Machine Learning Overview

The field that allows computers to learn without being explicitly programmed. This involves using algorithms to analyze data and make predictions.

Machine learning focuses on learning input-output relationships to make predictions, while statistical models focus on understanding the reasons behind relationships and drawing inferences.

Machine Learning models that predict a continuous outcome, like stock prices or house prices.

Machine learning models that identify patterns and group data points with similar characteristics.

Machine learning models that learn from labeled data, where each input is paired with a corresponding output, aiming to replicate the correct answers.

Machine learning models that work with unlabeled data, discovering hidden patterns and structures within the data.

Machine learning models that learn through trial and error, based on rewards (positive or negative) received after making decisions.

The process where a model learns a function, mapping an input to an output based on training examples.

A technique involving training multiple decision trees sequentially, with each tree trying to correct the errors made by the previous one.

Represents discrete data (like words, images) as points in a continuous, lower-dimensional space.

A relationship between two variables where a change in one directly causes a change in the other.

The relationship between two variables is not due solely to a third variable.

The cause happens before the effect.

A group of decision trees trained independently on different subsets of the data.

Combining multiple learning models to make a more robust and accurate prediction.

Training a machine learning model on a set of data to learn how to predict an output based on given inputs.

Combines various "weak" models to create a stronger model, similar to taking advice from multiple experts.

Technique used to train multiple decision trees in a random forest model. It involves creating multiple datasets by randomly sampling with replacement.

Boosting trains models sequentially, with each new model focusing on correcting the errors of the previous one.

A neural network with multiple layers of interconnected nodes, capable of learning complex nonlinear relationships.

Tuning hyperparameters, such as the number of layers and nodes in an MLP, to improve performance.

Repeatedly training and evaluating a model on different subsets of the data to avoid overfitting and improve generalization.

Randomly selecting features for each split in a decision tree in a random forest model, promoting diversity and reducing overfitting.

A technique used to stop training a model early when performance on a validation set starts to decrease, preventing overfitting.

A Decision Tree represents data in a hierarchical structure starting at the root node and branching out to represent decisions and outcomes.

The root node contains all the data, while split nodes represent decisions based on features. Each branch represents a specific split, and leaf nodes indicate the final class or value.

Entropy measures the disorder or uncertainty within a dataset. Information Gain determines which feature split provides the most information by reducing the entropy of the resulting subsets.

Decision Trees use a greedy algorithm to choose the best split locally at each step. This means it chooses splits that are optimal for the current node without considering the overall best solution.

Overfitting occurs when a Decision Tree becomes too complex and learns the training data perfectly but fails to Generalize to new data. Pruning techniques help to avoid overfitting.

Pre-pruning techniques, like maximum depth, minimum examples per leaf, and minimum impurity reduction, are performed before the decision tree is fully grown.

Feature importance in Decision Trees is determined by how much a feature contributes to reducing node impurity. This indicates which features are most influential in the decision process.

Metrics like Accuracy, Precision, Recall, Specificity, and MCC are used to evaluate the performance of Decision Trees based on how well they classify data points correctly.

Ethical concerns around AI, such as privacy violations, discrimination, and the spread of misinformation.

The potential for AI to harm people's privacy, like revealing sensitive information about their location or behavior.

AI techniques that make predictions and decisions while being transparent and explainable, providing insights into the model's reasoning.

Challenges in creating XAI systems, balancing the need for transparency with the model's performance.

Using AI to analyze climate data, simulate emission reductions, and track climate change impacts.

Utilizing AI to optimize energy use in buildings, transportation networks, and other areas, potentially reducing carbon emissions.

Potential impact of AI on the environment due to high energy consumption during training and operation.

Guidelines for developing and implementing explainable AI systems that are fair, ethical, and beneficial to society.

A method for understanding a model's predictions for individual data points by slightly changing inputs and observing the corresponding changes in outputs.

A technique for explaining predictions locally by analyzing the model's behavior around a specific data point. It involves perturbing the data point, generating predictions, training a simple model to explain the prediction.

A set of alternative data points that would have resulted in a different prediction from the original data point.

A technique that explains predictions by attributing the prediction to individual features through a game-theoretic approach, where each feature is a player and the output is the final prediction.

A graph that shows the relationship between any one feature and the predicted outcome by averaging over all other features.

A method that shows predictions for individual data points instead of averages, providing a more detailed understanding of the model's behavior.

An algorithm used to predict the probability of recidivism (re-offending) in American justice by assessing risk factors.

Techniques that provide insights into the underlying logic and decision-making process of a machine learning model.