Basics of Machine Learning

Machine Learning Basics

Introduction to Machine Learning

  • Humans learn from past experiences, while machines follow instructions.
  • Machine learning involves training machines to learn from past data, enabling them to perform tasks faster and more efficiently than humans.
  • Machine learning is not just about learning but also understanding and reasoning.

Paul's Song Preferences: A Classification Example

  • Paul likes or dislikes songs based on tempo, genre, intensity, and the singer's gender.
  • For simplicity, consider only tempo (x-axis: relaxed to fast) and intensity (y-axis: light to soaring).
  • Paul likes songs with fast tempo and soaring intensity and dislikes songs with relaxed tempo and light intensity.
Classifying Song A
  • Song A has fast tempo and soaring intensity.
  • Based on Paul's past choices, it's easy to classify that Paul will like Song A.
The Challenge with Song B
  • Song B has medium tempo and medium intensity.
  • It's unclear whether Paul will like or dislike Song B, indicating a need for machine learning.
K-Nearest Neighbors Algorithm
  • Draw a circle around Song B to consider nearby data points.
  • If there are four votes for "like" and one vote for "dislike," the majority predicts Paul will like the song.
  • This example demonstrates the k-nearest neighbors algorithm, a basic machine learning technique.

The Role of Machine Learning

  • Machine learning is useful when choices become complicated.
  • It learns from data, builds a prediction model, and predicts outcomes for new data points.
  • More data leads to a better model and higher accuracy.

Types of Machine Learning

  • Supervised learning
  • Unsupervised learning
  • Reinforcement learning
Supervised Learning
  • Example: Identifying currency based on weight.
  • Given one million coins of three currencies: one rupee (3 grams), one euro (7 grams), and one dirham (4 grams).
  • Weight is the feature, and currency is the label.
  • The model learns the association between weight and currency.
  • Supervised learning uses labeled data to train the model.
Unsupervised Learning
  • Example: Analyzing a cricket dataset.
  • Dataset includes players' scores and wickets taken.
  • The machine identifies patterns without labels.
  • Plots data with wickets on the x-axis and runs on the y-axis.
  • Identifies two clusters: batsmen (higher runs, fewer wickets) and bowlers (fewer runs, many wickets).
  • Learning with unlabeled data is unsupervised learning.
Reinforcement Learning
  • Reward-based learning using feedback.
  • Example: Identifying an image of a dog.
  • The system initially identifies it as a cat, receiving negative feedback.
  • The machine learns from the feedback and correctly classifies future dog images.

Generalizing a Machine Learning Model

  • Input is fed into a machine learning model.
  • The model produces output based on the applied algorithm.
  • If the output is correct, it's taken as the final result.
  • If incorrect, feedback is provided to refine the training model.

Quiz: Supervised vs. Unsupervised Learning

  • Scenario 1: Facebook recognizes a friend in a tagged photo album (Supervised).
  • Scenario 2: Netflix recommends movies based on past choices (Supervised).
  • Scenario 3: Analyzing bank data for suspicious transactions and flagging fraudulent ones (Supervised).

Factors Enabling Machine Learning Today

  • Availability of humongous data due to increased online activity.
  • Increased memory handling capabilities of computers.
  • Great computational powers of modern computers.

Applications of Machine Learning

  • Healthcare: Diagnostics prediction for doctors.
  • Sentiment analysis on social media.
  • Fraud detection in finance.
  • Predicting customer churn in e-commerce.
  • Surge pricing by Uber: Differential pricing in real-time based on demand, number of cars, weather, etc to match supply and demand.

Conclusion

  • Encouragement to identify everyday examples of machine learning.
  • Invitation to watch for more videos.