WEEK 6 — DIGITAL TRANSFORMATION

Week 6: AI/ML/DL

Agenda

  1. 1Review last week's material
  2. 2Introduce AI
  3. 3Different types of AI
  4. 4Machine learning
  5. 5ML algorithms
  6. 6Different paradigms of learning
  7. 7Conclusion
  8. 8Q&A
Dr. Davood WadiDigital Transformation
Two data documents on the left feed via a red arrow into a framed screen holding three nested realms: Artificial Intelligence on the outside, Machine Learning in the middle, and a red Deep Learning neural network at the coreArtificial IntelligenceMachine LearningDeep Learning

CHAPTER 01

Artificial Intelligence

CLASS DISCUSSION

  1. 1When you hear ‘Artificial Intelligence,’ what’s the first example that comes to mind?
  2. 2Have you interacted with AI today without realizing it? Let’s brainstorm hidden AI in daily life.
  3. 3How is a ‘smart’ app like ChatGPT different from a basic calculator? What makes it ‘AI’?
Two people in profile face each other, each with a speech bubble, in conversation

CLASS DISCUSSION

  1. 1If you asked ChatGPT for advice, would you trust it more than a human? Why or why not?
  2. 2What kind of topics do you post on your social media?
  3. 3If you trained an AI using only your social media posts, what might it ‘learn’ about people?
Two people in profile face each other, each with a speech bubble, in conversation

CORE CONCEPT

Artificial Intelligence

Artificial Intelligence: Intelligence shown by machines.

Three stacked selectable rows comparing 1. Animal, 2. Human, and 3. Artificial intelligence; 1. Animal is selected in red1. Animal2. Human3. Artificial

Intelligence:

Animal: Instinctual survival behaviors (e.g., birds migrating, bees building hives).

ARTIFICIAL INTELLIGENCE

Artificial Intelligence: Which One Is AI?

Which one is AI?

Four stacked selectable rows for Which one is AI: 1. Chess Game, 2. Music Recs, 3. Spell-Check, and 4. ElevenLabs; 1. Chess Game is selected1. Chess GameRules-Based AI2. Music RecsMachine Learning3. Spell-CheckDictionary Lookup4. ElevenLabsDeep Learning AI
  1. 1.A computer chess game playing against humans beats a human player. It uses rules-based algorithms to decide its next move.
Exercise · Which one is it?

Each scenario below shows a form of intelligence at work. Name whether it demonstrates Animal, Human, or Artificial intelligence.

Pacific salmon navigate thousands of kilometres back to the exact stream where they hatched to spawn.

Which concept does this case show?

CHAPTER 02

Machine Learning

DEFINITION

Machine Learning

A technique that uses Data and Performance Metric to automatically find how to solve a Task.

Contrasted with Rules-Based Expert Systems.

Gained popularity in the 21st century.

SIMULATED. Left: Rules-Based Expert Systems in blue, where a person writes explicit rules for a screen and performance stays flat at 62%. Right: Machine Learning in red, where Data and a Performance Metric automatically train the model on a screen; dragging the Data slider to 75% raises the Performance Metric to 91%Rules-Based Expert SystemsFixed RulesMetric: 62%SIMULATEDMachine Learning (21st Century)Performance Metric: 91%Solves TaskData: 75%

DRIVERS

Rise of Machine Learning

  1. 1.Increase in Compute Power
  2. 2.Increase in Data (Big Data)
  3. 3.Decrease in Storage Costs
SIMULATED. Three 21st-century drivers of Machine Learning at year 2020: Compute Power (78%, rising), Big Data (81%, rising), and Storage Costs (22%, falling), controlled by a timeline slider from 2000 to 2025SIMULATEDCompute Power↑ 78%Big Data↑ 81%Storage Costs↓ 22%Year: 20202000–2025
Exercise · Sort into groups

Sort each development into the force driving the 21st-century rise of Machine Learning that it illustrates.

        Tap a card, then tap the group it belongs to.

        CHAPTER 03

        Machine Learning Algorithms

        ML ALGORITHMS

        Decision Trees

        Imagine you are decision whether to go outside to play tennis.

        To make your decision you use a mentally formed Decision Tree.

        Interactive Decision Tree for playing tennis: root node Outlook branches to Sunny (Humidity: High -> No, Normal -> Yes), Overcast (Yes), and Rain (Wind: Strong -> No, Weak -> Yes). Active path: Sunny · Normal -> YesSunnyOvercastRainHighNormalStrongWeakOutlookHumidityWindNoSunny · HighYesSunny · NormalYesOvercastNoRain · StrongYesRain · Weak

        ML ALGORITHMS

        Linear Regression

        Imagine you have recorded transaction data for a retail store.

        You have the price of the items and their sales.

        You wish to predict sales based on the prices.

        SIMULATED. Linear Regression scatter plot of retail store transactions with Price on the horizontal axis and Sales on the vertical axis; at Price $25, the red regression line predicts Sales of 65 unitsPrice $25 → Sales: 65SIMULATEDSales9520Price ($10–$50)

        ML ALGORITHMS

        Logistic Regression

        SIMULATED. Logistic Regression S-curve for retail website click prediction: customer match score 68 yields 84% probability, predicting Click (1)Click (1) · 84%SIMULATED10.50Customer & Product Match: 68

        Imagine you have recorded historical click data for a retail website.

        You have product and customer characteristics in a database.

        You wish to predict whether a given customer will click on the product.

        Exercise · Which one is it?

        Each team below is choosing a machine learning algorithm. Name whether the task fits a Decision Tree, Linear Regression, or Logistic Regression.

        A grocery chain wants to predict the exact number of coffee bags a store will sell next week at different shelf prices.

        Which concept does this case show?

        ML ALGORITHMS

        Ensemble Learning

        Ensemble learning combines multiple models to improve prediction accuracy,

        Much like seeking advice from several experts to make a better decision.

        The idea is that the collective “wisdom” of diverse models reduces individual errors.

        Scenario: You want to predict if it will rain tomorrow.

        Single Model (Apple Weather):

        • Apple Whether predicts rain with 70% accuracy.
        • If it says “rain,” you’re 70% confident.

        Ensemble Approach (Apple Weather, AccuWeather, Google Weather):

        • Three apps, each with 70% accuracy, make independent predictions.
        • Possible outcomes:
          • All three say “rain” → Ensemble says “rain.”
          • Two say “rain,” one says “no rain” → Majority vote → “rain.”
          • If only one says “rain,” ensemble says “no rain.”
        Interactive Ensemble Learning weather forecast: Apple Weather (rain), AccuWeather (rain), and Google Weather (no rain) combine by majority vote (2 of 3) to predict rainApple Weather“rain”AccuWeather“rain”Google Weather“no rain”Majority Vote“rain”2 of 3 apps3 rain2 rain, 1 no1 rain, 2 no

        ENSEMBLE ALGORITHM

        Random Forest

        Random Forest is an ensemble learning algorithm that combines multiple decision trees to improve predictive accuracy and control overfitting. It is widely used for both classification and regression tasks.

        Five stacked selectable rows for Random Forest stages; 1. Ensemble is selected in red1. Ensemble2. Diversity3. Voting4. Robustness5. Advantages
        Ensemble Method: Builds a “forest” of decision trees, each trained on a random subset of the data and features.
        Exercise · Match up

        Match each core element of Random Forest with how it works when a logistics firm predicts package delivery delays.

        Random Forest Concept

        In the Delivery Delay Model

        Tap a term, then the example it goes with.

        CHAPTER 04

        Supervised vs. Unsupervised Learning

        LEARNING PARADIGMS

        Supervised vs. Unsupervised Learning

        Two stacked selectable rows comparing 1. Supervised Learning (Labeled Data) and 2. Unsupervised Learning (Unlabeled Data); 1. Supervised Learning is selected1. Supervised LearningLabeled Data2. Unsupervised LearningUnlabeled Data

        Supervised Learning: Algorithms learn from labeled data, where each input example is paired with a corresponding output (target). The model aims to map inputs to outputs for prediction or classification.

        • Example: Customer classification into hedonic vs. utility consumers

        Supervised:

        • Predicting house prices (regression).
        • Classifying emails as spam vs. not spam.
        • Medical diagnosis (e.g., identifying diseases from symptoms).

        COMPARISON

        Supervised vs. Unsupervised Learning: Comparison

        AspectSupervised LearningUnsupervised Learning
        DataLabeled (inputs + known outputs)Unlabeled (inputs only)
        GoalPredict outcomes or classify new dataExplore data structure, find patterns
        FeedbackDirect feedback (errors are corrected via labels)No feedback; model infers patterns autonomously
        Common TasksRegression, ClassificationClustering, Dimensionality Reduction
        Evaluation MetricsAccuracy, Precision, Recall, F1-Score, RMSESilhouette Score, Elbow Method, Reconstruction Error
        Visual comparison of Supervised Learning (top, in red) and Unsupervised Learning (bottom, in blue) for DataAspect: DataSupervisedInputs + LabelsUnsupervisedInputs Only12345
        Exercise · Sort into groups

        Sort each data characteristic, task, or evaluation metric into Supervised Learning or Unsupervised Learning.

            Tap a card, then tap the group it belongs to.

            CASE STUDIES

            Case Studies: Supervised vs. Unsupervised Learning

            Test your understanding by classifying these scenarios into Supervised vs. Unsupervised Learning

            Interactive Case Study selector (Cases 1 to 10): Case 1 (Customer Churn) uses Supervised LearningCase 1Case 2Case 3Case 4Case 5Case 6Case 7Case 8Case 9Case 10Case 1: Customer ChurnSupervised Learning · Predict Churn

            Case 1

            A company wants to predict customer churn (whether a customer will stop using their service) based on usage patterns and demographics. They have historical data showing which customers churned in the past.

            Answer: Supervised Learning

            Why? The model uses labeled data (past churn outcomes) to learn patterns and predict future churn.

            Knowledge Check

            A pharmacy chain launches two analytics projects: Project A uses five years of labeled prescription refill records (refilled on time vs. abandoned) to predict which patients will miss their next refill. Project B feeds 40 unlabeled store-level demographic and basket variables into an algorithm to group its 600 pharmacies into natural store formats. Which learning paradigms do Projects A and B use?

            CONCLUSION

            Conclusion

            KEY TAKEAWAYS

            1. 1Artificial Intelligence definition
            2. 2Rules-based AI vs ML
            3. 3Decision Trees
            4. 4Linear Regression
            5. 5Logistic Regression
            6. 6Ensemble Learning
            7. 7Random Forest
            8. 8Supervised vs. Unsupervised learning