasn’t on Google’s
own database system.
Kirill: That sounds good. Back at Canal Plus, where there any
specific tools that you use there?
Hadelin: Well, actually on Canal Plus I was mostly working on R and I
was building the recommender system on R. Actually just
on R and that’s where I improved a lot in my level in R.
Actually, I would like to say something. The best way to
improve skills in programming is by actually doing an
application by building something. And by building this
recommender system in R, I considerably improved my level
in R.
Kirill: It’s funny that you say that because a lot of students come to
me and they ask for more exercises, more challenges.
Because like you say, it’s the best way to learn, to actually
solve challenges based on real world data and real world
problems or even business problems.
Hadelin: Exactly, yeah.
Kirill: That’s something that we’re really looking into and in our
platform of Super Data Science. That’s going to be a big
focus. We’re going to have a heavy focus on actually not just
presenting the tutorials and lectures but also presenting
exercises on a monthly basis that people can come in and
refresh their skills. For instance, if somebody wants to learn
R or Python, they just do an exercise on that or on something
else. I think that’s an important part of learning.
Hadelin: That’s very important, yes.
Kirill: Definitely. Okay. And so, the next question is quite an
interesting one. If being a data scientist and learning a lot –
and by the way, people probably notice that today, I’m saying
data and data interchanging. That’s funny because I learned
recently that we only say “data” here in Australia, data
science. I’m trying to see if data science is going to be a good
option as well. How do you say it?
Hadelin: Data.
Kirill: Data Science, yeah. Coming from Google, probably you
would.
The question was, what was the biggest challenge you’ve ever
had as a data scientist?
Hadelin: I can think of two. The first challenge was actually to build
this recommender system. Because Canal Plus is a French
company that exists for a long time and Netflix just entered
the French Market in everything – in TV and shows and
everything and their Netflix program.
Netflix has a huge recommender system and that’s actually
their strength. It’s their weapon.
And the challenge was to build a recommender system for
Canal Plus that will be Netflix level. That was quite a
challenge because the Netflix recommender system was
based on a competition across several countries. And the
competition that last actually for several months. So, I have
to build it by myself. I was actually with somebody else and a
manager but we’re just a team of three. That was quite a
challenge to make a recommender system that would be the
same level as the Netflix one that was based on competition
with the best data scientists in the world. That was very
challenging.
So, I built a recommender system. I don’t know if it’s Netflix
level but I know for sure that I improved Canal Plus
recommender system so that’s a good thing. I wouldn’t say
it’s Netflix level but it was quite a challenge to make it
anyway.
The second challenge, well I didn’t mention it but I love
writing. I actually write a lot of stuff. Right now I’m writing a
book. This book is about data science in the future. It’s a
novel so it’s purely imaginative. So, I’m imagining a story in
the future where machines are really developing. I’m
imagining the world in 20 years how it’s going to be. And
that’s quite a challenge to imagine that and try to be logic
and try to predict what’s going to happen in terms of data
science and machinery. That’s the second challenge I would
mention about that.
Kirill: That’s so cool. That’s awesome. So, everybody who’s
listening, definitely check out the bookstores for Hadelin de
Ponteves’ book. It sounds like an exciting – I’ll definitely buy
that book coming from somebody who’s in the industry.
Hadelin: Thanks.
Kirill: Just on that second challenge, are you using Moore’s Law to
predict what’s going to happen in 20 years?
Hadelin: Well, I’m using mainly my imagination. Well, for example I
will tell you about my idea. I will tell you about what I believe
in.
I believe in the not too far future, there are only going to be
two jobs actually. There’s going to be the engineer and the
artist because machines are going to develop so much that
they are going to replace a lot of jobs. And, I think that in
the not too far future, the only jobs that are only going to
remain will be engineer and artist.
Because I don’t think a machine can replace an artist and we
will need engineer to continue building the machines and
improve them and mostly control them because we don’t
want have machine declaring a war to us, right?
So, that’s the kind of things I’m trying to picture in my book
and build in my book. So, that’s quite a challenge because
it’s not easy to predict in the future. But having that quite a
while, I think I’m on the right path.
Kirill: That’s really cool. I recently saw an infographic. I think it
was from Futurist.com which describes the predictions of
Ray Kurzweil for the next hundred years. Ray Kurzweil,
you’ve heard of him, yeah?
Hadelin: Yes, a little bit.
Kirill: He’s an American futurist who’s made predictions about the
future since 1980 and he predicted things like the iPhone,
the iPad, the modern machine, computing power of machines
and the way we live and self-driving cars and so on.
He just doesn’t predict it. He predicts with the exact date,
the exact year and so far, his predictions have been 80%
correct. Definitely check out that infographic and we’ll
include it in the show notes as well.
He predicts crazy things like by the year 2019, machines will
be voting for humans to recognize that they’re conscious
beings and give them rights to vote and things like that. It’s
really crazy as well.
But going back to your first challenge where you created the
recommender system, that by the way, that’s a huge
accomplishment. The recommender systems are like – the
Amazon has a very powerful recommender system.
Hadelin: Yeah, Spotify, Amazon, Netflix even Udemy actually.
Kirill: Yeah. Exactly, right. That’s the stuff that the ads or the next
recommended items that you should purchase that the
system comes up based on your previous experiences.
Amazon even took it to the next level. They introduced this
one-day shipping or same day shipping based on the
recommender system. Based on your previous purchases,
before they ask you to recommend you to buy something,
they’ll take that item and they’ll ship it to a warehouse which
is next to you. And then when you buy it, you’ll get in the
same day because they knew you will buy it before you even
knew you will buy it.
Hadelin: That’s great.
Kirill: That’s machine learning, isn’t that?
Hadelin: Yeah, that’s machine learning. That’s the power of machine
learning.
Kirill: So true.
Hadelin: That’s why it’s so exciting.
Kirill: Exactly. And so, two questions here. First one was about this
big challenge at Canal Plus. What were the main kind or how
long did it take to build a recommender system and what
were the main challenges there?
And the second question will be, how is it working in a team?
I know you said it’s a small team – a manager and you and
another data scientist. How is it working in a team of data
scientists building a project? Because a lot of the times data
scientists actually work on their own thing. Especially if
you’re freelancing, you’re just working on some project by
yourself.
Can you tell us a bit more, as a second part of this question,
what is it like to work in a team of data scientists?
Hadelin: Okay. So, the first part of the question was how long did it
take? It took six months because they were actually part of
my school program so I was doing that actually twice a
week in parallel of my engineering program. By working
two days a week in six months, I have to build this
recommender system .
And so that leads me to the second question. You have to
be very organized. You have to use most of the resources of
the team to be able to build this recommender system at
such a high level in six months. So, you have to
understand what’s the best skill in each person in the team.
You have to allocate the best resource from each person in
the team to organize in such a way that you can build a
recommender system faster.
Actually, I was better on the research part because I love
about machine learning theories so I was working a lot on
the research papers and trying to find the best models and
how to combine them.
The other guy was actually very good at coding. He was
more the developer. I was the researcher, he was the
developer. By doing this, we managed to build this
recommender system in due time.
Kirill: Fantastic. And what is your manager doing then?
Hadelin: He was supervising us. He was checking if we were going in
the right direction but he wasn’t part of the building
process. We were all by ourselves.
He also helped us at the beginning. He explained what was
Canal Plus recommender system at that time, what needed
to be improved. He really explained the context and the
goals, not much more than that. We did all the work by
ourselves.
Kirill: That’s lovely. I love it. I love how you described the two
parts.
And in this scenario, you split the work in a way that it
doesn’t interfere with each other. I think it’s always a better
approach when people are working, even within a team,
they’re working on individual parts of the project rather
than coding in the same code. That can lead to lots of
confusion and has to be managed differently.
Hadelin: Exactly, yeah.
Kirill: And you mentioned you love research. This is a great segue
way to my question. What is your most favorite thing
about data science?
Hadelin: My favorite thing about data science is creation. That’s the
first thing I said actually. I said that I chose data science
because you can create a lot of stuff. I also talked about all
the projects that I have in mind that I want to complete in
my life. A lot of these projects are related to creation. I want
to create a lot of stuff that is related to data science. That’s
what I love about it.
And you know, creating these stuff implies doing a lot of
research because for example if you want to build
something revolutionary, you have to spend quite some time
in research to know how to implement it, to see if it’s
possible. You have to read some research articles, to
understand which algorithm to use, how you can develop
them. So, that’s quite exciting. But for me, when I will be
doing some research, it will always be linked to creation, to
creating something. I will not do some research to invent a
new theory. It will be to create something.
Kirill: Okay, yeah. That’s very cool and that’s very in line with
what you said for your book that there will be artists and
engineers. You use software as more for an engineer.
Hadelin: Absolutely, yeah.
Kirill: This whole concept of robots in the future where this is all
going, aren’t you afraid? This is kind of like veering off to
the side here a little bit. Aren’t you afraid that by creating
so many things especially using data science, you’re going
to teach the machines to be very pragmatic, rational and
think on their own. And then at some point, they’ll decide
humans don’t have a place on this planet, that we’re just
making this planet worse. How do you see that going on?
Hadelin: I actually talked about that in my book. That’s a very true
subject. Actually, some movies in cinema have talked about
this subject. I can think of iRobot for example, Ex Machina.
There are a lot of movies discussing about this subject.
So, that’s why I think that engineering is going to be a very
important job in the future because it’s not only about
improving the machines and developing new machines. It’s
also about controlling the machines so that we can avoid
this to happen. We will need to improve the security
systems, to improve all the controls that there are in the
machines to avoid this.
I think that will be new subfield of machine learning, if it
doesn’t exist already. Actually, I think it already exists like
machine learning security or something like that. There will
be a lot of that.
Yes, I’m scared but I’m sure we’ll do the necessary to
prevent this on time and to predict what’s going to happen
and try to prevent this.
Kirill: Okay, so that is really cool. I love that you brought this up
because a lot of our listeners are always curious about
where everything is going and what career paths might
exist.
It is a fact that in the next top 10 jobs that are going to exist
in 10 years from now don’t even exist at the moment. It’s
all evolving so quickly. And what you brought up right now
is a very, very profound pathway that might develop which
is controlling machines, being a data scientist but in the
field of controlling machines so that we are always in charge
of them. That’s a great one. I’ve never heard that before.
Hadelin: Actually, you’re right. The world is going so fast. There is
going to be a lot of new jobs but I’m sure 100% that one of
these new jobs will be to control the machines and
guarantee the security of people against the machines.
Kirill: That’s really cool. That’s very interesting. What kind of
skill sets do you think will be required for that? Of course,
data science but what would you mix data science with to
get somebody who would be an expert in that field?
Hadelin: Well, I actually read an article not a long time ago about the
new skills that we would need to have in 2020. Actually,
the first one was creativity. I’m very happy to have a need
for creativity because it’s apparently a required skill.
There is creativity. There is also adaptation. You need to
adapt very fast because the world is going so fast that you
need to be able to learn fast, understand fast what’s going
to happen and actually predict.
It’s not only the machines that have to predict things, it’s
also you. You have to predict what’s going to happen. You
have to anticipate.
I think anticipation, adaptation, creativity are very
important besides having the skills of data science and
machine learning.
Kirill: Wonderful. I actually love to read that article if you can
share it, if possible.
Hadelin: Yeah, okay.
Kirill: We will include that in the show notes for our listeners as
well.
Hadelin: Okay, sure.
Kirill: It sounds like a great read. By the way, you mentioned a
movie, Ex Machina. If you’re listening to this podcast and
you haven’t seen Ex Machina, you have to see it. It is so
good.
Hadelin: It is very good.
Kirill: It is good. It’s got three or four actors in there total and at
the same time it blows your mind. It’s totally about
machine. It’s all about machine learning, about data
science, about robots, about where the future is going. I
could watch that movie every day.
Hadelin: And it’s about something that could happen and that’s
scary but that’s exactly what we’ve been talking about.
Kirill: It might be happening already.
Hadelin: Yes, maybe somewhere in the mountain.
Kirill: How they conveniently replaced Google with Blue Book.
The creator of Blue Book created this machine. It’s just
fantastic.
Hadelin: Yes, you have to watch it.
Kirill: You have to watch it, totally. We’re slowly coming up to the
end of our conversation.
So, just a couple of finish up questions in terms to give our
listeners some ideas of how to get inspired, how to go
forward in this career, what would you say was the biggest
inspiration or aspiration currently for you in data science
that has pushed you forward or something that is pushing
you forward every day to become a better data scientist?
Hadelin: Well, I have several aspirations, actually. I actually have
several mentors. I inspire from several mentors’ ideas.
I will start with Nelson Mandela who actually said that
education is the most powerful weapon which you can use
to change the world. I absolutely love this. That’s why I also
want to be an educator in everything that I’m doing. I’m
inspired from that a lot.
Then, I’m inspired from Larry Pages, CEO of Google not
because I worked in Google. That absolutely has nothing to
do about it. It’s because when he started Google, he started
his print and that’s exactly what I’m doing right now. I’m in
a print that I started not a long time ago. Now, I will do this
print until I realize all my projects that I have in mind. I’m
on this print and I will continue.
And I also have inspiration from Mark Zuckerberg who
started with Revenge Energy. I think Revenge Energy is one
of the most powerful energy to complete something to be
able to achieve something very, very difficult to achieve.
Revenge Energy can take it from all the people that put you
down. That’s called Revenge Energy. From these people,
you can use this energy to transform it into something very
powerful and very creative. That’s the third thing, Revenge
Energy from Mark Zuckerberg.
The fourth mentor I would mention about is Elon Musk who
is very imaginative, very creative. And I love his masterplan.
I actually will inspire from this because I think it’s great. I
love his ambition. And I want to have and ambition like
that too. I’m ambitious.
So, that’s all the inspirations that I use.
Kirill: Well, fantastic. I love it. We really meet on that, Elon Musk
and aspiration and inspiration. I also follow Elon Musk. I’ve
read his biography. By the way, listeners of this podcast if
you haven’t read it yet it’s by Ashlee Vance. It was released,
I think start of 2015. Great biography.
Elon Musk, definitely one of the pioneers of everything that
we’re doing right now that’s innovative – rockets, self-
driving cars and solar.
And I also loved Mark Zuckerberg. Revenge Energy - I
haven’t heard about that but it sounds like a great thing to
turn negative energy that’s coming towards you from others
which still happens even in this world. Unfortunately, it
happens.
Hadelin: It happens every day to everyone.
Kirill: It happens all the time.
Hadelin: All the time. And in every context. You can use this energy
to transform it into some very powerful energy that can
guide you and that can make you achieve some great
things.
Kirill: Yeah. The concept reminds of Judo or Aikido or some sports
used in martial arts where somebody is running at you and
instead of blocking them and fighting them back, you use
their weight and their energy to throw them down or in this
case, to create something beautiful.
Hadelin: I did a little bit of martial arts and I know exactly what
you’re talking about.
Kirill: That’s some inspirations. That was great for inspirational
characters in your life.
If our listeners want to learn about you, contact you in any
way, get in touch, follow you maybe, what would be the best
way to get in touch with you?
Hadelin: I think LinkedIn is a good way and Google Plus as well. I
don’t have a Twitter account yet. I should actually work on
that. But, I think so far it’s Google Plus, Gmail and
LinkedIn.
My LinkedIn is Hadelin de Ponteves. It’s my first name and
my last name. We’ll probably write it somewhere because
it’s not very simple for a non-French person. It’s Hadelin de
Ponteves on LinkedIn and Google as well.
Kirill: We’ll definitely include those in the show notes and as soon
as Hadelin includes that Twitter account or something else.
We’ll also include that and update it.
And final question, what is the one book that you would
recommend to our listeners that can help them become
better data scientists?
Hadelin: Okay. The one book, so my favorite book in data science is
called Data Science for Business but it’s actually not to
start data science. It’s a very good book once you have the
basics of data science. You must read it in anyway because
it’s really, really amazing. It talks about how you can use
data science to create added value in your business. That’s
great.
But If I had to mention a book to start from scratch, it’s
actually Data Science from Scratch with Python. Because
when you really start from scratch, you use Python which is
the best tool for data science. It’s a very complete book. It
tells you about the basics. So, I would say that yes, Data
Science from Scratch with Python.
Kirill: Wonderful. Thank you very much. Great recommendation
and I really appreciate you coming on the show. I’m sure
everybody’s going to get a lot of value out of this interview.
Thank you very much.
Hadelin: I hope so. Thank you.
Kirill: Take care. Bye.
So there you go guys. That was Hadelin de Ponteves. I hope
you enjoyed this inspiring and entertaining interview. I
definitely had a few good laughs while we were chatting and
at the same time, I did learned a lot. Hadelin is a very
impactful person and so I learned how to make myself work
harder, that I need to push myself further, and of course
I’ve learned a lot about machine learning as well. So,
speaking of which if you haven’t yet checked out our
course, it’s upcoming literally in a few days. It’s going to be
released and once it’s released or if you’re watching and
listening to this later further down the track, then it’s live,
definitely check out the Machine Learning A-Z course,
where Hadelin and I put our experience and our expertise
together to bring you the knowledge of machine learning.
And make sure to share this episode with your friends,
family, colleagues, whoever you think is interested in
machine learning and the future of data science and the
future of the world for that matter.
Also, you can get the show notes at
www.superdatascience.com/2 . So that’s just the number 2
for episode 2 and please leave us comments for Hadelin and
I at the bottom in the comment section under the episode.
We’d love to hear what you thought, what your ideas about
the future are and how you use machine learning in your
day to day role.
And I look forward to seeing you next time. Until then,
happy analyzing.