Visa Inc. · NYSE:V
Visa — Technical Presentation (May 2018)
34 slides from Visa's technical presentation, May 24, 2018
Slide 1: Machine Learning in the Payments Industry
Machine Learning in the Payments Industry 24 May 2018 John Steensen
Slide 2: Disclaimer: Forward-Looking Statements
Disclaimer Forward-Looking Statements The materials, presentations and discussions during this meeting contain forward-looking statements within the meaning of the U.S. Private Securities Litigation Reform Act of 1995. These statements can be identified by the terms “will,” new,” "continue," "could,
Slide 3: Disclaimer: Notice
Disclaimer Notice The information, recommendations or “best practices” contained herein are provided "AS IS" and intended for informational purposes only and should not be relied upon for operational, marketing, legal, technical, tax, financial or other advice. When implementing any new strategy or
Slide 4: Agenda
Agenda A Brief Introduction to Machine Learning Machine Learning Applications in the Payment Industry Decision Making with Machine Learning Threats from Machine Learning-based Attacks Managing and Monitoring of Machine Learning Questions and Answers 4 Machine Learning in the Payments Ind
Slide 5: A Brief Introduction to Machine Learning
A Brief Introduction to Machine Learning ©2017 Visa. All rights reserved. Visa Public 5 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 6: What is Machine Learning?
What is Machine Learning? Machine Learning is a part of computer science that "gives computers the ability to learn without being explicitly programmed“. Arthur Lee Samuel, 1959 Data from Predictions Machine about the the Past Learning Future Yesterday Today Tomorrow 6 Machine Learning in the Paym
Slide 7: What is Learning?
What is Learning? Learning is the process of acquiring a body of knowledge, usually with the intent of performing some actions based upon that knowledge. By identifying the most influential cause-and-effect relationships from the past, a machine can learn to make accurate predictions about the fut
Slide 8: Machine Learning is Becoming Widely Used
Machine Learning is Becoming Widely Used “In recent years many successful machine learning applications have been developed, ranging from data-mining programs that learn to detect fraudulent credit card transactions, to information-filtering systems that learn users’ reading preferences, to autonomo
Slide 9: Machine Learning Applications are Models
Machine Learning Applications are Models Models are simplified approximations to the real world. Most models are built to support a specific activity within a specific environment. 9 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 10: Models are Built upon Data
Models are Built upon Data In machine learning the data columns are referred to as “features”. Getting the data ready to use is called “feature engineering”. Messy, incomplete, or Complete, well- Used in Machine 80% 20% poorly structured data structured data Learning Although payment systems gen
Slide 11: Implementing Machine Learning
Implementing Machine Learning Machine learning uses many different algorithms. Algorithms are just step-by-step processes to calculate a result. Data is needed by these algorithms to “train” them and to “test” them. Some algorithms need data that is expertly prepared to exacting specificatio
Slide 12: Types of Machine Learning Algorithms
Types of Machine Learning Algorithms Supervised Learning Training Data Unsupervised Learning Reinforcement Learning Feedback Algorithm Test Data Hypothesis Performance Each of these have specific problem areas that they are best suited for. Part of the data scientist’s job is to know which
Slide 13: Supervised Learning
Supervised Learning Step 1: Identify the outcome variable which is to be predicted Step 2: Identify the set of input variables called predictors Step 3: Generate a mathematical function that maps the input variables to outcome variable. This mapping process is called “training” and is repeated
Slide 14: Unsupervised Learning
Unsupervised Learning Using this algorithm, there is not any specific target or outcome variable to predict or estimate. It is used for clustering a population into different groups. For example, it is widely used for segmenting customers into different groups for specific intervention or market
Slide 15: Reinforcement Learning
Reinforcement Learning Using this algorithm, the algorithm is trained to make specific decisions. The algorithm is exposed to an environment where it trains itself repetitively using trial and error. This algorithm learns from past experience and tries to capture the best possible knowledge to
Slide 16: Machine Learning Areas of Application in the Payment Industry
Machine Learning Areas of Application in the Payment Industry ©2017 Visa. All rights reserved. Visa Public 16 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 17: A Typical Payment Flow
A Typical Payment Flow Customer Customer Initiates Transaction Pays the Merchant Issuing Bank Checkout through Customer Transaction is Completed Credit or On completion of Debit Transaction the Card Accounts Issuing Bank Transfers Payment Merchant Money to Merchant’s Gateway gives Checkout Page Issu
Slide 18: Payment Flow Metrics Determination
Payment Flow Metrics Determination Is initiating the transaction? WHO Was the recipient of the transaction? WHEN Is the transaction initiated? WHERE Is the transaction initiated from? Was the transaction initiated? HOW Often were transactions initiated? Was the transaction trying to accompli
Slide 19: What are Machine Learning's Unique Advantages?
What are Machine Learning’s Unique Advantages? Based upon real data, not human interpretations of data. Can incorporate massive amounts of data. Can typically outperform business rule-based processing approaches. Algorithms can be combined into something called “ensembles” allowing for more
Slide 20: Fighting Fraud
Fighting Fraud Machine learning fraud fighting strengths: Facilitating real-time decision-making power that allows for evaluation of huge numbers of transactions Improving accuracy of decisions resulting in detection of subtle or non-intuitive patterns to help identify fraud and avoid erroneous
Slide 21: Decision Making with Machine Learning
Decision Making with Machine Learning ©2017 Visa. All rights reserved. Visa Public 21 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 22: Machine Learning "Accuracy" is a Business Decision
Machine Learning “Accuracy” is a Business Decision Machine learning is about understanding the past to make predictions about the future ~ and some of those predictions will be wrong. Inaccurate rules logic may result in rejections of valid transactions or “false negatives” leading to revenue lo
Slide 23: Tuning Your Machine Learning Algorithms
Tuning Your Machine Learning Algorithms Machine learning algorithms must be accurate: Algorithms tied too tightly to past behavior are “over-training”. Algorithms tied too loosely to past behavior are “under-training”. Determine the acceptable business balance between “false positives” (appr
Slide 24: Threats from Machine Learning-based Attacks
Threats from Machine Learning-based Attacks ©2017 Visa. All rights reserved. Visa Public 24 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 25: Machine Learning is a Tool for the Threat Actor
Machine Learning is a Tool for the Threat Actor Machine learning can allow threat actors to first gain a better foothold in the payment ecosystem to accomplish their mission. The threat actor may utilize machine learning to help in disguising a bad payment transaction as a good one. Remember t
Slide 26: Threat Techniques Enhanced by Machine Learning
Threat Techniques Enhanced by Machine Learning Cloaking or “Wolf among the Sheep” “Outrun the Cops” “Wait out the Cops” “Friendly Microbe” 26 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 27: Managing and Monitoring of Machine Learning
Managing and Monitoring of Machine Learning ©2017 Visa. All rights reserved. Visa Public 27 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 28: Managing Machine Learning
Managing Machine Learning Machine learning applications are models that are simplified approximations to the real world. If the nature of the specific activity, or the environment within which that activity takes place changes, you will need assurance that the model is still relevant. If your
Slide 29: Monitoring Machine Learning
Monitoring Machine Learning How do you know that a machine learning application is becoming outdated? Monitor for a change in effectiveness of the machine learning application. For example, are you getting more of a specific result than you expect? If possible test the machine learning applica
Slide 30: Summary
Summary ©2017 Visa. All rights reserved. Visa Public 30 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 31: Opportunities for Action
Opportunities for Action Machine learning has an important role to play in supporting a robust and safe payment environment. Acquire trained professionals or develop staff through training and certification. Train Internal Audit on risks associated with machine learning. Threat actors will t
Slide 32: Question & Answer
Question & Answer ©2017 Visa. All rights reserved. Visa Public 32 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 33: Visa Data Security Resources
Visa Data Security Resources ©2017 Visa. All rights reserved. Visa Public 33 Machine Learning in the Payments Industry – 24 May 2018 Visa Public
Slide 34: Visa Data Security Resources
Visa Data Security Resources Visa Data Security Website www.visa.com/cisp • Alerts, Bulletins • Best Practices, White Papers • Past Webinars Visa Global Registry of Service Providers www.visa.com/onthelist • List of registered, PCI DSS validated third party agents PCI Resources for Small Merchants h