What is Machine Learning ?
ML is a way to use standard algorithms to derive predictive insights from data and make repeated decisions.
Phases of ML
Collecting dataLabeling dataTraining using chosen metrics and objectivesEvaluate a modelDeploy a model
Good data characteristic
Has coverageIs cleanIs complete
ML vs AI
ML is a type of Artificial intelligence.Logic vs Machine LearningNeural networks & Deep learningUse AI responsibly — responsible AI = successful AI
Why ML now ?
Increasing availability of dataIncreasing maturity and sophistication of ML algorithmsIncreasing power and availability of computing hardware and software
Labeling data
Label is the true answer for a given input.Regression vs ClassificationEvery example needs to have features and a label.Ways to label your data
Use a proxy labelBuild a labeling systemUse a labeling service
Modeling Training
Continuous training keeps models freshhttps://github.com/tensorflow
Formulating the ML problem
Choosing input featuresGet labelsChoose an objective
Modeling Evaluation
Test data (20%)Confusion matrices measure performance relative to expectations for classification.
ML Best Practices
ML involves experimentationstart simpleDon’t use your test data during experimentationDo pilot projects with end-users
Human Bias in ML
Decisions made as you do ML have real world impact for you and your customersUnconscious biases exist in dataFairness in ML
Discovering ML Use Cases
Simplifying rule-based systemsStreaming business processesUnderstanding unstructured dataPersonalizing experiences
Adds significant value to users Recommender systems
ML in Series
Data Strategy
ML is about repeated decisions
Design a system so that you will have more data next yearBreak down data silosTransition from data lakes to data warehousesLearn about your dataIntegrate pilots into your toolsRun ML models on real-time data to extract the most valueCollect more data
Data Governance
Data access must be balanced against securityThree goals for ML and Privacy
Identify sensitive dataProtect sensitive data by removing, masking or coarseningCreate public governance documentation Types of sensitive data
Specific columns in structured datasetsPatterned text, e.g., credit card numbersUnstructured data, like audio, video and imagesCombination of fields Common principles for establishing a policy framework
Establish a secure location for documentationExclude sensitive information from documentationDocument all sources and processesEstablish a process to review and enforce policies Build your ML team
Data engineersML engineersData analysts
Create a Culture of Innovation
Starts with a dedicated mindsetFocus on the user10X thinkingLaunch and iterateChange is inevitable
Course Certificate
{% pdf Coursera-Google-ML-for-Business.pdf %}
