Retail Workshop - Dania Meira & Rachel Berryman Who we are Rachel Data Science Consultant Bridging the gap between code and customers Teacher at a Data Science Bootcamp Dania Senior Data Scientist ML models for predictive analytics Volunteer at DSSG Berlin Teacher at a Data Science Bootcamp The AI Guild is the go-to community for Data Scientists, Data Engineers, Machine Learners, and Deep Learners accelerating the adoption of AI in Europe 2
Retail Workshop - Dania Meira & Rachel Berryman Agenda Part I: Theory & Understanding • What is AI? • AI & Image Recognition: What are the use cases? • Live example with code: Recognizing Fashion Images ~ Lunch Break ~ Part II: Workshop • Rolling out your own Image Recognition project: Hands-on strategy development with AI Project Canvas and AI Transformation Playbook 3
Retail Workshop - Dania Meira & Rachel Berryman What is AI? • Question - what is autonomous driving? AI ANI (Artificial Narrow Intelligence) Can handle 1 task well AGI (Artificial General Intelligence) Can cope with complicated tasks with many unknowns 4
Retail Workshop - Dania Meira & Rachel Berryman What comprises AI? AI is based on Machine Learning (ML). Data Science (DS) is a field that uses scientific and statistical methods to draw insights from data, and build predictive models from data. ML is one of the main tools of data science. 5
Retail Workshop - Dania Meira & Rachel Berryman What comprises AI? Artificial Intelligence (AI) Machine Learning (ML) Data Science (DS) A multidisciplinary field that uses scientific, computational and statistical methods to draw insights and build predictive models from data. Statistical techniques and algorithms that computer systems use to perform a specific task without explicit programming instructions, but instead processing data to detect patterns and inference. Deep Learning (DL) The ability of a machine to perform cognitive functions we associate with human minds, such as perceiving, reasoning, learning, interacting with the environment and problem solving. Type of ML methods based on artificial neural networks, algorithms inspired by the human brain, that learn from processing vast amounts of data. 6
Retail Workshop - Dania Meira & Rachel Berryman How Does ML Work? Machine learning uses training data to build a model of the relationship between inputs and outputs, which can be used to make predictions on unseen data. 7
Retail Workshop - Dania Meira & Rachel Berryman ML Workflow Practical ML has a distinct cyclical nature that demands constant iteration, tuning, and improvement. Define Goal Evaluate Model Prepare Data Collect Data Train Model Deploy Model Make Predictions Monitor Predictions Gather and Analyze Insights Define MVP 8
Retail Workshop - Dania Meira & Rachel Berryman ML Use Cases Narrow AI is good at focused tasks with a clear outcome. • It works best when there is a very large amount of training data. • It works well for specific cases where other methods fail - outlier detection, sparse matrix work. 10
Retail Workshop - Dania Meira & Rachel Berryman Examples of what ML can do Input Output Application Picture Are there human faces? Photo tagging Loan application Will they repay the loan? Loan approvals Ad + User information Will user click on ad? Targeted online ads Audio clip Transcript of audio clip Audio transcription English sentence German sentence Language translation Sensors from plane engine Is it about to fail? Preventive maintenance Recipe ingredients + customer reviews Will customer like the food? Food recommendation 12
Retail Workshop - Dania Meira & Rachel Berryman How does Image Recognition work? Under the hood, each pixel of an image is represented by a float value. The entire image is a matrix of these values (ex: a 20x20 pixel image is represented by a 20x20 matrix). 17
Retail Workshop - Dania Meira & Rachel Berryman How does Image Recognition work? Image converted into numeric values Image fed into system Numeric values and label fed into algorithm Algorithm learns pattern between numeric values and label Pattern “saved” via weights in the algorithm Image converted into numeric values Image fed into system Numeric values and label fed into algorithm “Saved” pattern between values and label used to predict label Predicted label returned Algorithm Training Prediction 18
Retail Workshop - Dania Meira & Rachel Berryman How does Image Recognition work? New datapoint: not yet seen by model Trains Model Model Training dataset: images & labels Model Generates Prediction “Nine” Fed in 19
Retail Workshop - Dania Meira & Rachel Berryman Deciding on a use case: Practical questions that need to be addressed sooner than later to enable development of a successful solution. 25 (1) Heart (2) Ingredients (4) Financial Requirements (3) Bringing to the Customer
Retail Workshop - Dania Meira & Rachel Berryman Example: Mobile engagement in-store A 2016 survey from Euclid Analytics found that 83% of shoppers use their mobile while shopping in-store. Respondents want to: • Access the store’s best deals/sales for maximum savings • Allow me to quickly access my loyalty program information in stores • Display inventory count for all products • Provide me with a map of the outlet • A more personalized experience based on my shopping habits and preferences • Allow me to request the attention of a store associate Winewoo is a Shazam-like experience that allows consumers to discover and purchase wine. Photo by OneSave/Day on Unsplash 26
Retail Workshop - Dania Meira & Rachel Berryman Value Proposition: Start from the “end” When we are done and have an amazing model, that produces great output, who is going to pull a lever differently, or push a button differently? Those are the opportunities worth tackling 28
Retail Workshop - Dania Meira & Rachel Berryman 29 Mobile Engagement in-store Provide customers with quality wine: an app to explore and discover brands while on the shop
Retail Workshop - Dania Meira & Rachel Berryman Data Audit Checklist • Which are the data sources? • If more than one source: are they joinable? • Is it trustworthy? • How much data? • What is the required data format? • Where does the data source exist? • Do we have access to it? • How rare is the event we want to predict? • Is the data labelled well? 30
Retail Workshop - Dania Meira & Rachel Berryman Output and Key Metric • What is the output like? ◦ A number, a percentage, a boolean answer (yes/no) ◦ How often it needs to be produced? ◦ Is there a trigger for the output to be generated/updated? • Solution acceptance criteria: the single key metric you’re evaluating on ◦ Accuracy, f1-score, precision, recall, minutes spent using the service, etc. ◦ Optional: sufficing metric. Accuracy has to exceed 95% (key metric) while taking no longer than 1s inference time (sufficing metric). 31
Retail Workshop - Dania Meira & Rachel Berryman 32 Mobile Engagement in-store 200.000 varieties of wine: label images and product information JPEG or PNG Computer Vision App development 98% accuracy rate on wine identification Provide customers with quality wine: an app to explore and discover brands while on the shop
Retail Workshop - Dania Meira & Rachel Berryman Integration • Where will the project be used? • How will the customer engage with your model? • Will you use a microservice, monolith, or predict on-the-fly during streaming? 33
Retail Workshop - Dania Meira & Rachel Berryman Key Stakeholders • Who are the Key Stakeholders? ◦ internal departments like legal, UX, management, or ◦ external stakeholders like contractors, owners, non-profit groups, ... 34
Retail Workshop - Dania Meira & Rachel Berryman Customers • Who is the final customer? ◦ Does the customer really care about an accuracy improvement from 99.2% to 99.3% or would faster inference time suit them better? 35
Retail Workshop - Dania Meira & Rachel Berryman 36 Mobile Engagement in-store Provide customers with quality wine: an app to explore and discover brands while on the shop 200.000 varieties of wine: label images and product information JPEG or PNG Computer Vision App development 98% accuracy rate Wine shoppers Wine stores Winemakers App (iOS and Android)
Retail Workshop - Dania Meira & Rachel Berryman Financial requirements Costs • Time or salary spent from your team members • Do you need to outsource labeling? • Will you incur compute costs? Revenue • Will the product be sold as a service or as a new feature category for users? • Will the project reduce internal costs through automating processes or support an innovation initiative? 37
Retail Workshop - Dania Meira & Rachel Berryman 38 Mobile Engagement in-store Provide customers with quality wine: an app to explore and discover brands while on the shop 200.000 varieties of wine: label images and product information JPEG or PNG Computer Vision App development 98% accuracy rate Wine stores Winemakers App (iOS and Android) Wine shoppers 3 FTEs for 6 months Data annotation Cloud Computing) Freemium model for winemakers Affiliate program for stores
meiradania
ch1aki
lycorptech_jp
freee
oracle4engineer
meihei3
kaonavi
maroon1st
kawaguti
tebiki
biwashi
yutakikai
tatsuo48
smashingmag
notwaldorf
chrislema
lauravandoore
zakiyama
zenorocha
chrisshort
brad_frost
philhawksworth
chriscoyier
bcantrill
sstephenson
