Machine Learning & Artificial Intelligence for the NS2 Mission

[Music] there's a lot of depictions of artificial intelligence in popular culture here's three evil robots you might recognize and then there's Siri there's even those like Elon Musk who warned that AI is a fundamental risk to the existence of human civilization but what really is artificial intelligence what is your technical lead talking about when she says that machine learning could help your organization perform your mission better or faster why should you listen to your team when they want budget to add AI and machine learning to your processing and exploitation of data there's many overlapping terms that are used about this technology there's a lot of buzz around these approaches to automating data analysis my goal for the next few minutes is to help make clear what the nature and potential of this technology is and perhaps to get you thinking about how it might be useful in your particular mission or business area here's one definition I like for when I study this stuff machine learning provides a set of tools that use computers to transform data into actionable knowledge back in the day it used to be called data mining we were just lowly data miners toiling away in the mine but now they call them data scientists but let's break this down what does machine learning really mean so in a traditional computer program the programmer tells the computer how to process the data step by step then that program is executed and an output is produced just like if you write a formula in Microsoft Excel and run it against your spreadsheet you know what you want to compute and that formula is static unless you change it what is different about machine learning is this the data itself is used to alter the program that's what the learning means in machine learning the computation the math that you're doing on the data can actually be altered by the data you pointed at so we say that the program learns from the training data set humans still write the program but they do it in such a way as to allow the specific data being used to alter the computations that are executed what they're trying to do is build a mathematical representation of the relationship between data elements a symbolic picture of reality and of probabilities they call this set of formulas a model and any mathematicians in the room will recognize the Bayesian theorem up there which will go through that in detail now no so the model gets inputs from the data set data scientists train the model on a known data set such as a set of known bad actors or they can train the algorithm to recognize words to do what's known as natural language processing then they point that trained model at a test data set to see if it correctly characterizes the nature of the observed entities they evaluate the accuracy of the model by looking at the output in what they call a confusion matrix they want the model to find the greatest number of true positives and the least number of false positives in the data and vice versa so the data scientist Tunes the model and runs it again and again until they're happy with the output but they also have to be careful not to tune the model too closely to the training data or I won't find a thing when they use it with real-world data that's known as the problem of overfitting the model this is why experience data scientists are so valuable there's a subtle art to tuning the model but not overfitting it for example if this was a bank trying to evaluate which loans they should issue based on credit history and balances they would want the model to correctly identify people in the test data who defaulted on loans but not deny alone to someone that proved to be a good paying customer in the test data makes sense right finally after tuning they turn the model loose on the real-world data and they hope that it now produces insights and actionable knowledge so that's a summary of what machine learning really means but what kinds of machine learning algorithms are there let's summarize the major kinds and the types of business and mission problems they might help to solve the first category I'll discuss seeks to predict future values from past values we call this type of math regression you may not even realize this but I bet all of you have studied a lot of regression output recently it looks like this data scientists train the algorithm using the past latitude and longitude numbers of the hurricane and they alter the algorithm with variables such as water temperature the cone of uncertainty is the probable progression of future values based on the past now regression has obvious applications in weather forecasting but couldn't this type of algorithm also be used in the mission to give a probable location of an adversary based on terrain and past movement the next category I'll discuss is called categorization Association and clustering this is used to answer what I call the Sesame Street question which of these things belongs with the others clustering seeks to assemble groups that are most like each other and most different from all the other groups the commercial world uses them for customer segmentation people who bought this might also like to buy that can we think of a mission use where you'd want to look at a huge pool of data and get insight into which of the entities should be perhaps grouped together decision trees known as rule learners are used to divide data into smaller and smaller portions to identify patterns that can be used for prediction they're often used in the commercial world for credit scoring diagnosis of medical conditions or marketing studies of customer churn in the mission decision trees could be used to help analyze courses of action what is most likely to occur and what is most dangerous another type of algorithm used for predicting behavior is called Market Basket analysis it's actually named for the original use case which is analyzing point-of-sale data to understand which things are purchased together this is where we get the classic example of the relationship between diapers and beer analysis of sales receipts can tell us that when the wife sends the husband to the store at 11 p.m. to buy diapers he also buys a six-pack therefore put the beer near the diapers basically Market Basket analysis says if a given entity has feature a as well as feature B that implies it might also have features C so can we think of where this type of logic could apply to the mission how about in helping to predict self-radicalization the last type of algorithm I'll discuss is called artificial neural networks these are the ones that most closely fit the science fiction idea of artificial intelligence as the late arthur c clarke wrote any sufficiently advanced technology is indistinguishable from magic in neural network algorithms the math is processed in packets that are analogous to how dendrites and neurons connect in the human brain at each node the weighted inputs of the data are summed together then they're passed on to what's called an activation function which measures whether or not the input should be passed on to the next neuron the output of one node is passed to the next which also measures the weighting and performs its activation function the neural network learns because the weights of the connections are strengthened or weakened based on the exposure to the input data it's very similar to how a baby learns kind of blows your mind doesn't it so that's a sampling of some major types of machine learning algorithms now let's discuss two examples of use cases that are ripe for machine learning the first is in the cyber domain of course a firewall is a must as a first line of defense but it can't be your only defense because it can be circumvented through deception using social engineering or even by intentional insider threat so you need to constantly monitor your network in real time to look for subtle signs of compromised machine learning can help identify attacks at the speed that's required using clustering algorithms features such as subnets departments people's roles access locations shifts and so on can be used to identify what is normal and focus the attention of your responders on that which is not normal once the team has investigated the anomalies they can you the positives and false positives as input to subsequent machine learning to refine future results the end result is a system that surfaces problems faster and with fewer false positives here's a second use case where machine learning can help further the mission we know that foreign extremist groups are using the Internet to recruit and coordinate their activities counterterrorism and law enforcement organizations are investing in machine learning to help them analyze these data streams to learn how these groups are organized and to help prevent future attacks first natural language processing is used to prepare the incoming data then classification algorithms can bucket together the posts using subject matter sentiment location places and so on time series analysis can be used to see if chatter is increasing or decreasing around a specific topic or target and finally a decision tree can suggest groups with similar messages to identify influencers or cross memberships without the automation of machine learning the task of processing the volume of data needed to gain insight would be insurmountable these mathematical algorithms have been around a lot of years but what's different now is we have an explosion of big data we can use to train models and use machine learning to get signal from the noise this can help get to the true answers we need when they're needed but there are two big challenges in delivering that benefit the first is a people problem it takes a long time to train an experienced data scientist and the second is a technology problem machine learning using big data is a heavy computing performance problem here's the people problem look at the steps of the data science process they have to know how to wrangle the data from multiple sources they need the experience to prepare the data correctly they need to know which the pic and then train it with learning data set then they have to evaluate the output and tune the model for a given mission problem all of this takes a lot of experience and skill that's why I say that good data scientists are as rare and as talented as rock star unicorns but ns-two can help with this challenge to bring the power of machine learning to bear on your mission si P has developed a set of software tools which helped you democratize data science by automating the data preparation steps according to which algorithm is needed for a given mission or business problem the idea here is that you can arm your subject matter experts who are not mathematicians or data scientists with the capabilities of machine learning this is a powerful combination we can help your organization leverage the deep domain expertise of your analysts and empower them to execute machine learning functions on the mission problem that no one understands better than they do the second challenge to executing machine learning is that it's very compute intensive scientists using traditional computing environments routinely set up jobs that may run for hours or even for days this is caused by a technical problem that computer scientists call the disk to memory bottleneck this bottleneck also exists in distributing computing environments that are used MapReduce and Hadoop that's where NS 2 can help si P has made a multi-billion dollar investment in a new approach to data processing we call si P Hana it makes massively parallel processing available on standard servers that you already have in your in your data centers or in your cloud Hana is an in-memory they processing system which means that the time it takes to run machine learning jobs can go from days and hours down to minutes or seconds on unlimited Big Data and we've assembled this processing power into our digital innovation system called Leonardo this framework can provide the analytical flexibility and processing speed that can make all the difference when an answer is needed in real time at the moment of truth the best run commercial organizations are calling this the fourth Industrial Revolution machine learning is transforming the way businesses are applying automation to gain insight at Ennis - we believe that this innovation can be applied to the third offset and to fifth generation warfare our goal is not to replace the human analyst with artificial intelligence it's to help your smart people do what they already do so well but faster and scaling to ever-increasing sizes of data in the needle in a haystack problem the idea is not to try to have the computer find the needle it's not to replace humans with machines instead the goal is to shrink the size of the haystack so now I'll challenge our government leaders here today as well as our systems integrator partners in the room where could this innovative technology be brought to bear on the hard data problems that affect your mission thank you very much [Applause] we live in a connected world exabytes of data are produced every day with information comes opportunity commercial innovators have developed novel ways to collect and exploit massive data some of these advancements hold the potential to transform the national security mission wearable technologies collect biometric data to measure customer behavior and help retailers better regulate supply and demand the same real time processing can enable more rapid data fusion from is our sensors data analytics in the digital boardroom allow corporate executives to make better informed decisions the same technologies can deliver situational awareness and provide the real time common operating picture the mission demands managing a global manufacturing supply chain requires fusing data from multiple sources sensors streaming human and data stores it's the same multi infusion that drives reconnaissance logistics and force planning in the corporate world leaders use social media to listen to voice of the customer likewise sentiment analysis can provide deeper context into a subjects influences and intent commercial innovations can revolutionize the mission this technology is not a future State it's today's reality how will you innovate [Music]

Channel: SAP NS2

Views: 491

Rating: 5 out of 5

Keywords: SAP, SAP NS2, Machine Learning, Artificial Intelligence, AI, National Security, Defense, Intelligence, cybersecurity, SAP HANA, Neural Networks, In-Memory Computing, Data Analytics, Data Miner, Data Scientist, Bob Palmer

Id: LAZSGe5eALY

Length: 18min 51sec (1131 seconds)

Published: Mon Feb 05 2018