Industry sign banking illinois word now
hello everyone I'm Tim scarf I'm a data solution architect in the customer success unit and like Lauren was saying earlier I'm also a gardener we've got the grass down here and it's our job to go into the banks and to tell the Microsoft story on on the cloud we've got an incredible cloud story and we've got an incredible cloud hybrid AI platform and obviously I'm going to be talking about that today but generally speaking there are a lot of outdated perceptions about Microsoft and it's our job to go in there and update these perceptions because our company has been through a real revolution in the last five years it's really important for us to get those messages across clearly so in today's presentation I'm going to talk about FinTech and how it came about I'm going to talk about our experience talking to the banking sector in the UK I'm going to talk about some of the applications of artificial intelligence in the banks and some of the challenges that the banks have and then I'm going to talk about a partner spotlights and I'm going to finish off just with some key take-home messages so how did FinTech start about ten years ago after the financial crisis you had lots of newly minted computer science graduates and you had lots of people who worked in banking who are experts in banking who are either looking for a job or being made redundant at the same time you had the cloud computing revolution so you had this kind of mix of people that are very very skilled in banking people that are very technically skilled and a low-battery of option for technology so it created this kind of FinTech revolution that we see now the Chinese in some ways are ahead of us on FinTech I'll give you an example so about seven years ago applications were the new horizon mobile phone applications and people tended to have hundreds of applications installed on their phone but after a while they became bored and they spent most of their time just using social media applications and messaging applications like WeChat and what's that so the Chinese companies they've got quite canny to this and what they did is there was a company called pay key and they integrated banking technologies right into the keyboard inside the application so if you said to your friend I'll pay you for a coffee the banking icon would come up inside the keyboard and you could conduct the transaction right there in the context of the application now there's a lot of excitement about blockchain at the moment we'll talk about blockchain in the context of distributed ledger technology and smart contracts so I'll give you an example in trade finance if you have a loan application you have cheques and credits all the way down the line you have a slow long-running process and you need to have a smart contract because it needs to be tamper proof you need to have a consortium to validate and verify contract you need to have an audit trail now when you take smart contracts and you take AI then you put them together you have something which is going to be an absolute game changer for the industry and that's something that we're going to see over the next five years or so so AI is future of banking data volumes are growing and data volumes are going into systems and humans can no longer keep up with this volume of data it's just going too fast for them to read and infer and respond to so we need to have new systems of artificial intelligence to create automation of understanding this data to create a sense of value to create new revenue pools and to create new jobs jobs that we didn't think we're possible just a few years ago Microsoft is fueling innovation Microsoft Research is a worldwide function and we spend 12 billion dollars a year on research and development a week ago Microsoft ventures launched an AI competition with a prize fund of 3.5 million dollars so Microsoft is fueling innovation at the grassroots level and around the world so this is our man Satya these are satis words it's our goal to democratize AI to empower every person and every organization to achieve more so Microsoft is an incredibly innovative company especially around AI but we have some of the very very best people in the industry sitting on our bench working for Microsoft let's let's talk about an example so speech recognition a couple of months ago Microsoft released a paper on speech recognition which takes the average work error rate down to 5.1 percent this is a milestone it's an elixir that we've been trying to achieve for about twenty years there have been so many applications and deep learning over the last 10 years that have broken records and Microsoft is the world leader in speech recognition it's only really IBM that's giving us any competition at all I'll give you another example what about computer vision in 2015 Microsoft won the imagenet competition the image net is the annual computer sorry the annual Olympics of computer vision we published an architecture called ResNet and resna is now the de facto standard deep learning architecture which is used by all people that publish papers on machine learning and computer vision so let's talk about this democratization thing under Satya and we're a transformed company and people find it so surprising when I tell them that you can run Linux on the azure platform we are Platinum members of the Linux Foundation our deep learning frameworks the NDK is a hundred percent open-source it's available to anyone to use and this is part and parcel of how we're democratizing our AI platform because people from all backgrounds they can start to compose and create deep learning models and when I say compose they can you know use models together in completely new ways and this is how we're democratizing AI machine learning is becoming a form of software development and people are using code to create machine learning models and they're publishing their models on sites like github but what do we mean by this word democratization I think there's a lot of fear and misunderstanding about AI at the moment and by democratizing AI and by making it ubiquitously accessible and available to everyone it will make people understand that a AI is there to empower every single person to achieve more so AI is innovation AI is about bigger opportunities it's about bigger markets people shouldn't be afraid they are it's not just a tool it's a radical shift in technology it's making technology ubiquitous invisible and intuitive it's not about replacing humans it's about harnessing humans collective knowledge and experiences to make better decision and enriched our lives and helped us relate to each other better now there was a study a couple of years ago from the University of Oxford and it said that in 20 years time nearly half of jobs will be automated by artificial intelligence but the problem is that's only half of the story the other half of the story is that artificial intelligence is going to create a whole suite of new jobs jobs that we couldn't even have imagined being possible today I think that in five years time people working in financial services will have basic AI training it'll be mandatory so AI is not about robots it's about making our lives easier through intelligent services then anticipate our needs organize our environment perform time-consuming repetitive tasks and freeing up humans to be more creative and more productive no AI is about automating about supporting growth it's about innovating we're helping the banks innovate data and population is growing and this is a huge opportunity for banks to capitalize on data is the new oil and banks hire people that are very fluent with numbers that are intellectually curious and banks are an interesting industry because they have more data than any other industry so banks are in an incredible position to capitalize on this AI revolution so let's talk about the the data and compute explosion so data has been increasing exponentially over the last ten years one of the main reasons why machine learning algorithms have suddenly started performing so well and the results have become so good it's because of the availability of data and public data sets but not only that we have the advent of cloud computing cloud computing is completely democratized access to computing resources to average people so this is how we democratizing AI and particular machine learning applications that have been you know seeing improvements over the last ten years are things like natural language processing speech recognition computer vision and reinforcement learning so let's talk about the machine learning stack at Microsoft now Microsoft is a transformed company we're not quite as prescriptive as we used to be and we've been talking to data scientists about how they do their process and there are some key trends that have been emerging one of the first trends is that machine learning is a form of software development it's really important for data scientists to work with code and to be able to compose models in new ways and to have very very fine control over those models deep learning in particular is quite interesting because it has this form of composability pre-loaded deep learning model that's a convolutional neural network for doing image and then you can download a model which is for doing language processing and you can plug them together and you can train in your network that will generate a caption for an image even though you're using two modalities that were completely separate ten years ago machine learning algorithms were very different if you were doing language processing you had one set of algorithms if you were doing vision you had another set of algorithms this deep learning revolution is making some very interesting things possible another thing we've noticed is that every customer has very unique data requirements especially in financial services and people tend to have data on SPARC clusters sometimes data is in relational databases sometimes data is in blob storage sometimes data is on the person's own machine or on a docker container we need to be flexible and pragmatic we can't be prescriptive about how people do data science another thing we've noticed is that data scientists want to consume models from anyway they have very latency requirements for example they might need to consume a model on an edge they might need to be able to publish it on their own machine or on the cloud another thing we've noticed is the diversification of hardware languages frameworks and tools data scientists are very very particular about the tool set that they use and we shouldn't be prescriptive and telling data scientist how they should work we should come to them and we should work using their tools and their languages whether it's R or Python or tensorflow or CMD K or whatever it is so just talking about this diagram from left to right it shouldn't be it shouldn't be any surprise to you that financial services and the Big Bang's software is the service offerings or our platform as a service offerings and the reason for that is they don't want to put PII data in the cloud for doing machine learning what they want is the that we have on the right hand side now these are that the hybrid model of execution and their code first systems and they also support a novel prediction architecture because quite often data scientists and financial services need to do things that are quite unique so just in sort of summary we're not being prescriptive with a new Microsoft with open source an outward-looking you can start on your machine you can use a code first system and we have a hybrid model of execution and we support novel prediction architectures now this is the new machine learning snack this is what we call our hybrid cloud AI stack in this stack we support all languages all frameworks all compute context and code first data science so what that means is we move the computer to where your data sits wherever that is whether it's on a spark cluster whether it's in a relational database whether it's on your own machine it means that if you choose to operationalize or train your model in the cloud we have Elastic Compute we have an incredible compute platform we have the at the the azure ai supercomputer which I'll introduce on the next slide through our new experimentation service we support a model of collaboration so our experimentation service takes a lot of cues from the software world the devops world it uses um git which is a distributed software source control system we have a new data wrangling tool which is inspired by the flash fill feature inside Excel so you give a few examples of some kind of data transformation you might be extracting 2:00 p.m. to 3:00 p.m. from a date what happened is it will infer what you wanted to do it'll generate some code behind the scenes and then it will execute that code where the data sits because the data might not be on your machine you're just looking at a preview the data might be on a spark cluster somewhere else another thing is we've created some AI extensions for visual studio code visual studio code is probably the best code editor an IDE available right now form and a measure of how good it is is when Google do demonstrations they're using Visual Studio code so it's really really cool stuff so this is the azure aai supercomputer this is incredible so first of all we have SGX on clothes this means your data is encrypted at rest and in motion we are the first cloud provider to roll out tesla p100 graphics cards across the board for accelerating deep learning workloads we're the only cloud provider at the moment that has it we are making FPGAs available as Hardware micro services this isn't public yet but think the Bing team are already using it for their search algorithm so if you want to be doing AI Azure is the AI supercomputer this this is our vision so let's talk about the journey to AI automation in banking so first thing is this concept of reg tech so reg tech is the new FinTech because a lot of the back office machine learning applications in in banking and actually helping the banks comply with the regulations from the regulator so this is a huge area that they look at another thing we look at is you can kind of have a split between the front office applications and the back office applications the front office applications tend to be things like customer for Penn City models you know is the customer going to churn what's the next best action how's the customer going through his mortgage application is the customer about to phone up that the helpdesk that kind of thing the back office applications tend to be things like credit and risk compliance and anti money laundering and detecting suspicious behavior now these are models that need to be explainable now we have this problem in machine learning especially with deep learning that is very very difficult when a machine learning algorithm makes a prediction to explain what it's doing so this is this is one of the key challenges the other key challenge of course is the sensitivity of the data on the front office applications the the bank so far more likely to put that data in the cloud because it tends to be sort of CRM data or website data so we've got these kind of two challenges that we're working with the banks but you know we're on a journey and you know we're going to help them on that journey to the cloud but the back-office applications a bit more difficult not only that the banks have some pretty complex legacy they have very large application estates they have some very specific requirements around governance data security and their data tends to be siloed all over the place so I was kind of talking about this before but here's an example of some of the front office applications now we're dealing with the Tier one bank in the UK at the moment and they already have hundreds of machine learning models in production doing these kind of things and some of them are in the cloud one of the things I noticed is the dearth of deep learning applications in this list now the reason for that I mean that there are a couple you know the audio transcription of customer interactions that would be one scanning in the customer documents automatically that would be another one n
w this is where we have a bit of a gap in our offering because we have incredibly sort of well-trained SAS services for doing deep learning but if the customer wants to do deep learning on their own machines they can use the same composition on your network they can use the deep learning training toolkit but they don't have the millions of records that we do to train the network so you know that's a bit of a gap but generally speaking these are reasonably straightforward use cases when we look at the back office applications these are a slower journey to the cloud and the reason for that is they need to be explainable there are a few interesting exceptions here so for example when you're looking at a fraud model you don't need to explain to the regulator why you've made a decision if you look at a credit lending decision you do need to explain to the regulator if you're detecting suspicious activity you need to be able to say this is suspicious because of X Y and Zed it doesn't make certain another point I want to get across here is that it's still okay to have machine learning models because these are just augmenting you know rules based systems you can think about it as another expert in the room I mean at Microsoft we do revenue prediction in our finance function we have lots of machine learning models and they're working incredibly well they're saving us so much money that we used to have hundreds of you know Excel spreadsheet jockeys in our financial function so it's really really helping and our CFO just looks at the results of these models and makes the decision so it's really helpful to have other experts in the room but I'm going to introduce one of our partners so this is ASD now ASD is currently in some of the major banks doing things like anti money laundering and suspicious behavior detection now what really differentiates these guys is that they can tell you why they've made the decision so they're using artificial intelligence but they're using something different to machine learning they're using something called TDA which is topological data analysis what they do is they take the data set and they load it into a network structure which is the same as a graph for those of you that have done don't computer science and they colorize that graph depending on some kind of metadata so by looking at the shapes of these networks they can not only say is something suspicious or they can make a prediction but they can critically tell you why and that's why the banks are so interested in this company in this kind of technology the kind of things that they do are sort of segmentation intelligent alerts and intelligent topologies give an example of it apology they would know what a pharmaceutical company looks like they would know what you know the kind of payments that come into a pharmaceutical company you know would be from health care companies they would know what an oil company looks like oil companies tend to have high cash reserves for example so they can build this kind of topology they gave me an example of a big bank they're working with in America and one of the use cases was sales practice misconduct so this particular bank has two hundred thousand complaints a year about mortgage applications and the regulator comes knocking on their door and they say why do you have all of these complaints about mortgage applications so you know they really exposed there because first of all they want to understand why do people have trouble with the mortgage applications but secondly they don't want to be fine by the regulator because people are having so much trouble with their system so explain ability is really important so coming back to Satya we're not simply building tools we're building an intelligent fabric on four principles collaboration mobility intelligence and Trust everything we do here is built on trust the value of our cloud is our fabric we are the new pragmatic Microsoft we have an incredible culture now which is built on collaboration and empathy our vision is a ubiquitous AI platform built on the hybrid cloud so some of the key takeaway messages that I want you to kind of take from this presentation today is that we are the new Microsoft we're open and we're outward looking ai is exploding but we think that it's going to empower every person and every organization to achieve more reg tech is the new FinTech so a lot of machine learning applications at the bank and AI applications are about helping the bank comply with requirements and we are helping the banks on this journey and we'll be with them every step of the way thank you very much cool any questions at all I guess what does the competitive landscape look like especially with Google and stuff that they do in the AI space yes so Google are really in the consciousness in the API space at the moment and this is a real bug bear for me this is why I need to go out there and tell this story because the truth is we have the most incredible AI platform I mean Google they have a deep learning toolkit called tensorflow and Google have very cleverly in their marketing equated tensorflow and artificial intelligence but actually there's so much more than that you probably saw in the list of them front office applications that most of them were not deep learning most of them are just standard classification and regression workloads in fact 80% of the work those at the bank do are just really simple kind of machine learning models and data scientists don't necessarily want to be using tensorflow they want to be using R they want to be using Python they want to be using scikit-learn Google do not have the flexibility that we have in machine learning and that is why this message is so powerful cool so anyway on the last speaker today so thank you all so much for being here today it's been it's been a great day and you know thank you for taking the time out to come and you know and to be here thank you very much [Applause]
