Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Taming Dragons
A Breakthrough Approach to AI for Business Leaders
Presented at the O'Reilly AI Conference in San Francisco Page 1 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 2 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Here’s what we’re going to do today:
1. Provide clarity 2. Seed your plan 3. Prepare for launch
Presented at the O'Reilly AI Conference in San Francisco Page 3 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Terminology matters
Big Data Machine Artificial Learning Intelligence
Fuel Engine Mechanical Power
Presented at the O'Reilly AI Conference in San Francisco Page 4 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Automation Analytics
next-generation skip-generation opportunities opportunities
Presented at the O'Reilly AI Conference in San Francisco Page 5 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
The primary purpose of automation is to perform a repetitive task previously performed by humans.
Presented at the O'Reilly AI Conference in San Francisco Page 6 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Automation is no longer limited to mechanical applications
Vision Speech Language Robotics
Current Visual inspection Voice call trees OCR Robotic arms
Cheaper Video security Chatbots Translation Robotic order fulfillment
Object ID in image Faster Video transcription Sentiment analysis Precision farming posts
Better Radiology Voice ID Author ID Hazardous conditions
Presented at the O'Reilly AI Conference in San Francisco Page 7 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Imagery
If you could identify something specific in photographs, illustrations and/or video, what could you do with that data?
EXAMPLE If you could identify a candidate’s facial expressions, you could better evaluate his/her interest in the job.
Presented at the O'Reilly AI Conference in San Francisco Page 8 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you could identify ______in ______, you could ______.
Hint: sources could include anything from social media to surveillance footage to product/service-related image capture, to name a few.
Presented at the O'Reilly AI Conference in San Francisco Page 9 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Audio
If you could recognize something specific in live and recorded audio sources, what could you do with that data?
EXAMPLE If you could recognize a customer’s emotional state in his/her tone of voice, you could customize our response.
Presented at the O'Reilly AI Conference in San Francisco Page 10 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you could recognize ______in ______, you could ______.
Hint: audio sources could include data you capture (customer service interactions) as well as data you consume (podcasts).
Presented at the O'Reilly AI Conference in San Francisco Page 11 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Language
If you could interpret something in the language you generate and consume, what could you do with that data?
EXAMPLE If you could interpret sentiment in the claims forms you process, you could minimize settlement pay-outs.
Presented at the O'Reilly AI Conference in San Francisco Page 12 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you could interpret ______in ______, you could ______.
Hint: consider focusing on one source, something your company, customers, competitors or outside content producers produce.
Presented at the O'Reilly AI Conference in San Francisco Page 13 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Automation Analytics
next-generation skip-generation opportunities opportunities
Presented at the O'Reilly AI Conference in San Francisco Page 14 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Sample Data
Algorithm
New Data StatisticalModel Model Prediction
Presented at the O'Reilly AI Conference in San Francisco Page 15 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Training Data
Algorithm
New Data MachineStatistical Learning Model Model Prediction
Presented at the O'Reilly AI Conference in San Francisco Page 16 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
The Humble Regression Model
Presented at the O'Reilly AI Conference in San Francisco Page 17 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Sales Regression Model
Presented at the O'Reilly AI Conference in San Francisco Page 18 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
What’s changed?
• Hyper-connected world makes 100% sampling feasible • Massive amounts of data on each individual customer • We can predict their behavior without even asking
Presented at the O'Reilly AI Conference in San Francisco Page 19 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 20 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 21 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 22 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
At the heart of every AI application is a machine learning model.
Presented at the O'Reilly AI Conference in San Francisco Page 23 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
1 Number 2 Probability 3 Category
Presented at the O'Reilly AI Conference in San Francisco Page 24 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Number Models
• Similar to “traditional” regression models, just much more data • Useful for predicting an exact quantity in advance • Can be any continuous unit of measure: $, time, units, degrees
Presented at the O'Reilly AI Conference in San Francisco Page 25 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Number Model Historical Forecasting Quarterly Sales
$ per Demographics, interests, purchase history, … quarter Customer 1 $557 Customer 2 $667 Customer 3 $223 Customer n $1009
New Data
Presented at the O'Reilly AI Conference in San Francisco Page 26 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Number Model Historical Customer Forecasting Lifetime Value
Lifetime Demographics, interests, purchase history, … spend Customer 1 $3556 Customer 2 $456 Customer 3 $1225 Customer n $3778
New customer
Presented at the O'Reilly AI Conference in San Francisco Page 27 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Number Model Historical Forecasting Maximum Price
Max Demographics, interests, purchase history, … Price Customer 1 $23.00 Customer 2 $14.00 Customer 3 $38.00 Customer n $26.00
New sale
Presented at the O'Reilly AI Conference in San Francisco Page 28 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Number Model Process Optimization
26 3
5 13 6 6 13 5
7 10
8 11
Presented at the O'Reilly AI Conference in San Francisco Page 29 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Number Model Process Optimization
26 3
5 13 6 6 13 5
7 10
8 11
Presented at the O'Reilly AI Conference in San Francisco Page 30 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Number Model Historical Process Optimization Segment Times
time, day, date, weather, events, ... Time Segment 1 3 Segment 2 5 Segment 3 7 Segment n 8
New Machine Learning Model request
Presented at the O'Reilly AI Conference in San Francisco Page 31 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Number Model Process Optimization
Labs
Presented at the O'Reilly AI Conference in San Francisco Page 32 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Forecasting
If we could predict precisely how much each guest will spend in our casino over the lifetime of our relationship, we could determine how much effort to put into maintaining the relationship with each guest.
If we could predict precisely the maximum amount each customer would spend, we could maximize ticket and package prices.
Presented at the O'Reilly AI Conference in San Francisco Page 33 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you could predict precisely how much/many ______at any given moment, you could ______.
Hint: identify places where you currently rely on estimates and averages to make decisions, then imagine instead that you have information about each individual.
Presented at the O'Reilly AI Conference in San Francisco Page 34 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Optimization
If we could identify the fastest way to get from one place to another at a given time, we could save both time and relationships.
If we could identify the fastest way to route patients through the clinic at any given time, we could maximize the number of patients seen.
Presented at the O'Reilly AI Conference in San Francisco Page 35 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you could identify the fastest way to ______, you could ______.
Hint: think about processes in your business that require moving something from point A to point B with multiple paths to choose from, then imagine being able to predict the fastest path at any given time.
Presented at the O'Reilly AI Conference in San Francisco Page 36 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Probability Models
• Commonly referred to as recommenders or recommendation engines • Predict the probability of a given event occurring (e.g. purchase) • Useful when you have two long lists and need to make connections
Presented at the O'Reilly AI Conference in San Francisco Page 37 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
People Movies
?
Presented at the O'Reilly AI Conference in San Francisco Page 38 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Probability Model Personalization
John
Presented at the O'Reilly AI Conference in San Francisco Page 39 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Historical Probability Model Recommendation Personalization Outcomes
Purchased Customer info + movie info ? Recommendation 1 yes Recommendation 2 no Recommendation 3 no Recommendation n yes
New login Machine Learning Model
Presented at the O'Reilly AI Conference in San Francisco Page 40 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Probability Model Personalization
Products Articles Bees Birds Users Customers
Presented at the O'Reilly AI Conference in San Francisco Page 41 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Probability Model Prioritization
Leads Customers
Presented at the O'Reilly AI Conference in San Francisco Page 42 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Probability Model Advanced Expert Systems Response Recommenders
Responses Arguments Claims Issues
Presented at the O'Reilly AI Conference in San Francisco Page 43 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Probability Model Advanced Expert Systems
Guided Selling Training Coach
Next action Next exercise Sale status Goal & status
Presented at the O'Reilly AI Conference in San Francisco Page 44 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Personalization
If we knew which movies were most likely to be viewed by each individual user, we could maximize streaming revenue.
If we knew which articles were most likely to be read by each individual reader, we could increase reader satisfaction.
Presented at the O'Reilly AI Conference in San Francisco Page 45 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you knew which ______were most likely to ______by each individual ______, you could ______.
Hint: identify places where you use segments or groupings, and imagine being able to engage each individual directly.
Presented at the O'Reilly AI Conference in San Francisco Page 46 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Prioritization
If we could rank sales leads by their likelihood to convert we could focus on the most lucrative deals.
Presented at the O'Reilly AI Conference in San Francisco Page 47 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you could rank ______by their likelihood to ______, you could ______.
Hint: think of areas in your business where it’s difficult to decide what’s most important, or where to focus resources, and imagine eliminating that guesswork.
Presented at the O'Reilly AI Conference in San Francisco Page 48 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Advanced Expert Systems
If we could match incoming calls with proven resolution responses, we could improve customer service.
If we could match client requirements with ideal solution recommendations, we could reduce RFP response times.
Presented at the O'Reilly AI Conference in San Francisco Page 49 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Advanced Expert Systems
If we could match incoming calls with proven resolution responses, we could improve customer service.
If we could match client requirements with ideal solution recommendations, we could reduce RFP response times.
Presented at the O'Reilly AI Conference in San Francisco Page 50 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Category Models
• Commonly referred to as classifiers • Used to classify elements of a population into one of two or more categories • Binomial (two category) models used for “yes/no” type decisions • Multinomial (three more categories) used for selection among options
Presented at the O'Reilly AI Conference in San Francisco Page 51 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 52 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Category Model Email Binomial Detection Examples
Misspellings, grammatical errors, all caps, domain… Spam? Email 1 yes Email 2 no 100% Email 3 yes Spam Email n no 80%
Not spam
Uses rule YOU set Model calculates to decide the final the probability output that it is spam
New email Machine Learning Model 0% If > 80% then spam 85% If <80% then not spam
Presented at the O'Reilly AI Conference in San Francisco Page 53 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Category Model Previous Patient Multinomial Detection Diagnosis
Symptoms, medical history,… Diagnosis Patient 1 Eczema Patient 2 Allergy Patient 3 Eczema Patient n Psoriasis
Model calculates the Selects the probability for each most likely possibility
New Machine Learning Model Patient Eczema 86% Allergy 8% Psoriasis 6%
Presented at the O'Reilly AI Conference in San Francisco Page 54 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Category Model Historical Loan Binomial Prediction Outcomes
Customer + loan details, … Paid? Loan 1 yes Loan 2 no Loan 3 yes Loan n no
New Machine Learning Model application If > 90% then approve 85% If <10% then decline
Presented at the O'Reilly AI Conference in San Francisco Page 55 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Binomial Detection
If we could detect deceptive, distracting, unproductive, annoying, aggravating communication,fromhistorical communication data we could the good stuff.
Presented at the O'Reilly AI Conference in San Francisco Page 56 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you could detect ______from ______data, you could ______.
Hint: look for places where you make pass/fail type decisions in your current business processes.
Presented at the O'Reilly AI Conference in San Francisco Page 57 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Framing the Opportunity: Binomial Prediction
If we could predict the likelihood that a loan applicant will default on his/her loan from historical loan data, we could let ‘em have it anyway.
Presented at the O'Reilly AI Conference in San Francisco Page 58 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Your turn…
If you could predict ______from ______data, you could ______.
Hint: look for places where you make yes/no type decisions in your current business processes.
Presented at the O'Reilly AI Conference in San Francisco Page 59 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 60 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Map the opportunities you identified to your company’s functional areas…
Presented at the O'Reilly AI Conference in San Francisco Page 61 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
You are here, sort of
Presented at the O'Reilly AI Conference in San Francisco Page 62 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Coordinated Transformation
Presented at the O'Reilly AI Conference in San Francisco Page 63 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
• Determine potential ROI for each application • Conduct feasibility assessments • Estimate model development times • Create staged roadmap
Presented at the O'Reilly AI Conference in San Francisco Page 64 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
• Select data science platform • Create end-state architecture • Determine implementation plan synchronized with ML roadmap
Presented at the O'Reilly AI Conference in San Francisco Page 65 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
•Unprecedented teamwork •Data-driven decision making •Culture of experimentation •C-suite commitment
Presented at the O'Reilly AI Conference in San Francisco Page 66 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
• Upskilling plan for existing staff • Reskilling forecast & alternatives for displaced workers • Forecast, acquire and grow data science talent from within
Presented at the O'Reilly AI Conference in San Francisco Page 67 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Scarcity of data science talent?
Open jobs in US for Data Scientists in US Data Scientists in US “Data Scientist” “actively” seeking “quietly” seeking 19k 15k 22k
Sources: LinkedIn 8/7/18
Presented at the O'Reilly AI Conference in San Francisco Page 68 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 69 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 70 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 71 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 72 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
3-6 Months 1-2 Months 2-3 Months
Presented at the O'Reilly AI Conference in San Francisco Page 73 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
PMO
• Need to share learnings, encourage teamwork • Can’t centralize because too domain dependent • Moving toward multidisciplinary teams
Presented at the O'Reilly AI Conference in San Francisco Page 74 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
We’re a go!
Presented at the O'Reilly AI Conference in San Francisco Page 75 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
A glimpse into your future
Once you get a win like this, generating real revenue and increasing profitability, it really opens everyone’s eyes to where else could we use machine learning technologies to enhance the business.
Ken O'Brien, CIO - RR Donnelley
Presented at the O'Reilly AI Conference in San Francisco Page 76 September 5, 2018 Taming Dragons: A breakthrough approach to AI for business leaders milkandhoney.ai copyright: milk+honey [email protected]
Q&A
milkandhoney.ai [email protected]
Presented at the O'Reilly AI Conference in San Francisco Page 77 September 5, 2018
Automation Exercises
Framing the Opportunity: Imagery
If I could identify ______in ______, I could ______.
Hint: sources could include anything from social media to surveillance footage to product/service-related image capture, to name a few.
Framing the Opportunity: Audio If I could recognize ______in ______, I could ______.
Hint: audio sources could include data you capture (customer service interactions) as well as data you consume (podcasts).
Framing the Opportunity: Language If I could interpret ______in ______, I could ______.
Hint: consider focusing on one source, something your company, customers, competitors or outside content producers produce.
Number Model Exercises
Forecasting If I could predict precisely how much/many ______at any given moment, I could ______.
Hint: identify places where you currently rely on estimates and averages to make decisions, then imagine that instead you have information about each individual.
Optimization If I could identify the fastest way to ______, I could ______.
Hint: think about processes in your business that require moving something from point A to point B with multiple paths to choose from, then imagine being able to predict the fastest path at any given time.
Probability Model Exercises
Personalization If I knew which ______were most likely to ______by each individual ______, I could ______.
Hint: identify places where you use segments or groupings, and imagine being able to engage each individual directly.
Prioritization If I could rank ______by their likelihood to ______, I could ______.
Hint: think of areas in your business where it’s difficult to decide what’s most important, or where to focus resources, and imagine eliminating that guesswork.
Advanced Expert Systems If I could match ______with ______, I could ______.
Hint: think of opportunities to more effectively and immediately leverage your knowledge base to expedite and improve process.
Category Model Exercises
Binomial Detection If I could detect ______from ______,
I could ______.
Hint: look for places where you make pass/fail type decisions in your current business processes.
Binomial Prediction If I could predict ______from ______, I could ______.
Hint: look for places where you make yes/no type decisions in your business today.
Additional Application Exercise Examples
Automation I would love to be able to recognize faces in images from surveillance cameras in order to improve security at our factories.
I would love to be able to interpret customer emotional state in audio from chatbot conversations in order to improve customer relations.
I would love to be able to identify prospective customers in text from blog posts in order to increase sales.
Number Models If I could predict precisely how long terminally ill patients have to live at any given time, I could better provide them with the care they need.
If I could predict precisely how much electricity was going to be demanded at any given moment I could reduce costs by optimizing the power grid.
If I could predict precisely how many more months each engineer was going to stay at any given time I could start sourcing replacements sooner.
Probability Models I would love to know which candidates are the most likely to be successful for each individual job opening in order to minimize unnecessary interviews.
I would love to be able to rank outstanding repair orders by level of potential damage in order to minimize risk of injury.
I would love to be able to capture and share product configuration expertise with operators in our call center in order to maximize productivity.
Category Models I would love to be able to predict whether middle school students are at low, medium or high risk of dropping out from school record data in order to focus counseling resources appropriately.
I would love to be able to predict which suicide hotline callers will actually attempt suicide from their call history and current emotional disposition data in order to initiate actions to prevent it.
I would love to be able to detect children that have been victims of abuse from school and health record data in order to provide them with early mental health care.
Data Science Candidate Profile ZR_520_CAND
Experience in role Candidate Overview Development Role History low high Business Solution Model Model Model Continuous Current Title: Director of Data Science Objective Design Creation Evaluation Deployment Improvement Work Experience (yrs): 15 Machine Learning Degrees held: Master of Science;Bachelor of Science Data & Platform
Tool & Technology Skills Ama Python;Ja Windows;U Programming Math Computation zon va;C/C+ Unix Shell MATLAB Tableau Development Platforms nix/Linux OS's languages Scripting languages Tools Visualization Tools Web + TRUE Amazon Web Services TRUE Unix/Linux TRUE Python TRUE Unix Shell TRUE MATLAB TRUE Tableau TRUE IBM Cloud TRUE Windows FALSE R FALSE awk FALSE Octave FALSE QlikView TRUE Google Cloud Platform FALSE MacOS TRUE Java FALSE Perl FALSE Mathematica FALSE TIBCO Spotfire ### Microsoft Azure TRUE C/C++ FALSE Julia FALSE SAS JMP ### On-premise/private Microsoft IBM SPSS Statistical Analysis IBM Business Intelligence DataRob ML Productivity Jupyter Notebooks;Tensorflow;Hadoop/Hive/Pig;Spark/Mllib;scikit-learn SQL Python Tools & Libraries Data Mining Tools Statistics Tools Cognos Tools ot Tools Databases Server;Post TRUE Jupyter Notebooks FALSE MS Excel Data Mining FALSE SAS Base FALSE SAS FALSE RapidMiner FALSE Oracle SQL TRUE Tensorflow FALSE MS SQL Server Mining TRUE IBM SPSS Statistics FALSE SAP Business Objects FALSE Cloudera TRUE Microsoft SQL Server ### PyTorch FALSE SAS Enterprise Miner FALSE Oracle R Enterprise FALSE Predictive Analytics FALSE Amazon ML TRUE PostgreSQL TRUE Hadoop/Hive/Pig FALSE Oracle Data Mining FALSE Stan FALSE KNIME FALSE Azure ML Studio FALSE Other SQL TRUE Spark/Mllib FALSE IBM Watson FALSE Minitab TRUE IBM Cognos FALSE Orange FALSE MongoDB ### Flume FALSE Statistica FALSE Angoss FALSE IBM SPSS Modeler FALSE Amazon DynamoDB TRUE scikit-learn TRUE DataRobot FALSE Couchbase FALSE Salford Predictive Modeler FALSE Other NoSQL
Machine Learning Application Experience Level of expertise low high
Functional Automation Numeric Prediction Recommender Classifier Business Area image audio/ natural robotics/ forecasting/ process personalization option expert system strategy/ condition outcome processing speech language industrial projection optimization prioritization sequential detection prediction processing processing control actions Accounting & finance Administration & operations Customer service support Human resources Legal
Logistics & supply chain Marketing & promotion Product/ service development Production/ manufacturing Research & development Sales
Other
9/4/2018 Data Science Candidate Profile ZR_520_CAND
Training & Education Summary
Linear algebra Accredited degree program Basic machine learning Accredited degree program Natural language processing Bootcamp or certificate program Calculus Accredited degree program Advanced machine learning Accredited degree program Image processing/computer vision n/a Basic statistics Accredited degree program Deep learning Bootcamp or certificate program Voice & audio processing n/a Advanced statistics Accredited degree program Reinforcement Learning Employer provided training Other: n/a Data visualization Individual online course Data privacy/governance n/a Other: n/a
Machine Learning Experience Level of expertise low high Algorithm family Numeric Recommendation Classification Prediction item action sequential image audio/ speech natural data/multi Association Rule actions language
Bayesian
Clustering
Collaborative Filtering
Content Filtering
Hybrid Filtering
Decision Tree
Deep Learning
Dimensionality Reduction Ensemble
Instance-based (kNN)
Matrix Factorization
Neural Network
Regression
Reinforcement Learning
Support Vector Machine
Other
Awards & Accomplishments
9/4/2018 Data Science Candidate Profile ZR_520_CAND
Data Science Project Portfolio Project Name Project Project ML Functional ML Algorithm Level of Project ML Model Model Model Model Data Data Data Data Model Data dumm2 duration type application Business Area implementation family contribution scoping application development evaluation deployment evolution platform integration platform visualization automation governance design architecture plan development BOW Topic 1-3 months class Classifier - Marketing & Classification - Regression Team n n n Classification condition promotion natural language member Mushroom PCA & 1-3 months class Automation - Marketing & Classification - Dimensionality Team n n n clustering audio/ promotion image Reduction member Final Project 1-3 months class Automation - Product/ service Classification - Deep Learning Team n n n n image development image member email classification 1-3 months class Classifier - Administration Classification - Bayesian Team n n condition & operations natural language member spam detector 1-3 months class Classifier - Administration Classification - Bayesian Team n n n condition & operations natural language member product recommender 1-3 months class Recommend Marketing & Recommendation - Regression Team n n n er - promotion item member tweet clustering by 1-3 months class Classifier - Marketing & Classification - Instance- Team n n n source condition promotion natural language based (kNN) member joins, book EDA, 1-3 months class Automation - Administration Classification - Bayesian Team n n n synonyms natural & operations natural language member shortest path wikipedia 1-3 months class Recommend Administration Recommendation - Decision Tree Team n n n n er - & operations item member pagerank 1-3 months class Recommend Marketing & Recommendation - Decision Tree Team n n er - promotion item member wordcount & sort with 1-3 months class Classifier - Customer Classification - Instance- Team n n n spark condition service support natural language based (kNN) member binary classification 1-3 months class Classifier - Customer Classification - Regression Team n n n outcome service support data/multi member CTR with logistic 1-3 months class Recommend Logistics & Recommendation - Regression Team n n n regression er - supply chain item member decision trees 1-3 months class Numeric Marketing & Numeric prediction Decision Tree Team n n n n Prediction - promotion member adder in tensorflow 1-3 months class Numeric Sales Numeric prediction Deep Learning Team n n n Prediction - member word similarity 1-3 months class Automation - Marketing & Classification - Deep Learning Team n n language model natural promotion natural language member sentence probability 1-3 months class Automation - Marketing & Classification - Deep Learning Team n n n natural promotion natural language member POS tagging, pipe 1-3 months class Automation - Administration Classification - Deep Learning Team n n n cutting word id natural & operations natural language member fake news detector 1-3 months class Classifier - Administration Classification - Deep Learning Team lead n n n n n n n condition & operations natural language election results tweet 1-3 months class Classifier - Marketing & Classification - Regression Team n n n n n analysis condition promotion natural language member Capstone - crisis 3-6 months class Automation - Product/ service Classification - Bayesian Team lead n n n n n navigator natural development natural language Digit Classification 1-3 months class Automation - Administration Classification - Instance- Team n n image & operations image based (kNN) member
9/4/2018 Data Science Job Opening Summary Acme Tech - Data Science Manager - ZR_36_JOB
Job Opening Overview Job Opening Scope Business Solution Model Model Model Continuous Team Role: Manager Objective Design Creation Evaluation Deployment Improvement
Experience Requirement: 5+ years Machine Learning Education Requirement: Masters Data & Platform
Tool & Technology Skills Ama Programming Math Computation zon Unix/Linux Python;R Unix Shell MATLAB Tableau Development Platforms OS's Languages Scripting Languages Tools Visualization Tools Web TRUE Amazon Web Services TRUE Unix/Linux TRUE Python TRUE Unix Shell TRUE MATLAB TRUE Tableau ### IBM Cloud FALSE Windows TRUE R FALSE awk FALSE Octave FALSE QlikView ### Google Cloud Platform FALSE MacOS FALSE Java FALSE Perl FALSE Mathematica FALSE TIBCO Spotfire ### Microsoft Azure FALSE C/C++ FALSE Julia FALSE SAS JMP ### On-premise/private SAS Microsoft Statistical Analysis IBM Business Intelligence DataRob ML Productivity Jupyter Notebooks;Tensorflow;Hadoop/Hive/Pig;Spark/Mllib Base;IBM SQL Python Tools & Libraries Data Mining Tools Tools Cognos Tools ot Tools Databases SPSS Server;Post TRUE Jupyter Notebooks FALSE MS Excel Data Mining TRUE SAS Base FALSE SAS FALSE RapidMiner FALSE Oracle SQL TRUE Tensorflow FALSE MS SQL Server Mining TRUE IBM SPSS Statistics FALSE SAP Business Objects FALSE Cloudera TRUE Microsoft SQL Server ### PyTorch FALSE SAS Enterprise Miner FALSE Oracle R Enterprise TRUE Predictive Analytics FALSE Amazon ML TRUE PostgreSQL TRUE Hadoop/Hive/Pig FALSE Oracle Data Mining FALSE Stan FALSE KNIME FALSE Azure ML Studio FALSE Other SQL TRUE Spark/Mllib FALSE IBM Watson FALSE Minitab TRUE IBM Cognos FALSE Orange FALSE MongoDB ### Flume FALSE Statistica FALSE Angoss FALSE IBM SPSS Modeler FALSE Amazon DynamoDB ### scikit-learn TRUE DataRobot FALSE Couchbase FALSE Salford Predictive Modeler FALSE Other NoSQL
Project Details
Functional Automation Numeric Prediction Recommender Classifier Business Area image audio/ natural robotics/ forecasting/ process personalization option expert system strategy/ condition outcome processing speech language industrial projection optimization prioritization sequential detection prediction processing processing control actions Accounting & finance Administration & operations Customer service support Human resources Legal
Logistics & supply chain Marketing & promotion Product/ service development Production/ manufacturing Research & development Sales
Other
9/4/2018 Job Opening Candidate Match Summary Acme Tech - Data Science Manager - ZR_36_JOB
Experience in role Candidate Overview: ZR_520_CAND Development Role Comparison low high Business Solution Model Model Model Continuous Current Title: Director of Data Science Objective Design Creation Evaluation Deployment Improvement
Work Experience (yrs): 15 Machine Learning Degrees held: Master of Science;Bachelor of Science Data & Platform
Tool & Technology Skills Match Assessment Ama Python;Ja Windows;U Programming Math Computation zon va;C/C+ Unix Shell MATLAB Tableau Development Platforms nix/Linux OS's Languages Scripting Languages Tools Visualization Tools Web + TRUE Amazon Web Services TRUE Unix/Linux TRUE Python TRUE Unix Shell TRUE MATLAB TRUE Tableau TRUE IBM Cloud TRUE Windows FALSE R FALSE awk FALSE Octave FALSE QlikView TRUE Google Cloud Platform FALSE MacOS TRUE Java FALSE Perl FALSE Mathematica FALSE TIBCO Spotfire ### Microsoft Azure TRUE C/C++ FALSE Julia FALSE SAS JMP ### On-premise/private Microsoft IBM SPSS Statistical Analysis IBM Business Intelligence DataRob ML Productivity Jupyter Notebooks;Tensorflow;Hadoop/Hive/Pig;Spark/Mllib;scikit-learn SQL Python Tools & Libraries Data Mining Tools Statistics Tools Cognos Tools ot Tools Databases Server;Post TRUE Jupyter Notebooks FALSE MS Excel Data Mining FALSE SAS Base FALSE SAS FALSE RapidMiner FALSE Oracle SQL TRUE Tensorflow FALSE MS SQL Server Mining TRUE IBM SPSS Statistics FALSE SAP Business Objects FALSE Cloudera TRUE Microsoft SQL Server ### PyTorch FALSE SAS Enterprise Miner FALSE Oracle R Enterprise FALSE Predictive Analytics FALSE Amazon ML TRUE PostgreSQL TRUE Hadoop/Hive/Pig FALSE Oracle Data Mining FALSE Stan FALSE KNIME FALSE Azure ML Studio FALSE Other SQL TRUE Spark/Mllib FALSE IBM Watson FALSE Minitab TRUE IBM Cognos FALSE Orange FALSE MongoDB ### Flume FALSE Statistica FALSE Angoss FALSE IBM SPSS Modeler FALSE Amazon DynamoDB TRUE scikit-learn TRUE DataRobot FALSE Couchbase FALSE Salford Predictive Modeler FALSE Other NoSQL
Application Experience Comparison Level of expertise low high
Functional Automation Numeric Prediction Recommender Classifier Business Area image audio/ natural robotics/ forecasting/ process personalization option expert system strategy/ condition outcome processing speech language industrial projection optimization prioritization sequential detection prediction processing processing control actions Accounting & finance Administration & operations Customer service support Human resources Legal
Logistics & supply chain Marketing & promotion Product/ service development Production/ manufacturing Research & development Sales
Other/ unspecified
9/4/2018 [email protected] www.milkandhoney.ai