Supervised Machine Learning Tutorial

An In-Depth Exploration of Supervised Machine Learning in Machine Learning Tutoring.

Machine learning has transformed industries and applications across the board, and one of its foundational techniques is supervised learning. Supervised machine learning forms the basis of predictive modeling, classification, regression, and much more. In this comprehensive guide, we delve into the world of supervised machine learning. We begin with the basics, explaining what supervised learning is and its key components. We then explore various algorithms, techniques, and real-world applications. Whether you're a beginner looking to grasp the fundamentals or an experienced practitioner seeking a deeper understanding, this guide has something for everyone.

Table of Contents

1. Introduction
• What is Machine Learning?
• The Role of Supervised Learning
• Why is Supervised Learning Important?
2. Foundations of Supervised Learning
• Data: The Fuel of Supervised Learning
• Labels: The Supervision in Supervised Learning
• The Training and Testing Split
• Performance Metrics: How to Measure Success
3. Common Supervised Learning Algorithms
• Linear Regression
• Logistic Regression
• k-Nearest Neighbors (k-NN)
• Decision Trees
• Random Forests
• Support Vector Machines (SVM)
• Naive Bayes
• Neural Networks
4. Model Training and Evaluation
• Data Preprocessing
• Feature Engineering
• Model Training
• Cross-Validation
• Overfitting and Underfitting
• Hyperparameter Tuning
• Model Evaluation
5. Classification Problems
• Binary Classification
• Multi-Class Classification
• Imbalanced Classes
• Receiver Operating Characteristic (ROC) Curve
• Area Under the Curve (AUC)
6. Regression Problems
• Linear Regression Revisited
• Polynomial Regression
• Ridge and Lasso Regression
• Gradient Boosting for Regression
• Time Series Forecasting
7. Ensemble Methods
• What are Ensemble Methods?
• Bagging: Bootstrap Aggregating
• Boosting: Improving Weak Learners
• Stacking: Combining Multiple Models
8. Real-World Applications of Supervised Learning
• Healthcare: Disease Diagnosis and Predictions
• Finance: Credit Scoring and Fraud Detection
• E-commerce: Recommender Systems
• Natural Language Processing: Sentiment Analysis
• Image Classification: Object Recognition
• Autonomous Vehicles: Perception and Control
• Social Media: Personalized Content
9. Challenges and Considerations
• Data Quality and Quantity
• Bias and Fairness
• Ethical Considerations
• Model Interpretability
10. Future Trends in Supervised Learning
• Transfer Learning
• Explainable AI (XAI)
• Federated Learning
• Reinforcement Learning in Supervised Settings
11. Conclusion
• Recap of Supervised Learning
• The Endless Potential of Supervised Machine Learning
12. References

1. Introduction

What is Machine Learning?

Machine learning is a subfield of artificial intelligence (AI) that focuses on the development of algorithms and statistical models that enable computers to learn from and make predictions or decisions based on data. In other words, it's about teaching machines to perform tasks without being explicitly programmed for each step.

Machine learning algorithms can be broadly categorized into three types:

• Supervised Learning: In supervised learning, algorithms learn from labeled data, making predictions or decisions based on input data. It's like teaching a machine to recognize patterns by showing it examples with clear labels.
• Unsupervised Learning: Unsupervised learning deals with unlabeled data, where the algorithm seeks to find hidden patterns or structures within the data. Clustering and dimensionality reduction are common tasks in this category.
• Reinforcement Learning: Reinforcement learning is about training agents to make sequences of decisions in an environment to maximize a cumulative reward. It's commonly used in robotics, gaming, and autonomous systems.

The Role of Supervised Machine Learning

Supervised learning is a foundational and widely used subset of machine learning. In supervised learning, the algorithm learns a mapping function from input data to output labels by training on a dataset that consists of both input-output pairs. The primary goal is to generalize from the training data to make accurate predictions or decisions on unseen, new data.

Why is Supervised Learning Important?

Supervised learning is crucial because it enables machines to learn from historical data and apply that knowledge to new, unseen data. This capability has far-reaching implications across various domains, including healthcare, finance, e-commerce, natural language processing, image analysis, and many others.

For example, in healthcare, supervised learning can be used to predict disease outcomes based on patient data, aiding in early diagnosis and treatment planning. In finance, it powers credit scoring models that assess creditworthiness, and in e-commerce, it drives recommendation systems, improving customer experiences. In essence, supervised learning is the backbone of predictive modeling and decision-making in the modern world.

2. Foundations of Supervised Learning

Before diving into specific algorithms and techniques, it's essential to understand the fundamental concepts that underpin supervised learning.

Data: The Fuel of Supervised Learning

At the heart of supervised learning is data. Data is the raw material that algorithms feed on to learn patterns and make predictions. In supervised learning, the dataset typically consists of two main components:

• Features (Inputs): These are the variables or attributes that describe the input data. For example, in a spam email classification task, features could include email content, sender, and timestamp.
• Labels (Outputs): Labels are the target values or categories that the algorithm aims to predict or classify. In the spam email example, labels would be binary: spam or not spam.

Labels: The Supervision in Supervised Learning

The term "supervised" in supervised learning refers to the presence of labeled data. Labels provide the algorithm with supervision, allowing it to learn the relationship between features and the corresponding labels. The process of associating features with labels is often referred to as "training" the model.

The Training and Testing Split

To evaluate the performance of a supervised learning model, it's common practice to split the dataset into two subsets: a training set and a testing set.

• Training Set: This portion of the data is used to train the model. The model learns patterns and relationships from the training data.
• Testing Set: The testing set is kept separate from the training data and is used to assess how well the model generalizes to new, unseen data. It helps estimate the model's performance on real-world scenarios.

Performance Metrics: How to Measure Success

The choice of performance metrics depends on the specific task and the nature of the data. Some common evaluation metrics for supervised learning include:

• Accuracy: Measures the proportion of correct predictions out of the total predictions.
• Precision and Recall: Useful for imbalanced datasets, precision measures the proportion of true positive predictions out of all positive predictions, while recall measures the proportion of true positives out of all actual positives.
• F1 Score: The harmonic mean of precision and recall, providing a balanced measure of a model's performance.
• Mean Absolute Error (MAE): Commonly used for regression tasks, MAE calculates the average absolute difference between predicted and actual values.
• Mean Squared Error (MSE): Another regression metric, MSE calculates the average squared difference between predicted and actual values.

Choosing the appropriate metric depends on the problem at hand. For example, in a medical diagnosis task, recall (minimizing false negatives) may be more important than precision, while in a recommendation system, accuracy and user satisfaction might be the key metrics.

3. Common Supervised Machine Learning Algorithms

Now that we've covered the foundational concepts, let's explore some of the most common supervised learning algorithms.

Linear Regression

Linear regression is one of the simplest yet powerful algorithms for supervised learning. It's used for regression tasks where the goal is to predict a continuous numeric value. Linear regression aims to find the best-fit linear relationship between the input features and the target variable.

Logistic Regression

Logistic regression, despite its name, is a classification algorithm. It's commonly used for binary classification tasks, such as spam detection or disease diagnosis.

Logistic regression models the probability that a given input belongs to a particular class. The logistic function, also known as the sigmoid function, is used to map the linear combination of features to a probability value between 0 and 1.

k-Nearest Neighbors (k-NN)

k-Nearest Neighbors is a versatile algorithm used for both classification and regression tasks. It works on the principle of proximity, meaning that objects that are close to each other in feature space are likely to belong to the same class or have similar output values.

In k-NN, the algorithm finds the k-nearest data points to the input and assigns the class label or output value based on the majority among the k-neighbors. The choice of k is a hyperparameter that can be tuned.

Decision Trees

Decision trees are a popular choice for both classification and regression tasks. They are intuitive to understand and visualize. A decision tree splits the data into subsets based on the most significant feature at each node. The splitting process continues until a stopping criterion is met, such as a maximum depth or purity threshold.

Decision trees are easy to interpret, making them valuable for explaining model decisions. However, they can be prone to overfitting if not appropriately pruned.

Random Forests

Random forests are an ensemble learning method that builds multiple decision trees and combines their predictions to improve accuracy and reduce overfitting. Each tree in the forest is trained on a random subset of the data (bootstrap samples) and a random subset of features, providing diversity among the trees.

Random forests are robust and perform well on a wide range of tasks. They are less susceptible to overfitting compared to individual decision trees.

Support Vector Machines (SVM)

Support Vector Machines are powerful for both classification and regression tasks. SVM aims to find a hyperplane that best separates data points belonging to different classes while maximizing the margin (distance) between the classes. The hyperplane is chosen to minimize classification errors.

SVM can handle high-dimensional data and is effective when dealing with datasets with a clear separation between classes. It's particularly useful for binary classification problems.

Naive Bayes

Naive Bayes is a probabilistic algorithm commonly used for text classification tasks, such as spam detection and sentiment analysis. Despite its simplicity, it often performs surprisingly well.

Naive Bayes is based on Bayes' theorem and assumes that features are conditionally independent, which is a simplifying but unrealistic assumption in many real-world scenarios. Despite this simplification, it can be very effective for certain types of problems.

Neural Networks

Neural networks, particularly deep neural networks, have gained immense popularity in recent years, driving advancements in various fields, including computer vision, natural language processing, and speech recognition.

Neural networks are composed of layers of interconnected nodes (neurons) that learn hierarchical representations of data. Deep learning models with many layers (deep neural networks) can capture complex patterns and features in large datasets. Common architectures include feedforward neural networks, convolutional neural networks (CNNs), and recurrent neural networks (RNNs).

These are just a few examples of supervised learning algorithms, each with its strengths and weaknesses. The choice of algorithm depends on the problem at hand, the nature of the data, and computational resources available.

4. Model Training and Evaluation

Building an effective supervised learning model involves more than just choosing the right algorithm. It requires careful consideration of data preprocessing, feature engineering, model training, and evaluation.

Data Preprocessing

Data preprocessing is a crucial step that involves cleaning and transforming raw data to make it suitable for modeling. Common data preprocessing tasks include:

• Data Cleaning: Handling missing values, outliers, and noisy data.
• Feature Scaling: Ensuring that features are on a similar scale to prevent some features from dominating the learning process.
• Feature Encoding: Converting categorical features into numerical representations (e.g., one-hot encoding).
• Feature Selection: Choosing the most relevant features to improve model efficiency.
• Data Normalization: Scaling data to have a mean of 0 and a standard deviation of 1.

Feature Engineering

Feature engineering is the process of creating new features or transforming existing ones to improve a model's performance. It requires domain knowledge and creativity. Effective feature engineering can significantly impact the model's ability to capture meaningful patterns in the data.

For example, in a natural language processing task, feature engineering might involve creating features based on the frequency of specific words or phrases in text data. In image analysis, it could entail extracting features from pixel values or using pre-trained deep learning models to extract features automatically.

Model Training

Model training is the process of teaching the algorithm to make predictions or classifications based on the training data. The algorithm learns the relationship between input features and labels by adjusting its internal parameters.

During training, the algorithm aims to minimize a predefined loss function, which quantifies the difference between the predicted values and the true labels. Optimization techniques, such as gradient descent, are used to update the model's parameters iteratively.

Cross-Validation

Cross-validation is a technique used to assess a model's performance and generalization ability. It involves splitting the data into multiple folds, training the model on different subsets of the data, and evaluating its performance on the remaining data. Common cross-validation methods include k-fold cross-validation and leave-one-out cross-validation.

Cross-validation helps detect overfitting by providing a more robust estimate of a model's performance on unseen data. It also helps tune hyperparameters effectively.

Overfitting and Underfitting

Two common challenges in supervised learning are overfitting and underfitting:

• Overfitting: Occurs when a model learns the training data too well, capturing noise and irrelevant patterns. As a result, it performs poorly on new, unseen data.
• Underfitting: Occurs when a model is too simplistic and fails to capture the underlying patterns in the data. It performs poorly on both the training and testing data.

Balancing the model's complexity to avoid both overfitting and underfitting is a critical aspect of model selection and training.

Hyperparameter Tuning

Hyperparameters are parameters that are not learned from the data but are set prior to training. They include parameters like the learning rate, regularization strength, and the number of hidden layers in a neural network. Tuning hyperparameters is essential for optimizing model performance.

Hyperparameter tuning can be done manually by iteratively adjusting hyperparameters and evaluating the model's performance or using automated techniques like grid search or random search.

Model Evaluation

Once a model is trained, it's important to evaluate its performance on the testing set or using cross-validation. The choice of evaluation metric depends on the specific task:

• For classification tasks, metrics like accuracy, precision, recall, F1 score, and the area under the ROC curve (AUC-ROC) are commonly used.
• For regression tasks, metrics like mean absolute error (MAE), mean squared error (MSE), and R-squared are used.

Model evaluation helps determine whether the model meets the desired performance criteria and whether further improvements are necessary.

5. Classification Problems

Classification is a common type of supervised learning task, where the goal is to assign input data to one of several predefined categories or classes. Let's delve deeper into classification.

Binary Classification

In binary classification, there are two possible classes or outcomes. Examples include spam detection (spam or not spam), disease diagnosis (positive or negative), and sentiment analysis (positive or negative sentiment).

To perform binary classification, algorithms typically use decision boundaries to separate the two classes. The output is often a probability score, and a threshold is applied to make the final classification decision.

Multi-Class Classification

In multi-class classification, there are more than two classes to choose from. Examples include handwritten digit recognition (digits 0-9), image classification (various object categories), and natural language processing tasks like language identification (multiple languages).

Multi-class classification algorithms extend binary classification techniques to handle multiple classes. Common approaches include one-vs-all (OvA) and softmax regression.

Imbalanced Classes

In many classification problems, class imbalances can be a challenge. Imbalanced classes occur when one class has significantly more instances than the other(s). For example, in fraud detection, the majority of transactions are non-fraudulent.

Addressing imbalanced classes may involve techniques like resampling (oversampling the minority class or undersampling the majority class), using different evaluation metrics (precision-recall instead of accuracy), or employing specialized algorithms like SMOTE (Synthetic Minority Over-sampling Technique).

Receiver Operating Characteristic (ROC) Curve

The ROC curve is a graphical representation of a binary classifier's performance at various thresholds. It plots the true positive rate (TPR or recall) against the false positive rate (FPR) as the threshold changes. The ROC curve helps assess the trade-off between sensitivity and specificity.

Area Under the Curve (AUC)

The AUC measures the area under the ROC curve and provides a single scalar value that quantifies a model's ability to distinguish between the positive and negative classes. A higher AUC indicates better discrimination.

In summary, classification is a fundamental problem in supervised learning with applications in various domains. It involves making decisions about which category or class an input data point belongs to, and the choice of classification algorithm depends on the specific problem and dataset.

6. Regression Problems

While classification deals with discrete categories or classes, regression is a type of supervised learning that addresses continuous prediction problems. In regression, the goal is to predict a continuous numeric value or output.

Let's explore some common regression techniques and scenarios.

Linear Regression Revisited

Linear regression, introduced earlier, is a straightforward regression technique that models the relationship between input features and a continuous target variable. It assumes a linear relationship between the features and the target.

For example, in real estate, linear regression can be used to predict the price of a house based on features like square footage, number of bedrooms, and location.

Polynomial Regression

Polynomial regression extends linear regression to capture more complex relationships between features and the target variable. Instead of fitting a straight line, polynomial regression fits a polynomial curve to the data.

Polynomial regression is useful when the relationship between the features and the target is nonlinear. However, it can lead to overfitting if the degree of the polynomial is too high.

Ridge and Lasso Regression

Ridge and Lasso regression are regularization techniques used to prevent overfitting in linear regression models. They add a penalty term to the linear regression loss function to discourage large coefficients.

• Ridge Regression: Adds an L2 regularization term to the loss function, penalizing the sum of the squared coefficients. This encourages the model to have small but non-zero coefficients.
• Lasso Regression: Adds an L1 regularization term to the loss function, penalizing the absolute value of coefficients. Lasso can drive some coefficients to exactly zero, effectively performing feature selection.

These techniques are particularly useful when dealing with high-dimensional data, where overfitting is a common concern.

Gradient Boosting for Regression

Gradient boosting is an ensemble technique used for regression tasks. It builds an ensemble of weak learners (typically decision trees) and combines their predictions to create a strong, accurate model.

One of the popular gradient boosting algorithms is XGBoost, known for its speed and performance. Gradient boosting models iteratively correct errors made by previous models, leading to improved accuracy.

Time Series Forecasting

Time series forecasting is a specialized type of regression used to predict future values based on historical time-ordered data. It's widely applied in finance (stock price prediction), weather forecasting, demand forecasting, and many other domains.

Common time series forecasting methods include autoregressive integrated moving average (ARIMA), exponential smoothing, and machine learning approaches like recurrent neural networks (RNNs) and long short-term memory networks (LSTMs).

Regression is a versatile technique that finds applications in numerous fields, including finance, economics, environmental science, and more. The choice of regression method depends on the nature of the data and the complexity of the relationship between features and the target variable.

7. Ensemble Methods

Ensemble methods are a powerful technique in supervised learning that involve combining multiple models to improve predictive accuracy and reduce overfitting. Ensemble methods work by aggregating the predictions of individual models to make a final prediction.

There are several ensemble methods, each with its own approach to combining models. Let's explore some of the most popular ones.

What are Ensemble Methods?

Ensemble methods use the wisdom of the crowd principle, where multiple models, when combined, often perform better than any individual model. These methods aim to reduce the variance (overfitting) and bias (underfitting) of individual models, leading to improved overall performance.

Bagging: Bootstrap Aggregating

Bagging is a technique that involves training multiple instances of the same model on different subsets of the training data. Each subset is created by randomly sampling the training data with replacement. The term "bootstrap" refers to this resampling process.

The final prediction is obtained by averaging or taking a majority vote of the predictions from each individual model. Bagging is particularly effective for reducing variance and improving the stability of models.

The most famous bagging algorithm is Random Forest, which builds an ensemble of decision trees using bagging. Random Forests are widely used due to their robustness and ability to handle high-dimensional data.

Boosting: Improving Weak Learners

Boosting is an ensemble technique that aims to improve the performance of weak learners (models that perform slightly better than random guessing) by iteratively training new models that focus on the errors made by previous models.

One of the most popular boosting algorithms is AdaBoost (Adaptive Boosting). AdaBoost assigns different weights to data points, with higher weights given to points that were misclassified by previous models. Subsequent models are trained to focus on these challenging examples.

Another powerful boosting algorithm is Gradient Boosting, which builds an ensemble of decision trees where each new tree corrects the errors made by the previous ones. Gradient Boosting can be used for both regression and classification tasks.

Stacking: Combining Multiple Models

Stacking, also known as stacked generalization, is an ensemble technique that combines multiple diverse models into a meta-model. Instead of simply averaging or voting on predictions, stacking trains a secondary model (meta-model) on the predictions of individual base models.

The idea is to let the meta-model learn the optimal way to weigh the predictions from different base models. Stacking often leads to improved performance, especially when base models have complementary strengths and weaknesses.

Ensemble methods are a valuable tool in a data scientist's toolkit, and the choice between bagging, boosting, or stacking depends on the specific problem and dataset. These techniques have been instrumental in winning machine learning competitions and achieving state-of-the-art results in various domains.

8. Real-World Applications of Supervised Machine Learning

Supervised learning has found its way into a wide range of real-world applications, driving innovation and improvements in various domains. Let's explore some notable examples:

Healthcare: Disease Diagnosis and Predictions

In healthcare, supervised learning plays a critical role in disease diagnosis and prediction. For instance:

• Medical Imaging: Supervised learning models analyze medical images (e.g., X-rays, MRIs, CT scans) to detect and diagnose conditions like cancer, fractures, and neurological disorders.
• Patient Risk Assessment: Algorithms predict the risk of diseases like diabetes, heart disease, and sepsis based on patient data, enabling early intervention and personalized treatment plans.
• Drug Discovery: Supervised learning aids in drug discovery by identifying potential drug candidates and predicting their effectiveness.

Finance: Credit Scoring and Fraud Detection

The financial industry relies heavily on supervised learning for various tasks:

• Credit Scoring: Algorithms assess creditworthiness by analyzing applicants' financial history and demographics, helping banks make informed lending decisions.
• Fraud Detection: Machine learning models detect fraudulent transactions by analyzing patterns and anomalies in transaction data, protecting both consumers and financial institutions.
• Algorithmic Trading: Trading algorithms use supervised learning to make buy and sell decisions based on historical market data.

E-commerce: Recommender Systems

E-commerce platforms leverage supervised learning to improve user experiences:

• Product Recommendations: Recommender systems analyze user behavior and preferences to suggest products, increasing sales and customer satisfaction.
• Personalization: E-commerce sites use personalization algorithms to tailor the user interface, content, and marketing messages to individual users.

Natural Language Processing: Sentiment Analysis

Supervised learning is at the core of many natural language processing (NLP) applications:

• Sentiment Analysis: Algorithms classify text as positive, negative, or neutral, helping companies gauge public sentiment and make data-driven decisions.
• Chatbots and Virtual Assistants: Chatbots and virtual assistants use NLP models to understand and respond to user queries and commands.

Image Classification: Object Recognition

Image classification is a classic supervised learning task with applications in various fields:

• Autonomous Vehicles: Self-driving cars use image classification to identify pedestrians, vehicles, and road signs for safe navigation.
• Medical Imaging: Radiologists use image analysis models to detect and classify abnormalities in medical images.
• Agriculture: Supervised learning helps identify plant diseases, pests, and crop quality issues in agriculture.

Autonomous Vehicles: Perception and Control

Autonomous vehicles rely on supervised learning for perception tasks like object detection, lane following, and obstacle avoidance. These models analyze sensor data (e.g., cameras, LiDAR, radar) to make real-time driving decisions.

Social Media: Personalized Content

Social media platforms employ supervised learning to personalize users' content feeds and recommend friends, posts, and advertisements based on user behavior and preferences.

These real-world applications highlight the impact of supervised learning in various industries. As the field of machine learning continues to evolve, the range of applications is expected to expand, further improving efficiency, accuracy, and decision-making across domains.

9. Challenges and Considerations

While supervised learning offers tremendous benefits, it also comes with its set of challenges and considerations that practitioners and researchers must address.

Data Quality and Quantity

• Data Availability: Obtaining high-quality labeled data can be challenging and expensive. Data collection and annotation processes may require substantial resources.
• Data Imbalance: In classification tasks, imbalanced class distributions can lead to biased models. Strategies like resampling or using different evaluation metrics are necessary to mitigate this issue.

Bias and Fairness

• Bias in Data: Biased training data can lead to discriminatory models that disproportionately affect certain groups. Careful data preprocessing and fairness-aware algorithms are necessary to address bias.
• Ethical Considerations
• Supervised learning models can inadvertently propagate societal biases present in the training data. Ethical considerations are essential when deploying models that impact individuals or communities.

Model Interpretability

• Black Box Models: Complex models like deep neural networks can be challenging to interpret, making it difficult to understand why a model made a particular prediction. Explainable AI (XAI) techniques are emerging to address this challenge.
• Regulatory Compliance: In some industries, like healthcare and finance, regulatory requirements mandate model interpretability and transparency.

10. Future Trends in Supervised Learning

The field of supervised learning continues to evolve with ongoing research and emerging trends. Here are some future directions and trends to watch for:

Transfer Learning

Transfer learning is the practice of pretraining models on large datasets and then fine-tuning them for specific tasks. This approach reduces the need for massive labeled datasets and accelerates model development. Transfer learning is gaining popularity in natural language processing and computer vision.

Explainable AI (XAI)

As machine learning models become more complex, there's a growing need for interpretable and transparent models. XAI techniques aim to provide insights into model decision-making, making AI systems more accountable and trustworthy.

Federated Learning

Federated learning enables model training across decentralized devices or servers while keeping data localized. It's particularly useful in privacy-sensitive applications like healthcare, where data must remain on the user's device.

Reinforcement Learning in Supervised Settings

Reinforcement learning, traditionally used in autonomous systems and gaming, is increasingly being applied to supervised learning tasks. This integration allows models to make sequential decisions and adapt to changing environments.

11. Conclusion

In this comprehensive guide, we've explored supervised machine learning, a foundational and powerful branch of artificial intelligence. We began with the basics, including data, labels, training, and evaluation. We then delved into common algorithms, techniques, and real-world applications across various domains.

Supervised learning has revolutionized industries, from healthcare and finance to e-commerce and autonomous vehicles. Its ability to make predictions and decisions based on data has paved the way for innovation and automation.

However, it's essential to approach supervised learning with an understanding of its challenges, such as data bias and model interpretability. As the field continues to evolve, trends like transfer learning, explainable AI, federated learning, and reinforcement learning in supervised settings will shape the future of machine learning.

Whether you're just starting your journey in supervised learning or you're an experienced practitioner, this guide serves as a valuable resource to deepen your understanding and stay informed about the latest developments in the field. Supervised learning, with its endless potential, continues to push the boundaries of what's possible in the world of AI and machine learning.