How To Sign Minnesota Banking Form

How to industry sign banking minnesota form computer

our next speaker is art Wang / Wang he's a distinguished engineer at US bank he's applying AI sorry he's applying ml and AI to solve problems across the organization art leads machine learning and artificial intelligence projects across corporate and retail banking in this role he identifies and evaluates business opportunities with strategic product potential market and technology trends key providers legal regulatory climate and profitability art also conducts research and development representing US bank to internal stakeholders customers and partners please join me in welcoming art to the stage thank you very much so the disclaimer the views and opinions expressed in here in this presentation are those of the authors me and do not necessarily reflect the policy official policy or position of any agency of US bank examples and analysis performed within the this presentation are only examples they should not be utilized in real world analytic products as they are treated only on very limited and dated open source information assumptions made within this analysis are not reflective any position of US bank furthermore nothing contained in this presentation should be considered investment advice okay let's see here so let's talk about transforming with artificial intelligence and how to unlock possibilities of working smarter with machine learning and AI Vienna my name is art Wang para Wong I'm distinguished engineer for artificial intelligence at u.s. bank let's talk about finding the wisdom on your AI journey so my AI journey your ki journey your company ai journey is all actually different because we're all trying to figure out how do we use AI for our business or our productivity and development and we're on different paths and different parts of the journey and it might not lead to different places so what I'd like to do in this presentation is to talk about what has worked and what hasn't worked from my personal experiences and perspective and hopefully some of it could resonate with you and maybe afterwards we can share stories about what works and what doesn't work because things are changing very fast and it's all learning process for all of us so first I'll start off with what is AI defining that landscape AI in banking the banking perspective prospects in AI with AI again this is my personal opinion and with it and then the vision for the future talk about some ambient computing resources and some ideas about that so it used to be like Oh 2020 is the future but we're already 2018 so I'm gonna we moved that 2028 so what would that be whether it'd be a utopia or a dystopia and the truth is like anything used for good can also be used for bad so I'm gonna restrict or limited to this presentation to what I see are beneficial and useful cases of machine learning artificial intelligence for Humanity and personally I'm very optimistic and I understand there's a lot of pessimism and a lot of attention around potential issues with AI Oh talk about that a little bit but in just 15 years like 52% of the S&P 500 has vanished and that rate has accelerated over time if you look at that chart on the left of the lifetime of a large corporate company especially operating globally and you know a lot of this can be attributed to technology and as we all know you know a lot of the big tech companies only ten years ago we they weren't in the S&P 500 so how did you go bankrupt two ways gradually and then suddenly this is a quote from Ernest Hemingway in the Sun Also Rises so with that that's that's why a lot of companies and corporations pay heavy attention to any development in technologies these days like blockchain when blockchain was a big hype in 2017 and early 2018 a lot of companies paid attention and didn't want to lose out on that and it's the same thing that's been happening with machine learning and AI in the past ten years so like what why is that why why do large companies disappear from the S&P 500 like that well the issue is that incumbents often lack ability to adapt and incorporate emerging technologies faster than those startups can gain distribution so that's why there's always startups coming up and providing value and growing faster before people realize it and then it's too late all right so let's define AI for a bit for the purpose of this discussion this is the chart that I'm going to frame AI around like you have this big data analytics which involves data in general and then AI can be a part of that where you're using that data for artificial intelligence purposes and artificial intelligence has been pursued through the machine learning means and machine learning can involve things and such as supervised learning reinforcement learning unsupervised learning and and deep learning can involve those things so and these definitions are there's it's not black and white there's gray area and people have different ways of defining it but for the purpose of this discussion we'll stick with this framework in order to have AI in a corporation you need to have that system to support it and when I'm talking about ecosystem I'm talking about not only the artificial intelligence and machine learning models but also the robotic process automation and orchestration infrastructure to support those models and the results of those data and processing capacity secure cloud enablement so to scale these days cloud has been a very useful tool that can be leveraged especially when we're deploying models that scale at large scale and finally strong customer authentication choices you know privacy is a huge concern these days and and the requirement for that is strong customer authentication so finally this is one summary of defining AI as a set of algorithms that leverage data to provide capabilities that are designed into user experiences and business processes to create business value again this is for the purposes of this discussion in how we're defining it so we need to come up with a system of technologies operating together to achieve this goal and then let's talk about strong versus weak AI there's a there's a lot of work into strong AI which is AI that can learn by itself that can achieve superhuman tasks be better than humans on the on the on the left there is a an example from jeopardy where humans are being beat in game in in trivia tasks and games of go but for the purpose of this discussion we're talking about we're not talking about emulating the human mind or performing any tasks human can like we're not trying to create artificial humans it for for the purpose of this discussion and and for the use of AI and businesses at the moment like there's there's efforts in in general AI and and strong AI but we're talking about weak and narrow AI so machine learning methods that are commonly used are supervised learning so as you know labeled data is very important there's a lot of startups springing up just to help companies get labeled data cloud services are now offering their label data labeling services and supervised learning for those not familiar is essentially you have inputs and you have targets or the labels for those inputs that are predefined so that you can train your model to predict those inputs and targets eventually such as home prices that depend on the inputs which are whatever the home details and then there's deep learning which has been a breakthrough driving this 10-year boom in AI the interest in AI which is learning from hierarchical relationships so in deep learning you have a deep neural network such as a convolutional neural network and each layer of that neural neural network learns a certain hierarchy of features in that data set that define that eventually define what you're trying to predict and this chart here shows the performance versus the amount of data deep learning has the opportunity to better utilize data to get higher performance and really these deep learning models have up to millions of parameters that are involved so it's essentially like its own memory system if you think about it alright let's talk about that AI landscape well let's I was supply meets demand so AI has become more useful and interesting because of access to data computational power and advanced models so these three developments in the past 10 years have made AI and machine learning more practical and I want to make a comment about access to data it used to be that you needed a lot of data to train AI models but actually more recently that's not the case anymore there's transfer learning where you don't need that much data to Train good AI models to do what you need to do transfer learning and fine tuning are two things that I'm very interested in and have been working a lot on so if you have questions about that talk to me afterwards but from the business perspective there's the profitability that comes with machine learning ai helping with Cost Reduction revenue gains and risk reduction and there's also the competition there's an arms race with companies in terms of trying to gather AI talent and AI models and AI capabilities and then there's regulations on reporting best execution anti money laundering that machine learning and AI could potentially be applied to or has already been applied to so with that maybe we should believe the hype that machine learning and AI is here to stay and will produce a lot of value this is a chart from Google Trends of the term machine learning but we shouldn't believe all the hype you have to be careful so here's an example of Chihuahua or a muffin it turns out that these advanced deep learning machine learning AI mottos can't tell the difference necessarily and maybe we can you can't tell difference so I can't tell the difference between an AI or Chihuahua and all of these if I give you one minute or less to go through and label them let's see what your accuracy is so science fiction might not have gotten it all right you know talking about cyborgs and artificial humans let's talk about in banking what have been the uses of machine learning AI like the actual current day statistics 65 percent in customer service 52% in back-office operations 42% for financial advisers 31% in fraud detection and 29% in risk management so this is actually where machine learning AI is being used again it's not some some phat science fiction AI general AI yet but it is a weak AI that is practical and yields benefits and it we can think about AI as superpowers for our customers where our customers a hero ai helps them with offense and defense when I'm talking about offensive superpowers I'm talking about using natural language understanding to have voice and conversation features for better customer experiences or biometric security and then sales and marketing can also use machine learning and I for instance in lead conversion and churn prevention shown here is a research study I published previously for telecommunications purposes where customers were modelled as images and we use the same ideas for deep convolutional neural networks applied to image recognition using that to recognize when customers might churn or they're unhappy with their service so we can better better address those issues all right so let's talk about the next chapter of where AI might go and again this is educated guesses at this point and and how will it be volved that's the exciting and that's what we're on a journey for and we're all on a journey building as well so I'm gonna restrict it to banking for now so digital transformation and the fourth Industrial Revolution is one way of looking at it where we have intelligent ambient agents proactively working on our behalf you know we don't have to actively generate data for the AI could just be collecting data and our surroundings and of course privacy is an issue that we have to be cognizant of but we have things that are always on always connected smart devices these natural interfaces that are IOT enabled contextual that provide contextual experiences and it's it could be proactive as well and I'm sort of speaking about vaguely but I think we all have an idea of what those are in our lives so we can have updated locations for customers we can switch default payment cards for businesses provide packing advice for a travel you know if it's going to rain provide advice on that on based on weather automatic check-ins provision room keys push room notices order room service shades down music become soft these are things that are happening all around us that we can help utilize and provide synergistic experiences holistic experiences for customers so self-driving cars is another exciting aspect well you know checking for your investment performance well automatically while you're in a car ordering go Suri's automatic updating your locations paying your bills using a payment micro pay to pass other drivers ecommerce with delivery to your trunk these are ideas where banking can be involved and if you're familiar with large tech companies the fine line between banking and tech and e-commerce it's it's all getting blurred and and that's that's the word like we're used to separating these ideas but is it really separate so we should keep an open mind on these topics biometric based authentication ari mentioned that advanced fraud detection world view forecast predictive analytics collections optimization computer vision awareness of the supply chain smart payment instruction repair IOT upstream insights in customer service voice based banking and chatbots is a is is an exciting area personalized website navigation and intelligent dashboards like customized dashboards are intelligent that's suitable for your needs natural language search like on-the-fly report creation with NLP is being developed and will be come a big thing I see that as a very promising Avenue leveraging data rich payloads from real-time payments so there's again there's a lot of data being generated in the billions in in the banking industry integrated receivables automatch incoming payments with invoices incomplete separately attacks or missing remittance info who's paying for what reduce manual effort with automation a lot of automation that could happen matching algorithms watch experience account receivable clerk learn to correct exceptions and prevent future errors post receivables faster improves straight through reconciliation of incoming payments to help post receivables faster decrease days sales outstanding and free cash flow I mean these are getting really specific into certain areas of banking customer service like natural language understanding I already mentioned that but you know imagine ability to ask what time is a branch nearby my home open and it'll automatically respond 6 p.m. so that you can get the answers to the questions you need faster so that you can be more efficient so you don't have to search through websites or get on the phone to call customer service and wait for minutes 2 hours reporting is a big thing there's a lot of human and auto-generated reports so dependencies and references between reports so how do you reconcile that other banking applications include the report of analyzing using OCR and NLP for these reports so a lot of reports are handwritten or their photographs or scanned and how do you automate all that into a honey digitize all of that that's a huge area that there's been progress but there's actually still limitations and OCR if you're interested in that talk to me I I can tell you all about my research with OCR so cross comparing reports in time and content so temporal analysis is a big thing we now have the ability with machine learning AI to look at how things vary with time as well and so these improvements should help with accuracy and compliance and you know the list goes on automated trading risk analysis management conversational banking farmers and prevention payments but what business problem are you solving for that's the question that you need to ask for yourselves right in here I've been throwing out a lot of things and there is a lot of things and and it is not exhaustive so so let's talk about key drivers of risk quickly and then we'll wrap up because time is running out accuracy and objective there's a model accurately forecast the outcome that you want and first you have to efine what outcome you want as many times the business lines that I deal with they're not able to articulate what does that value they need because we need as machine learning engineers I like I need to work with the business line to understand that business objective so in order to come up with an accuracy metric and there there are a lot of metrics out there and what metrics do you use f1 score micro macro average AUC you know precision recall all of that there's there's a bunch of accurate uh metrics out there which which is the correct one to use so you don't want to forecast and then have unintended consequence unintended consequences happening with the business line so I I also see that there there's a educational need even if you're not actively an engineer or working as a tech contributor like even if you're in the business or marketing there's a need to understand what again metrics are accuracy metrics and machine learning at a general level what it can and cannot do transparency is a big thing especially in banking you need to be able to explain why your algorithm algorithm reached a certain conclusion in human terms this there were some sessions about credit analysis and credit extending credit that's that's probably the most obvious one you know as your is your algorithm racially discriminatory in any way because that would be illegal so those are things to think about so there has to be clear documentation on the model the development the deployment the maintenance and and what what that model how is how it was developed what would what are those aspects so explained ability and transparency with the models is important and then that leads to bias and ethics just just because we can do certain things doesn't mean that we should do it so that's the thing that we have to be cognizant of as well so that the model contribute to unfair treatment of certain individuals you know these are things that we have to consider as well it's not just trying to get the most accurate model it's how do you get the most fair model all right so my time is up but you know do your homework avoid fear of missing out make it be educated on what you pursue and yeah I'll be available to talk afterwards if if you guys are interested okay I think that's it [Applause]