Latest / Agentic Conversations (formally mlops.community) / Machine Learning in Cyber Security // Monika Venckauskaite // MLOps Meetup #70

Transcript

Machine Learning in Cyber Security // Monika Venckauskaite // MLOps Meetup #70 · Agentic Conversations (formally mlops.community)

0:02
Monica is a keen and curious, engineer loving to build systems
0:20
own computer. She started in machine Learning
0:31
as a master student looking for Higgs boson and dark matter
0:40
within the CERN data later on. Monica move to the IT industry.
0:52
She worked on various machine learning projects including Open
1:00
source intelligence to and a distributed system for ML.
1:09
Cyber security analytics. Oh well, currently Monica works
1:18
as an MLA lobs engineer. Improving the ml Ops platform
1:27
that is used in production. Action for shipping models to a
1:34
45 million. Use a platform Monica.
1:40
Wow, your bio is incredible. It's just stunning and I hope we
1:48
can all learn a lot today. Thank you for coming on.
1:54
The MLS Community Mead of Make sure did waitress it sounded so
2:03
like very very long life story. Well that was because you had a
2:12
an awesome, awesome bio. Now for those that do not know,
2:20
I'm going to give a few announcements and we're going to
2:22
jump into this full conversation with Monica.
2:25
Thank you all for joining us. It is an absolute pleasure to
2:29
have. Have Monica here and everyone
2:31
else here. The first announcement that I
2:33
will make is we want this to be interactive.
2:38
So feel free to ask questions throughout the presentation.
2:44
Monica's got a quick few slides that we're going to go over and
2:49
then we're going to be just chatting and so at any point
2:54
feel free to jump in. Ask a question you can put it in
2:58
the Q&A tab. That we have or just throw it in
3:01
the chat fuel like you are part of this because that is the
3:06
beauty of being here live in real time in this Meetup that
3:10
you get to interact with us and we get to interact with you.
3:13
Next thing I will mention is that we have an MLs Community
3:18
slack. I see a lot of familiar faces
3:20
and names in here, but if you're not in the MLS Community slack,
3:24
I highly encourage you to get in it.
3:26
We've got over 5,000 people now, that is Is unreal to me because
3:31
just a year ago we were at less than 100 and as you all know ml
3:36
Ops is blowing up. So lots of good questions, lots
3:39
of smart people that are very, very kind with their time and
3:43
wisdom. And if you would like, I'll put
3:47
a link to that community in the chat glass thing over the last
3:53
six years, I have been trying to create a Sorry I said six years,
4:00
I meant six months it's definitely not six years over
4:04
the last six months, big difference, but it still has
4:09
been a big Endeavor for myself. A six months undertaking trying
4:13
to create a landscape that will help people understand what each
4:18
tool does where it fits in the ml Ops life cycle.
4:23
And right now on our ml Ops Community website, we have
4:28
released the First two of the ml Ops life cycle sections, and
4:34
we're calling it The compare page so you can go on there.
4:37
You can compare different tools within the monitoring space and
4:42
the feature store space and our goal is that we are going to
4:48
continue doing this for every piece of the stack.
4:52
So we demystify the different tools that are out there because
4:57
I don't know about you but a lot Sometimes I go to a website and
5:02
it's an ml Ops platform or they just call themselves an AI
5:05
platform. And then I poked around for
5:08
quite a bit and at the end I go. So what do they do?
5:13
And that is what I want to make sure no one has to put up with I
5:19
want to make the ml Ops community site, be a place where
5:23
you can find information and see what the differences are
5:29
between. Between different tools and what
5:31
opinions, they're taking, why they differentiate their selves
5:34
in one way or the other. So you can check that out on the
5:38
MLS Community website and that is as easy as just going to ml
5:43
Ops doc community and you'll find it there without further
5:47
Ado. I think we can start Monica
5:50
maybe before we jump into the slides you can tell us a little
5:53
bit of background about yourself and how you came to be where you
5:57
are right now. Dragon M.
6:00
So first of all, thank you for such a nice and tall and very
6:05
glad to be here. I actually discovered Lots
6:09
meetups when I was looking for mation about the tools.
6:12
So I'm very glad that I ended up actually here and.
6:18
Well, I started out on the originally and machine learning
6:23
while I was still studying as University, I did projects for
6:28
sale Iran and family Library. I looked for his Exposé on and
6:31
dark magic and it turned out that well, it won.
6:36
The events are very rare such as the production of Higgs boson.
6:41
You always need to such Advanced analysis, methods as machine
6:45
learning in order to find them happening.
6:49
For those who are not too interested in physics, I was
6:53
just mentioned as expose on is particle responsible for all.
6:58
all the map to having Mass so later on actually moved into
7:05
industry and I use my experience in machine learning and also
7:09
combine the experience of service Administration and made
7:17
made myself into a professional analogs, which I'm very glad
7:23
because it combines The Best of Both Worlds and it Actually lets
7:30
me use all the skills I have, is what I like about the mobs that
7:35
it's still very Dynamic. It's evolving.
7:39
So there is a lot of space for creativity as well as Discovery.
7:45
So I think, yeah, that thing. These are the reasons I like the
7:48
field in general and I'm glad to be here today talking to you
7:53
guys. Please shoot questions.
7:56
So, whenever you want to, I'm looking That's house as well so
7:59
you can can come at any time. Yeah, so let's get into it and
8:06
what we're going to do today because I think I find it ironic
8:11
Monica and I were talking beforehand and trying to make
8:14
everything work properly. We did a sound check.
8:17
And then we tried to see if Monica's screenshare would work,
8:20
but because she's on her work computer.
8:23
The screen share is disabled for Zoom due to security purposes.
8:27
I Zoom. And so it is my pleasure to
8:32
present what Monica's slides are.
8:35
And what we can do is just basically anytime Monica, you
8:39
would like me to change slides, you can do a hand wave, and then
8:44
I'll change so that we're both on the same page and let's just
8:51
double check. Make sure that you can see my
8:53
screen. Everything looks good.
8:57
Yep, perfect. We are today we'll be talking
9:00
more about machine learning site.
9:02
Actually, we can touch upon animal upside as well, and if
9:07
you can have questions, we can discuss it.
9:09
But my goal today is to present the cybersecurity world news
9:17
like why we are using machine learning why we cannot
9:21
absolutely do without it. And yeah, what are the benefits
9:26
and and whatever. As well, pitfalls of machine
9:31
learning in our world today. So yeah.
9:34
Glad to be talking to you and let's come to another site.
9:39
So, first of all, we'll be reviewing a little bit about to
9:46
landscape of cybersecurity today.
9:48
What are common? First have a eyes being hacked,
9:52
as well as will be talking. Fake news information workers,
9:56
how we are dealing with them. There are two special games for
10:00
them return to Coruscant open source.
10:03
Intelligence also will be talking about machine learning
10:07
agent as a threat Hunter. And later on, you can dial into
10:13
a future cyber security. What kind of systems are being
10:16
developed for defense and forth, start counting So just start and
10:23
a higher note to you, probably all learned about neural link as
10:28
well. Most of you have seen the movie.
10:30
How Ghost in the Shell, it's about a future where humans and
10:34
machines merge, and I think it doesn't seem so far-fetched
10:40
possibly very soon. We're gonna have a human
10:43
computer interfaces and we can see that cybersecurity is gonna
10:48
touch us personally, It's way more important than our everyday
10:54
lives and it's going to be even more and more important in the
10:58
future. Because currently, our world is
11:02
also run by machines that most of medical records are also on
11:10
computers so it does Justice very personally.
11:17
And it's talking about your cybersecurity first.
11:20
Well then there is a classification for them and most
11:26
of our companies around the world report threats and there
11:34
is I don't think so that there is a company in the world which
11:38
existed for a couple of years, which have sinned experienced,
11:43
at least one of those. So, most of the company's
11:47
experience, external hacking about half of them.
11:51
There's also malware incidents, social, engineering, spam, as
11:56
for social engineering, I think all of you experience phishing
12:00
attacks when you get a malicious Link in the email and it's kind
12:06
of hard to defend against those. So first of all, we have to be
12:10
smart ourselves and and be aware of it also.
12:15
So, they have been incidents of inside of data, leakage, and
12:20
theft Facebook is very often. So you've probably been have
12:30
been experiencing a few of those and it's coming the Precedence
12:38
of this taxes increase in the future.
12:44
Can we move on and talking about cyber security incident
12:48
response? It also has a cycle very similar
12:52
as an ellipse has a cycle. So, cyber security incident,
12:55
response also has a cycle. First of all, the threat is
13:01
identified. Usually it is done through
13:04
looking into the locks in your computer or any other device
13:10
that's being under attack. Every system produces system
13:16
message logs and also we can use additional tools to log them if
13:23
you are very curious about cyber security.
13:27
Log monitoring tools, you can look up for open source tools,
13:32
such as surakarta or Zeke, they are and there are also more
13:37
tools and the second response part is Protection.
13:45
So some protection can be done in advance such as taking care
13:49
of your access controls. Not giving too much of
13:54
permissions, also being aware. So that's why companies has to
14:00
train the individuals, and also companies have to protect
14:05
information. And number three is detection,
14:10
detection can be done. Also, using cyber Your tier logs
14:17
and also the systems you have on top of it, you can have a
14:24
machine learning agents looking for archaea locks and helping to
14:28
determine what kind of attack it is.
14:31
Or you can have an analyst sitting and watching the
14:35
dashboards and trying to determine that.
14:38
While something is wrong with the system.
14:40
Obviously machine learning engines are faster but there are
14:44
also limits to them. Them.
14:45
Because in order to train a machine learning agent, you need
14:50
to provide it with the data. You need to provide us with
14:54
expert intelligence. How that does certain attacks
14:58
look in the data. So machine learning models so
15:03
far, are as intelligent as cyber Security Experts.
15:10
But but they are faster. So these are the upsides of it.
15:15
And number 44 cycle is respond. So it's important to respond as
15:22
quickly as possible in for very serious attacks.
15:27
So, maybe it's like CEOs, account being hacked or your
15:33
services. It is experiencing denial of
15:37
service then you need to respond as quickly as you can.
15:44
And the number 5 of course is recover recover, your service?
15:49
You can continue Communications, restore your service and go on
15:54
with 16 glue, the vulnerabilities in your system
15:59
so for every incident response is cyclic, you need Go through
16:06
all the steps there is actually pretty standardized process of
16:13
the response so all the cyber security teams are being planned
16:17
on it and of course, you need always to improve your system to
16:22
eliminate the vulnerabilities as much as you can.
16:26
So I've got a quick question here before we move forward on
16:30
this detect part, where you were talking about, how machine
16:32
learning models are involved. But they're only as good as the
16:37
humans that are doing it. It's just faster.
16:43
Does that mean that any time there's some new attack that
16:48
comes out? These machine learning models
16:51
are not going to be able to pick it up or is it like can you
16:55
explain how that works? And yeah.
16:58
Yeah there are I would say a couple of approaches some
17:02
systems combine both is this An advanced machine learning
17:07
detection, system can have both. So one of them is pretty simple.
17:11
You simply look for anomalies. So here, you can also detect a
17:19
new kind of attacks with you. Thanks an anomaly.
17:22
It could be a different network. Patterns are also some different
17:29
to logging in patterns the machine Learning System could
17:33
simply stands. An anomaly.
17:36
And also, there's another type of machine learning system which
17:44
detects already behavior that behavior is was trained on.
17:48
For example, it could detect like that.
17:51
Somebody locked in into the system and try to elevate their
17:56
privileges to become the root user.
18:00
So, in this case, the system should be trained on. on
18:05
scenarios an advanced machine Learning System could have both
18:11
it because those have both anomalies and scenarios, but of
18:17
course, I think nothing can change cyber security expert
18:23
with the experience because well, as files for machine
18:27
Learning Systems, of course, we train them but I think they're
18:33
can always possibly We be blind spots and and experience cyber
18:40
security expert could notice that.
18:43
So, I think the best way to see it is that machine learning and
18:50
cybersecurity should work in Tandem and that machine learning
18:56
could automate some usually regular the task, like a regular
19:03
text. And Make the word from expert
19:07
faster, but in general, I think so far we cannot cannot replace
19:13
experts completely. Yeah, cool.
19:17
Thank you. Okay.
19:21
So another reason to use machine learning is actually at the
19:24
experiencing, a huge expert shortage.
19:28
So if anybody watching is, still considering a career in cyber
19:33
security, you it's a good decision because the news is
19:38
only is only gonna grow in the future.
19:41
And for now, we already missing about 4 million experts
19:47
worldwide. So this means That.
19:52
Well, the contacts first has to work harder, to have to find
19:55
ways to make everything more tomatoes and faster.
20:00
And basically, we cannot do away with out machine learning,
20:04
because it helps to helps to at least reduce the stress
20:10
experienced, by the shortest of Specialists and looking as the
20:15
current world. I mean, the cybersecurity
20:18
threats are all around and us. First of all, all of you
20:23
experience the damage. Fake news can do and sadly
20:30
machine learning only made easier to produce fake news and
20:34
and six videos. But on the other side of the
20:38
coin, machine learning is also being used to combat fake news.
20:43
So the algorithms similar algorithms could be used.
20:48
Also to detect further videos. Fake.
20:50
When whether the comments are 22.
20:59
Different, and sentiment meaning is fake news.
21:02
Use usually targeted to to produce a reaction in users, and
21:09
if we hear you have a very contrasting opinions, and
21:15
comments, you could also detect that it could be a fake news.
21:20
Thickness. And another thing that's very
21:26
relevant today is hacking machine Learning Systems.
21:32
So turns out that face recognition algorithms and not
21:38
absolutely accurate. They can be hacked by with
21:42
facemask and various different methods and This might mean that
21:50
well, it's not an absolutely ideal method of personal
21:57
identification and the more accurate, the more safe wanted a
22:02
is still using fingerprint recognition.
22:06
And as far as talking about the defense systems today, they're
22:11
actually there are drones being used for automated defense
22:19
systems and cyber security. Here is also very crucial but
22:25
different system itself is also something.
22:30
Well, it could be considered as part of cyber security because
22:34
it's very Automated money, machine learning, algorithms are
22:40
also being used to pick targets. So basically the same
22:46
recommendation engines that are being developed to sell us
22:51
relevant products. They are also being used to pick
22:54
relevant targets and also optimize which, which type of
22:59
Rocket should be used in different systems.
23:04
And another one, I would like to draw your attention to is cyber
23:08
security and and hospitals, actually, sadly.
23:13
Last year we had a first victim of cyber security incident.
23:19
The first death in the hospital because the hackers hacked into
23:25
the hospital and encrypted the medical records of the patients.
23:30
And doctors were not able to help a patient.
23:34
So we are seeing that well cybersecurity importance is
23:39
growing and these days is also a question of life and death as
23:45
well as this he would talk about automated autonomous cars.
23:52
So you possibly remember also from last year that tells us
23:58
where hacks with modifying the speeding limit signs.
24:04
Which resulted in a car accelerating where it should
24:10
have been driving at 30 miles per hour.
24:15
So it's accelerated much faster because they perceived it as 80
24:20
miles per hour so it could have ended in an accident while this
24:29
time, the driver was lucky, but We realized, well it's cyber
24:35
security is going to be more important in future and machine
24:38
learning also can be hacked in very unstated forward ways.
24:45
I would say it's very often unpredictable, and we should be
24:50
careful about using machine learning to it.
24:53
It has its course and in certain scenarios.
24:56
We should have I think systems to protect us because human
25:02
life. Visit steak.
25:07
And as talking about internet, so that the usually use a small
25:16
small, part of the whole web of the word is much much deeper,
25:23
some of you might have heard of dark web.
25:26
So, this is where our open source intelligence tools.
25:32
Come to action, most of the web we use Is publicly available is
25:39
Surface, Prep is public profiles everything on Wikipedia?
25:44
Who wants such all such platforms also.
25:47
Deep fat is more like forums what you can read inside the
25:51
phones. It's not easily indexed and
25:56
searchable. But the most interesting part I
26:00
would like to draw attention to this dark crap, because this is
26:04
where the most crimes happen. So, So it's both people Tackett
26:10
trafficking and drug sales and and all the illegal activity so
26:18
cyber security is also responsible.
26:23
I would say for the dark web security, we have tools also,
26:29
government is, is making actions to be able to track what's
26:35
happening on doctor. to be able to find criminals to, to find a
26:43
legal action and this is also where machine learning and web
26:49
crawlers are coming in and it forms one huge area of cyber
26:55
security which is called open source intelligence.
26:59
We will be discussing this today as well.
27:02
So if you have questions, so feel free to drop them in the
27:05
chat. So you can change slide.
27:10
I see there is a question coming through in the chat.
27:12
Yes, asking about how to get into cyber security as an
27:17
outsider. What kind of skills are needed?
27:21
Yeah. So it depends which side of you
27:24
if you want work because you can work as as an expert in
27:30
monitoring the systems and also the Professional School
27:37
Analysts. And also, you can work as a
27:40
product developer, which I had chance to do that as well.
27:44
So I basically, I came from machine learning background and
27:49
I joined in the project, actually, two projects with used
27:55
my versatile because as a developer, you can also
28:00
contribute as an L analyst if you like to digging through
28:04
system locks and discovering hacker, Turns learning about
28:09
malicious behavior, is that your kind of thing?
28:12
Then you would be aiming to be anomalous and well, in some
28:18
cases so you can, of course, graduate from computer science
28:22
also, but they are also many online courses you can, we can
28:28
use and it's in cyber security. We have quite a few standard
28:34
certificates. So you can, you can after you,
28:42
you learn in the courses, you can pass a certificate and you
28:46
can try to entry role in cyber security.
28:52
Or if you like creating products, you can just, you
28:56
know, develop your coding skills and and try to join a company
29:02
where products are being developed.
29:04
And there is something also interesting that I've heard of
29:07
and maybe we can talk about this because I just saw something on
29:11
LinkedIn of a connection of mine.
29:14
Got a 60 thousand dollar Bounty because they did they found a
29:22
bug or a potential security threat for Microsoft.
29:27
And so once you get to that level, I imagine that Hey
29:30
completely different level where you're the one that knows a lot
29:34
about the vulnerabilities and you're able to go in and try and
29:38
exploit those vulnerabilities. Yeah.
29:40
So I don't know how it's gonna end but I think this some is
29:46
pretty low for for the I mean, for the impact, it was castle
29:52
and Microsoft's users, so I don't know.
29:54
It might end well for the for the person with yeah, in general
30:00
is very common and usually usually there are proxies but
30:10
usual procedures for for the users who discover bugs to
30:14
report those, but I think yeah. And this case they will just
30:19
want some Some process. So it might work.
30:23
It might work because for Microsoft, I think it's a minor
30:27
thing that In this last year yet.
30:30
It's very, it's very common and also, yeah, and also asking
30:39
asking money for for encrypted data and so on.
30:45
So it's cyber security is it's crucial thing and and sometimes
30:52
even like in the case of of the crime company style.
30:59
So willing to pay bounties to get their data back, even before
31:05
the crime, can be solved by the police because they The data so
31:10
much. So yeah, yeah, yeah, along those
31:15
lines. I it's fascinating because I
31:16
think I just saw some news about how some states in the US are
31:21
considering making it illegal to pay the ransom Ware and so that
31:27
would put companies in a very big pickle if it were to happen
31:32
like that. But just moving on.
31:35
I see there's another question in the chat and it's asking do
31:40
you think governments and organizations are serious about
31:44
cybercrime? Yeah, so I think the area is
31:49
rolling very quickly and even though governments protecting
31:57
the very most important assets. It's, I mean, there's always
32:03
room for vulnerability. I know.
32:09
This, that governments are taking actions, they're aware of
32:14
it. But the problem is that hackers
32:19
evolve very quickly, and they also use the most modern tools,
32:24
and it's hard for governments to keep up.
32:27
So that's why we still having such problems as no power plant
32:31
being hacked, which could end in catastrophe reports.
32:35
And you also have like governmental systems being
32:40
hacked for very often for political reasons, as well as,
32:46
as part of information were between countries.
32:50
So, yes, governments are taking action.
32:56
I think, I hope so that in the future is there is there is
33:02
going to be even more focused because because hacker tools are
33:07
becoming Advanced they exploit. The machine learning as well and
33:12
also, Also various automations, they are smart.
33:18
They are fast. And yeah, so we need actually
33:24
the same people sitting on, on our side, on the, protecting the
33:29
critical infrastructure, because I think the best, the best
33:36
people to, to identify threats, and, and descendants for
33:40
structure, the ones who actually know the best ways to Hack them.
33:45
So that it's also making events where they let people
33:51
participate in what was called with like white hacking where
33:57
you participate and try to hack some some system and present
34:04
your findings. So so yeah, I think in the
34:07
future we don't have even more attention to that.
34:12
Excellent. Yeah, and the do I mentioned
34:17
open source intelligence so what does that is actually a system
34:23
which is gathering the data on the internet, which calls the
34:28
internet. As you see on left hand side, it
34:32
can crawl information, gather it from your public profiles from
34:38
from chats from social, all the so, Phone networks also from all
34:46
your posts. So for example, if you're
34:48
posting something you attach the the location tag, it can also
34:53
crawl that and it uses machine learning algorithms, to make
35:01
sense of that data. So, I was also working on open
35:07
source, intelligence to, and we used in machine learning to
35:12
recognize It's in photos. We also used it to channelize
35:21
Sentiments and text. So you can sense that it
35:25
basically was a machine learning.
35:27
You can know like gather all the information about the subject
35:34
you looking at we version usual jargon is for this subject is
35:41
entity so entity can be anything.
35:43
It can Person. It can be a company, it can be
35:47
any University, it can be an organization, may be some group
35:56
of activists as well. So a sin helps to gather this
36:01
information or times. So basically, you can use the
36:05
rollers regularly together. Information about something
36:11
that's happening. Maybe you want to for example,
36:14
gather information about But you Demetrius.
36:20
So the things I could find it could be like the university
36:25
went to like the places to dive in your photos and so on, so on.
36:30
So as far as saying, yes. Yes.
36:34
So this is pretty sensitive information.
36:37
So they use of these tools is also limited usually for safety
36:43
reasons. For governments, for For the
36:48
institutions which have contracts also as social media
36:52
Pro platforms. So for example Facebook has a
36:57
contract with the government of France so they can provide some
37:04
information for them. So these this use requires
37:09
special permissions but it but this is also I would like you to
37:14
know that this is a very powerful tool for Tracking.
37:18
What's happening on the internet.
37:20
So for example, if we would have some kind of fake fake news,
37:27
action organized by some good, we could already detect the
37:33
group that's making those spreading Rose, fake news
37:35
because we would see the whole graph of interconnected people
37:42
interconnected Topics, which are happening within that Network.
37:49
And we could almost compare Point everybody participating
37:53
and they action. So it's a very powerful tool.
37:56
It collects more information than one human can comprehend
38:02
and analyze. So this is the power of machine
38:07
learning basically and and statistical methods.
38:10
And actually this kind of tool could have prevented elections
38:22
from being affected by the also by the targeted advertising
38:28
which you probably are wealth of conventional Attica Scandal as
38:34
well. So this kind of tool could help
38:37
to notice that some kind of Organized action is happening on
38:43
the internet regarding some some event as for example as
38:48
elections. So I think in the future these
38:51
tools are going to be even more popular for governments.
38:56
Many governments are already have something like this, and
39:00
they're already developing to make it like smarter more
39:05
intelligent and this is actually also a very interesting career
39:10
path. You like working with machine
39:12
learning because you get to combine many different
39:18
algorithms. So both like image recognition
39:22
and sentiment analysis and and many more.
39:26
So, using a scent actually, you can learn how to use machine
39:33
learning in real time. Usually, the architectures of
39:37
these tools include streaming real-time data analysis and also
39:45
graph representation of different kind of entity spread
39:50
on once you're on your analysis. So, yeah, I think it's very
39:56
interesting, it's very extensive.
40:00
Good thing is that you don't need to train all the
40:03
algorithms. You can already use many of
40:06
advanced algorithms developed, by, by Google In Facebook, for
40:11
example, the bird language algorithm also use the image
40:17
recognition algorithms for object detection and so on.
40:21
These kind of tools can also be used to track to track and
40:29
military equipment on the internet.
40:32
So for example, governments very often and have those tools as
40:37
well to basically look at the new types of military machines
40:47
are Imaging, or do they have some Some more action, more, for
40:54
example, trainings, military, trainings happening.
40:57
And so on. So it wasn't is basically an eye
41:01
for for the internet for what is happening on the web.
41:06
Wow. So, this is really interesting
41:09
to me, and it's a little bit outside of the scope of our chat
41:12
right now, but this tool of course, with great power comes
41:20
great responsibility, it feels like it could be very much
41:25
abused, if any of these governments wanted to use this
41:28
for nefarious reasons, Yes, so that's why I actually is social
41:34
networks are trying to protect their data like a while ago.
41:40
It wasn't such a concern but when as like the last year's use
41:50
of various scraping and mechanisms, he also has evolved.
41:56
So yeah all of those social networks.
42:00
Spurs have some kettles which actually don't allow to use it
42:10
and it can be used only with within governmental context so
42:17
it means you would be has to working for some governmental
42:22
project in order to have permission to do that.
42:25
So but I'm glad that you know that Social networks these days.
42:31
Try to protect save data. They try to To protect against
42:38
bots scanning their systems. So yeah, these are good
42:43
intentions, and I hope all data is safe in general.
42:47
Of course, it cannot be absolutely safe because hackers
42:51
also getting smarter, but, yeah, yeah, it's interesting.
42:58
It's also a little bit. I mean, even with government
43:03
agencies, being able to use it, that also kind of scares me.
43:06
I don't know about you, but the news and the trust that I have
43:12
in meat coming from the US, the trust that I have in the US
43:16
government is not that high in how they are using and spying on
43:21
its people. But that's a whole nother
43:24
conversation so we can continue with this and get into that.
43:28
Conversation and the ethics peace in a different time.
43:33
Okay. Yeah.
43:35
So B.
43:49
Also, under Angels, ye used to analyze the data.
43:55
Usually in, in real-time, data can be streamed from various
44:00
sources, so, meaning, it could be autonomous cars.
44:04
It could be a power plant, even a nuclear power plant.
44:08
So you realize how important it is to protect it.
44:13
Also other power, power power generators save could be your
44:19
Laptop or it could be a communication satellite and so
44:23
on. So most of those systems are
44:26
crucial to our everyday lives to our everyday communicating
44:33
driving and basically use of electricity.
44:37
So it's becoming even more important to protect those and
44:42
machine learning agents analyze data from those systems in real
44:47
time. And we can An alert humans so
44:52
far are usually, humans are the ones who making decisions like
44:57
what kind of action to take maybe in the future.
45:01
When we are closer to a GI, we're going to have machine
45:05
learning, agents, deciding what to do with those systems.
45:09
But so far humans are usually involved.
45:14
And what I like to emphasize that, you know, even though like
45:20
you don't like to be in like being spied as well, but with
45:26
cyber security, tools can also help to protect the most crucial
45:32
Assets in our lives. So there are two sides of the
45:38
coin and and yeah, and I think in this Futura golden be even
45:45
more attention to protecting protecting these assets.
45:49
And there is also Special like area of cybersecurity is power,
46:00
plant security, because power plants also have like their
46:06
internals way great, internal ways of working of controlling
46:10
it. So people also studied how power
46:13
plants work and how to protect them, how to detect to the
46:17
malicious activity and so on. So I think think this is also
46:22
one area where you combine, you can combine both like
46:26
interesting industry and also your cybersecurity skills as
46:31
well. So you can see that
46:33
cybersecurity is like in many areas of our lives.
46:37
And yeah and and you can work, it can have fun working in
46:43
different different environments.
46:48
Yes. So I had a chance.
46:49
Also to work on a project where we made distributed machine,
46:57
learning cyber security system where we had to have agents on
47:05
each. Each note of the system, meaning
47:07
each each computer I could say because we could not move the
47:13
data to central location. There was a requirement, From
47:19
from our customers that they don't want to transfer the data
47:23
from a system to central location or due to safety
47:27
reasons. So we developed an unique
47:30
distributed system which could analyze the data on each each
47:34
note separately. So yeah, these are, these are
47:38
the use cases and yeah. And yeah, I could also recommend
47:48
you if you like to learn a bit more about where we going and
47:53
also a current state of artificial intelligence.
47:58
You could read the book, AI superpowers very carefully.
48:02
It's a very interesting book. You could also sense approach.
48:07
China is taking when they using actually.
48:11
Most of the data human can generate.
48:15
And what are they cheating with it?
48:17
I think some of it is gonna land a here, as well, just in another
48:25
enough form in different kind of approach.
48:29
But because like this people monitoring it can also help us.
48:36
Like, for example, safety in the streets and and so on.
48:41
So I think in the future we're going to see some Some
48:45
approaches of that in our society, to talk about that a
48:51
little bit more. What approaches are like, what's
48:55
the difference main difference? Is that your, you see between
49:00
China and, and the Western world.
49:03
And I think the main one is the approach to the use of data.
49:08
It's very, I would say little regulated like, And a lot of
49:16
data is openly used by the government.
49:19
For example, people are being monitored, how we cross the
49:23
streets, do they behave? Well, and also do they drive?
49:27
Well, do they, I mean a lot of of behavior is openly monitored.
49:36
There are upsides of static increases safety and for us it
49:40
might seem maybe strange, maybe for our Like for more democratic
49:46
view, it seems like our Iowa privacy is being disturbed,
49:53
totally, but I think, in some way, we gonna see it, for
49:56
example, maybe the traffic could be monitored in the streets, in
50:02
order, you know, to prevent accidents or to like to get more
50:08
more information, where accidents happen on how they
50:11
happen. Some, I think some of those are
50:14
Yes, this is coming to us as well.
50:17
So I think it's a good look into how far away I can go.
50:23
When, when they use of data is very little like well it's not
50:31
very limited and this is also the reason why China is
50:39
advancing so much in this area but I think I think it's fun to
50:45
try to imagine how it can look in a democratic and Democratic
50:51
kind of culture. Maybe I think we are going the
50:57
same direction but our implementations might be
51:02
different. Yeah and I think it's really
51:06
interesting. Also what's happening in the EU
51:10
with the new regulation that's coming out because I know that
51:14
they Specifically said in the new EU regulation that you can't
51:18
do this, kind of social scoring, that's not going to be allowed.
51:22
And those those things that they explicitly say no to but then
51:28
like you say, how is it going to sneak in in other ways and how
51:33
is it going to happen within a democratic Democratic Society as
51:38
opposed to what's going on in China?
51:41
So this is super fascinating, I really I enjoy talking to you
51:45
about all of this. Thank you Monica for the
51:48
presentation is is there another slide?
51:51
I think that's it for the slides.
51:52
I'm going to stop sharing right now and Are there any any other
51:59
questions from anyone? I think?
52:04
I had one. We got some people saying that
52:07
they loved it. I had one back at the beginning
52:12
when you were talking about and I know we're almost out of time
52:16
but when you were talking about the different kinds of hacks and
52:23
how different hacks happen. But now I can't I can't find it
52:27
nor remember it. So I think that is a sign that
52:30
we should probably just stop and I'll try and talk to Offline
52:34
about the question that I had if it ever comes back to me, okay?
52:38
And you can guys to come find me on LinkedIn.
52:40
So if you still have questions, feel free to ping me.
52:45
And I had a really fun talking to you.
52:47
So thank you for all the great comments.
52:51
Yeah, this is so amazing. And there's so many insights in
52:55
ways that machine learning is being used that I had no idea
52:58
about. I really appreciate you coming
53:03
on here showing these slides, the slides were beautiful too by
53:06
the way, anyone who is listening to this in the future, you may
53:10
want to check out the slides on YouTube if you can but I guess
53:16
that's it. We're going to call it a day and
53:19
Monica you. Thank you.
53:20
We'll see everyone later. Thank you all for joining us and
53:25
yeah, take care. Thank you guys.