Evolutions of AI, ML, DL To Prevent Cyber Attacks

well thank you everyone for joining its in the afternoon and hope everyone's lunches was great today's conversation is going to be around the evolutions of artificial intelligence machine learning and deep learning um it's part of my passion I really enjoy this to speak in this presentation I'm going to present to you and and it'll make you think outside of the box as well as you as you go to these conferences and I go to blackhead and DEF CON you hear these these words right AI machine learning deep learning and what does it actually mean and today I'll have images for you and in explanations for you so that you can be a little bit smarter when you have those

type of conversations at the end of today's call I'm gonna also show you a website that you can see if there's any jobs opening for my my company at deep instinct I always like to introduce that so that if you're looking for a job we do have openings we are the mid-sized startup and we're growing exponentially this year one moment you just uh all right a little bit about myself I'm gonna pause here and talk a little bit about myself my credentials and and where I came from so I'm a cyber or security solutions engineer a deep instinct and I'll tell you a little bit about the company but I really want to just focus around the conversation and

my company but I want to take a second to pause if you see this QR code I I believe any time I go to a conference I try to meet as many new people as I can so what you can do with your phone and I'll give you a few seconds is just open up your camera app and point it at this QR code it will point you right if you have the LinkedIn app you'll point you right to my LinkedIn page or just simply hit connect and say hi or something I'd really appreciate that so that we had this deep conversations with each other and if we need to continue it in the future or any anything about

cybersecurity I can I can assist my background I am a mentor for CyberPatriot I do that it was one of my passion to help high school students you know learn about cybersecurity I volunteer and help this team get into cybersecurity competitions and then win scholarships so local to Northern Virginia this is something that I do on the side and that's how I started my cybersecurity career well my service call it adventure saying I help students I learned you know the Security+ techniques and cissp and I train these students how to do some basic level pen testing or defending of their machine boxes and this is through cyber patron I've grown beyond that got a job into

the cybersecurity I was a firewall you know cyber solutions engineer as a checkpoint before joining here at defense thing moving from then about four years ago I joined deep instinct and now I help local Department state and local hotel chains University and colleges in consult them about cybersecurity and the threat landscape and and how we adapt to this changing culture of you know I'm I have this product I have this product how do they combine what what is the Best of Breed technology and that's that's something I enjoy doing every day and I do it at nighttime too I consulted the Asia Asia teams over in Japan and Taiwan and that's quite fun the translations I have to do but that's my

past graduate from Virginia Tech I'm very proud about my alma mater and thank you so deep learning there's content available at the these links actually are you can click it within the PowerPoint but you can't click it if you're looking at the PowerPoint I'll share these slides with the team and if you actually message me on LinkedIn I'll send you these links personally and these are just some other deep learning there's a deep learning dummy for dummies here on the right hand side I can give you a free copy so that you can learn the basic level of deep learning and have a book for you to read in the next couple of days and then there's

also ebooks and videos that I can share with you but let's start what is deep learning right you may have heard it you may haven't you're actually using it every day or it's being used against you every day and I'll tell you have but first before I tell you about deep learning which is a subset if you look at the timeline from 1950 to 2010 it's an evolution it's the way technology has evolved artificial intelligence started in 1950s about the time when I would say computer science came about the the need for a computer to do what a human can do calculations the human teaching a software to do things that you know to automate things nuts came out artificial

intelligence not the matrix of course the matrix actually before the matrix and then became machine learning machine learning was the thought of all right now I can teach computer what to do I can teach a robot to do sir things like you go into a factory and the robot can pick up you know certain objects but machine learning takes it a little bit the stuff ahead in 1980 machine learning the algorithms became a little bit smarter to say I well let's correct for error right you pick up a bottle but it misses so what about next time can it fix it so they can correct for error automatically without the you without the human going in and doing it

for them and that a good example of machine learning was about late teen 1980s there was this IBM came out with his own eternal blue I believe or it was Project Blue where it was a chess player it's a computer chess player that was taught how to beat the best chess player in the world using algorithms using a learned mathematical algorithm to defeat the best player quickly and that was machine learning right it didn't it didn't come out of saying or I I taught it to do something and it did it it was more of let's learn how artificial intelligence these algorithms learn how human behavior works oh I gave a question here and I believe we're going

to be tough it's a I'll answer questions at the end actually you know what give me one moment let me see if I can view the questions while I talk I've got it I've got it he said I'm deep blue be blue okay people oh yeah thank you so much all right thank you for correcting me I never get these things yeah I have the chat window up so if anybody types in they're all I'm on mute but I'll try to get off mute quick for you awesome yeah okay yeah and keep asking me questions or and add anything to my conversation because it's kind of a small group here so I am appreciate your

time and then deep learning deep learning evolved out of machine learning where instead of your teaching robot what to do you actually give it examples right in terms of deep learning I'll give you some examples today say for example you give it a hundred ways that a robot instead of telling the robot what to do you say robots supposed to do this and the robot is then able to learn that because you show it what to do but if you show it a hundred times what to do then you don't have to teach it anything you have to create these algorithms it actually teaches itself and I'll go into a little bit more detail and then that came out

in 2010 a lot it around the age of graphic processor units and the need for videogame systems to learn human behavior and I'll show you some logos here that you might be interested to see all right so let's go into a little bit more detail about the the three the three categories artificial intelligence right these are some of the topics that came out in the 1950s the the basic example of any technique that enables a human of computer to mimic human behavior so that's robotics and language processing and in logic and planning evolved from that is machine learning where you now have decision trees Bayesian models it's a technique for computers to learn without being

explicitly programmed to do so right so the nearest neighbor learning about trial and error with reinforcement learning there's a lot of terminology here and if you ever take computer science and learn math and there's a math portion to it you'll learn some of these techniques that are used in machine learning evolved from that is the artificial neural networks the deep learning the now it it uses all the concepts from artificial intelligence and machine learning and adds on to that where now you're looking at neural network here you're now perceiving and predicting the future almost and it's it's multi-layered it's instead of it being linear you have different layers to it now I'll show you two pictures

what I mean by that so let's go backwards since 1950s 1980s classical machine learning and now I'll tell you then how this applies to cyber security I know it's a little you see cats and dogs and you're not sure how this can apply to cyber security but I'll tell you how these are all evolving into how we're protecting our home today so the the general machine learning approaches you label a cat a cat but human goes and labels a dog a dog based on human right the human goes and says this this cat is able to cat this is a dog because we know it's a dog you trained them you trained the system and

you create a model the model say art now whenever you see this picture it's a dog whenever you see this cat it's a cat and it takes features and what we have to do is there's two features that we take from these pictures from a human perspective to apply mathematical mathematical algorithms to it so what you do is look at supervised learning versus unsupervised learning the basic concept if you hear those two terms just think of input data is labeled versus input data is unlabeled in supervised learning you say this is the size of this is the length of the ear for cats so that's a label versus unsupervised learning is you just see a picture of

the ears of the cat and you say that's a cat and you haven't supervised you haven't labeled it yet so you haven't created the attributes and so there's two different types and it requires both supervised and unsupervised learning in order to build these algorithms let's take a look at that right so with machine learning you take raw data you take a picture of a dog and then you do manual feature extraction you say this dog because it has a long style that's what makes it a dog the eyes are a certain size the ears flop down that's what makes it a dog so you extract in that case three features you create handcrafted feet you use you create an

algorithm and then now you have a machine learning to say anytime these three things combined equals a dog and this is just a little bit deeper example of that where in in centimeters and you're actually creating measurements against it what about machine learning how does that apply to misleading features right so when you teach a dog when you teach an algorithm to say this is what a dog looks like this is what a cat looks like in your eyes you know what the difference of dog and cat is but if you only talk the machine algorithm machine learning algorithm and right now I'm going to debunk machine learning if you only taught it that the ears are long

then what about if a cat's ears are long or the face of a dog is round in the cat's face is also round what is this part in cybersecurity when you when you um when you teach something a feature and it catches it incorrectly I just call false positives now you have a false positive that this cat is actually a dog or this this cat could be labeled a dog what about noise a lot of times cyber criminals what they'll do is they'll create and I'm trying to create analogy here instead of being a normal original picture these are original pictures they add features to it they create noise instead of it being a PDF they change

them hour into a Word document and now it's changes structure and it's hiding itself or they create noise and how do you teach a machine learning if it only looked at the ears being long but then that years are now missing and it didn't account for the face right so this is just a picture analogy of you know how cyber criminals are creating mutations of malware for example here this is an example of a malicious font file on the left here this is malicious malware you can see on the bottom here it's this is where the the code ends the malicious content ends and if you look at the malicious mutation only changing it by one character now bypasses

signature-based attacks right if it's a signature now the hash has changed or the Charlotte Russe of 56 has changed or if you taught them the machine learning to only look at this row and it changes rows then no and in this binary this binary of maurer on this malicious font file and it changed the locations it's going to say all right well this is not malware anymore because it's not in the same location but you only taught it be that certain sequence that location that's just an example of signature base or the way machine learning would think about it so yeah one byte offsets completely changes the verdict so why not let a model decide on features and and that's

that's now we're moving into the generations of 2010 deep learning so now deep learning is very similar to the way the brain works training the brain and when you teach a child the difference between a dog and a cat you don't teach it based on features you teach it based on examples in a lot of examples every time you walk your your child down the street and you see a dog you point out that dog and say that's a dog next time you go to a dog park and you say those are all dogs your child is then able to quickly learn through examples that that's going to be a dog and then you teach it for the cat and now it's

able to differentiate different trees all in a Canon and even the fact that you can look at different types of dogs this is a chihuahua this is a small dog versus big dog this is all that you should stay away from right so the detection becomes instinctual and that's now we're applying that to cybersecurity where we're creating artificial brain this is artificial brain by deep learning is very similar the way the human brain works so you have in human brain stoma and then dry sand axons and internal moles and the synapses that teach you know the behavior of you know something hurts or you see a dog and a cat changes different parts of your

brain smells and everything well the neural network is very similar you create a lot of inputs you put weights to it saying is this right or this wrong and then you create a output to say based on what I've learned before and all the inputs the millions of inputs that I've received creates an output of this is what it is this is what I perceive it is and it requires all the supervised unsupervised layers and the neural network and I'll go into loving more detail them on these four topics so leave learning is not a marketing term it's actually something that came out first publications in the 1960s and 2000's if you actually look up deep

learning an example is Cova 19 they're using deep learning to correlate early detection of the lungs x-rays of lungs in pictures of lungs and the computer is able to say with a high I think 90 plus percent efficacy that this lung is going is covert versus a flu or influenza based on how the virus is impacting this person's lungs and this is using deep learning some just it's impacting the medical industry it's it's also now applied to cybersecurity so there's theoretical advances and in 1990s a lot of labeled data sets this requires a lot of data right it's not something that you could say based on five samples I can make correlation it's more like five thousand or five million

and it's done by data Sciences so it requires a lot of GPU computing around the time of you know when graphics processors became very very powerful I would say Nvidia and AMD and all those companies have crazies these the the most powerful part of your computer is your GPU if you have a nice one and that GPU can crunch a lot of data and we're utilizing the GPU to crunch all these mowers mom our samples and in the scientists are crunching all the data from their research studies to create algorithms instead of having to create algorithms themselves and in 2008 open source frameworks you create these frameworks I think there's five main frameworks for deep learning and and my company I'm

proud to say we created our own framework for cybersecurity all right so let's review a little bit um okay so let's review a little bit about deep learning versus machine learning on the top is machine learning you have the raw data you have manual feature engineering right so you have a person to say this is what malware is this is what malware isn't because of these 10 things you create an algorithm it could be mathematical it could be theoretical and and then you create a machine learning algorithm right but deep learning is different instead of doing instead of having a human come in in human being in the picture it just takes the raw data it

takes hundreds of pictures of those cats and dogs you tell it what the difference is organic cat is and then it's able to learn based on examples or malware you give it millions of malware samples looked at the binaries and then is able to then create this correlation to say based on the binaries based on the core of a program of a Mauer and we've trained it there's these millions of ways that these mutations that could happen that we can we can actually identify it as malware and and quickly to so it it's much quicker to use deep learning and here's a little bit more difference between machine learning and applying because it requires that human

interaction a human can only tell what they see right they only can tell what the picture of the dog and the cat looks like and the features of it and the only attract extracts certain features and this is a really only utilizing two percent of the data if you have am our sample for example that is that is a safe example ten megabytes it is a human able to analyze every single bit of the code in that ten megabytes effectively and and we've tested it and said no because humans have you know time it takes too much time or they have to work on different projects and in they had to prioritize certain parts of the sum our

sample we're deep learning at the bottom here you feed it a hundred percent of the data and it's able to create an autonomous intuitive automated nonlinear model of correlation right so now it's not from A to B or A to C there's actually a three dimensional way that we look at mauers in our code and also machine learning limited to certain file types if you look at total executables or only to look at documents you only categorize you know what the latest threats are for PDF files for example it's it's it's really going to be limited your machine learnings enemy limited the file types that you can train the human has a train versus what

deep learning you give it a million font files benign and malicious and is able to tell the difference between what's what's Delta right what is the mutated one what is the one that includes a ransomware inside of it and this is just reviewing the debunking machine learning in that not debugging machine learning is great it's it's good with his uses but again it's just the limited human knowledge and expertise right you can only human only goes through and looks at the part of the 10 megabyte file versus the full raw data and then mutations if you only teach it one way to look at malware or ten ways to look at malware what about the 11 or

the 20th or the hundredth mutation of it and here's a nice picture here to summarize what I was talking about this is the way the human looks at a file right you write code line by line you highlight when I was an engineer a reverse engineer I would highlight things that would make the anomalies look malicious they're the ones that I think that this is calling out to a command and control server or something but deep learning has made it so much more simple I just feed a bunch of these files into my model set and it's been able to say based on a hundred samples this is 90 percent ransomware and it's able to categorize and tell me what

malware is and isn't so in that case I'm feeding it a hundred percent of the data its autonomous it's automating my my um my my data sets and categorizing it and I'm able to train it on the unlimited training samples and apply the inputs from many different types of devices so there's different operating system agnostic so you can actually learn malware in any type of operating system including Android phones right we have a lot of malicious apps that we're looking at and we have a lot of apps in or Google Playstore even Chromebooks that we're looking at and we're feeding this neural network of all the malicious things and then and then taking out protections and prevention capabilities

against it more statistics this year on the top is machine learning the accuracy with unknown malware so this is actually with known malware machine learning signature-based can protect unknown malware where it's mutated and new and actually if I have some time today I'll show you an example of some unknown malware and I'll execute it on a computer to show you that with deep learning you can actually prevent an unknown malware in fact I'm a mutated one so with that you could see with accuracy is 99% and very low false positives with deep learning so here's some real-world applications of deep learning you might see here some familiar I know you know these logos but let me just show you one facial

recognition by Facebook at five or ten years ago Facebook wasn't really able to do really good their image recognition was very subpar it was it was used uploaded a picture and based on location or based on your friends list they were able to categorize and say that could be you but now image recognition has done so much advanced it has advanced so much that now with even a picture I have a hat on and if I'll upload my picture right now to Facebook you'll actually be able to tell me that that is me and not my cousin or not my brother oh and even your friends right if you have a picture of a group of 10 people and they only

see like one part of your face the image recognition using deep learning is able to to say are well based on the percentage we're going to say this we're going to tag this friend in your picture in that's advanced so much now the deep learning is involved it's trained on your pictures it trains on your in and now Facebook has there's neural network of all your pictures and the pictures around the world and it can correlate without using the small features like geolocation and that being your friend is able to correlate based on part of your face or even bottom of your nose and it's doing that in image recognition in many ways and Netflix right able to

do recommendation systems and learn your behaviors the highlight deep learning deep cybersecurity deep instinct is the first and only and this is my company and and I want to represent my the data Sciences engineers who created on via this cybersecurity platform where were the first and only cybersecurity to provide end-to-end protection with deep learning utilizing deep learning neural networks to prevent cyber attacks nuts that's quite an app eat in many ways 90% in computer vision speech recognition you go through Google Translate in able to actually caption your voice and caption the translation of of what you're trying to catch utilizing deep learning and if you actually look at jobs you go to Facebook jobs or Google jobs or even Tesla I'm

not we're not affiliated with them by just one say for example like Tesla and you look up deep learning you'll see that you know they require some of those those that knowledge of deep learning in order to produce their algorithms for you're looking at stop signs and in an image recognition it's not so much using cybersecurity yet so we're actually a new frontier only two percent of companies are utilizing deep learning today and so it's it's an open market for for cyber security and there's a lot of traditional machine learning right feature extraction but this pure deep learning is only um it's very new and I'm proud to be part of the frontier so the benefits of deep learning to cyber

security which is B side so let's talk a little bit cyber right prediction of unknown threats zero time detection and prevention and I'll have some time today it looks like we have about 15 minutes left I'm actually going to show you an example of zero time detection and prevention of unknown malware or zero you're a time malware yeah sorry not I know it's going to be malware that is found within the last 12 hours yeah and it's it's cross device multi operation system so as you saw on my maybe three four slides ago there was some it's vendor its operating so big agnostic so we've trained this brain that's now sits on your computer your

desktop or laptop utilizing deep learning to protect against cyber attacks and it doesn't require any internet connectivity so you can have protections on your computer without having to have connectivity to a cloud server you can if Kent connectionless because your brain your brain doesn't actually it doesn't get information from an external source it's everything in here it's in your noggin right that's deep instinct in deep learning it's everything is now on the computer it's like an analyst that checks everything for you as it gets written to disk in here pom this is a good one and then this is what I'm going in my last slide before I I show you some examples let's show you the way that deep instinct has

applied deep learning from training to prediction to response so it starts off with malware third party sources dark net homemade malware mutated malware millions of malware in malware file types executables PowerPoint Excel essence and as you saw on my previous light where where human has to go in and highlight all the malicious parts of a word document and then create a signature for it if we're just giving them files we just give this neural network all the files and it's able to crunch it and say based on these million files are merely a 10,000 of them were malicious and what of the 10,000 malicious ones are makes it malicious and then it creates this whole Pro

pattern of our samples in be able to predict alright well let's give it another it's give it a separate sample set it's that bird sample set of another hundred thousand mile where do we see some similarities between the first sample said and the second sample said and then creating correlation against that and creating another algorithm or another pattern a deep learning brain right and you just teach it this way hopefully that's clear I'll pause a little bit for you guys to kind of read through this so it requires a lot of these data sampling supervised training and creating noise you know mutating malware tweaking a little bit maybe adding a different line and then and

then going with with the protections and engines through this deep neural network we retrain it through NVIDIA GPUs so Nvidia is one of our our partners that we use their business GP years not those you know 1080p eyes or the 220 atti super 2080 super now I think it's the best one for consumer base I'm a I like the game okay I like the game a lot but so 2020 80 those our consumer base but then they have these business grade GPUs actually did sit in the data and via data centers that we we partner up and we train our brain and our neural network with their GPUs and his own can be done on GPUs because of how much data

that's being added to it once it's trained takes about 24 hours a train its then then we create these petabytes of data into one single agent and this is where defensing is we are an endpoint security company we provide endpoint security on devices laptops desktops mobile devices and we look at any file any device any OS and it's a lightweight agent that runs on any very low CPU and only requires two updates a year so this is a deep instincts you know that's our market it's the threat prevention threat detection in response or predict and respond it requires only two updates a year and that's that's actually pretty powerful because if you think of it how

many times you have to teach your brain what the difference between a dog and a cat is you'll have to teach it anymore because you already learned it ten years ago or twenty or thirty years ago same thing for the brain you've already taught it what malware binary looks like and then what malware binaries could be mutated to look like in the future then you don't really need to have signature so it doesn't require signatures it just requires a new brain so we update the brain two times a year so with that is there any questions um before I start I'll do a little demo so that you can see this in action but any

questions so far if I intrigued you guys based on the topic today yep so we do have something that has come in to us we have seen stories of bias in learned algorithms ie mortgage decisions main maintained racial bias from the people that were watched or computers seen so many pictures that they assume coolers coders are men and women who are in the kitchen are you seeing that bias in the learning in in the cybersecurity implementations examples and how to fix them yeah yeah yeah there's bias yeah bias comes from the human yeah there's a lot of it even in deep learning I can say that bias is always added by the way you train it the

weights that you put on it so there are examples so say for example let me show you here I don't have a picture but is to kind of put into perspective if you teach malware too much if you teach the known networks too many times and you teach it too many malware samples what will happen is then you'll get another neural network that only learns on I only can find malware it doesn't know the difference between malware and benign is so what you have to do is you have to create a baseline and the way to create a baseline is you say what is clean versus what is malicious um you don't always teach it on one thing right

the only catch on one thing and that comes out false positives because if you only teach you know malware and you teach it too many times then you over teach you over you overcompensate the the algorithm and they may there's two things I could happen one you won't catch anything you taught it so much that you won't be able to say the difference between wrong and right is or you know white and black is or the second thing is you'll you'll catch things but it will be false positives because you don't know what um what's clear and what was actually the gray area and so you fix it you fix it by and then part of the the training

actually if I go back a little bit part of the training is a balanced data set right you want to balance want to balance and not not create too many too much noise into the samples and just enough and that's it's a it's also part of defensins IP or data sciences they have this train they know how to not do too much or too little and that's the whole brain behind putting it into a client and protection in more details Jeff I can go offline I do have different I have actually has different slides on peer deep learning and we can go into a little more detail if you like to look at that in examples in the real

world without outside of this call but hopefully that's able to answer your overfitting right here that was overcompensating what I mean by that is overfitting avoid memorizing the specific files available in the data set right you don't want to teach it only one thing and it starts to only look at that when that one thing and that's overfitting is a deep learning terminology it's yeah that's and if I go to back to one more example how do you fix it it's right the for the example with Cove in nineteen and they're using deep learning you show a lot of pictures of the Cova nineteen lungs and you show a lot of pictures of the influenza lungs

but just enough for both so that you don't only teach it that info on that Kobe nineteen long and you know everything else is negative and false positive I just wonder in my head you know how many influenza patients are actually being diagnosed as Cova nineteen I don't know but that could be an issue right now all right false positives that's really bad no false positives okay here let me show you an example I've created a lab for you me go into my VM I create a sandbox environment and this is a typical one that I share with some prospects and clients and run through POCs so virus total virus total is it's a beta

database of uploaded malware and uploaded malware samples and a consumer-based virus will allow you to upload any document that you find suspicious into this database to see if any other vendor has detected as malicious and once it's uploaded it creates signature and and other vendors are able to use that what deep instinct what I like to do is I look at my word that is less than 12 hours old so everyone can see my screen it's Q and 13th and 12 hours from now I'm looking at portable executables and just to create a nice little so I'm not looking at false positives I look at 30 other vendors that find it in less than 45

detected so let's take a look here I have the deep instinct client installed on my device it's not using any signatures and this is the client that sits on them on the mobile device or laptop or even server and it's able to as I'm downloading each one prevent each one in real time right so it's able to identify this is a poor that's too deep static analysis this brain is able to detect that it's a pool and I made sure you know their example download air before the file so every file that's being written to disk it's intercepted by this brain to say all right I'm gonna prevent this right now before you could actually and it goes so

quickly that only a tenth file is is found not even the full files even being able written to disk let me show you another sometimes quarter barbecue bowls are easy right you can name it a potentially unknown application what about a PDF that isn't that's actually a malicious PDF so chapter one or a story so when you're looking at other samples right here PDFs do the trojans out of PDFs or I'm going to close this this PDF fail to air so this was only it was probably less than a megabyte and it's already prevented in real in pre executions this is called pre execution where it doesn't sit on the desk doesn't require sandboxing it never actually

executed or intercepted and at the look at here it's like at the chrome cache level right before it can even be ridden to this it's already been analyzed and prevented what about another way a lot of you know machine learning doesn't some engines don't look at RTS which is a font file or a file that has certain macros inside so right here is saying I couldn't find ones that were today because not many malware come in RTF format so I had to a couple days but that's beside the point let me just download an RTF file fail to download so quick right so and then I can go into the year with virustotal and I think I click on something by

accident I can look at this hash and these are all the other signature based protections and this is some of your your consumer based signature protections are able to count as malicious and then there's some of the machine learning companies out there that undetected because it didn't isn't one maybe not able to look at you know process the file type on the file type is RTF and it's not added in the machine learning it wasn't trained on RTS or the the is undetected because there's no signature and there was no machine learning algorithm that was that was taught about you know how PDFs can be manipulated for RTF in this example like to point that out and keep in sync the

actual client is actually three months old the signature is three months old so three months ago was the latest update on the client itself and deep instinct endpoint protection and we're protecting our pre execution and that's the way it works online and offline how about this since we have a little bit of time I'm going to disconnect my internet so now there's no internet connectivity and in here I downloaded the top 25 malware on samples you know extract here so the passwords and on the top 25 from virustotal and as each file being written to disk is actually being deleted quarantined removes before it can infect the computer it doesn't even have time to sit on the disk and this brain is able

to understand I haven't ran this one before yet so this is brand-new and this is offline this is no internet connectivity 25 out of 25 samples prevented in zero time this is what deep learning this is deep instinct and this is a what we've created the algorithms from AI machine learning now deep learning to to apply to an endpoint security product all right without that yeah that's just some of the cool examples here I hope you guys were intrigued either those I think that's fun I'm glad all 25 samples were caught I feel safe and the user never in phishing attacks right if you get an email with the PDF that's malicious boom then the engine will well the brain will

then detect and prevent on the decipher the cyber attack us so would they have a minute left what I would like to also show you in a recap let me see if I agree have a recap slide recap um here so we have the past the past is the Aviara signature based in heuristics the present is machine learning the era where we have detection on human selected features it's limited but now the future deep learning has a higher detection rate that she skips the human engineering capabilities and now be able to defeat it raw data in a file in actually train on the algorithms create a brain now presented on the computer and that's the future with deep deep

instinct and deep learning now I promised you at the end and I know we might be going a little bit over join the revolution of cybersecurity and find your passion we have a team here at defense think we're growing we're about 150 employees and we're looking for a lot of talent if you know anybody who's interested in deep instinct just type in deep instinct and go to a career opportunities page and I welcome you join my team if you have the passion for cyber also a passion for emerging technologies please feel free to reach out to me or if you want to further the conversation around deep learning and deep instinct and and how this can apply

to your corporate environment or your small business please reach out to me as well as I am I shared with you my LinkedIn so with that I really thank everyone for your time I knows quite a lot of information a lot of slides I think it was about 40 slides but we went through it quickly I hope you uh you have a great day I hope everyone here has a great day and stay safe I'll leave this picture up if you want to take your picture just take your phone open up your camera app and and you'll automatically point you to LinkedIn to my LinkedIn page it Sabean I actually did that at the beginning and you're

right it took me exactly to your LinkedIn page to connect so also you had mentioned about sharing your presentation with the group so I did touch base with Michael as we you were going through the presentation and he said that you got an email kind of from the the speaker to send the bio and different things there were some communication there he said if you just send it to there we'll make sure we can share it very good I'll do that I'll make sure I send it to the to Michael okay and then Jeff did say thanks that he was just serious on the message response is there any other questions or anything from this session comments anything you guys

want to post in the chat yeah awesome good stuff I like it you know this makes me happy that is it's it's relevant and it's fun