AI is the broader concept of creating machines that simulate human intelligence. Machine Learning is a subset of AI where machines learn from data. Deep Learning is a subset of Machine Learning that uses neural networks with many layers.

Supervised learning is a type of machine learning where the model is trained on labeled data, meaning each training example is paired with an output label.

Unsupervised learning involves training a model on data without labeled responses, aiming to find hidden patterns or structures (e.g., clustering).

Overfitting occurs when a model performs well on training data but poorly on unseen data. It can be avoided using regularization, simplifying the model, increasing training data, or applying cross-validation.

Classification predicts a label (e.g., spam or not spam), while regression predicts a continuous value (e.g., predicting house prices).

A confusion matrix shows the performance of a classification model by comparing actual and predicted values, including metrics like True Positives, False Positives, False Negatives, and True Negatives.

Precision measures the accuracy of positive predictions: TP / (TP + FP). Recall measures the ability to find all positive samples: TP / (TP + FN).

Bias refers to the error from overly simplistic models. Variance refers to error from complex models. The tradeoff is finding a balance to minimize total error.

Cross-validation assesses how well a model generalizes to unseen data by splitting data into folds and evaluating performance across subsets.

Regularization techniques like L1 and L2 penalize model complexity to prevent overfitting and improve generalization.

Bagging reduces variance by training models on different data subsets (e.g., Random Forest). Boosting reduces bias by sequentially improving model errors (e.g., AdaBoost).

L1 regularization (Lasso) encourages sparsity by penalizing the absolute value of coefficients, while L2 (Ridge) penalizes their square, keeping all features.

Gradient Descent is an optimization algorithm used to minimize a loss function by iteratively adjusting parameters in the opposite direction of the gradient.

A Decision Tree splits data based on feature values to make decisions, using metrics like Gini impurity or entropy to select the best split.

A Decision Tree is a single tree structure, while a Random Forest is an ensemble of trees that uses bagging to reduce variance and improve accuracy.

An SVM is a supervised learning algorithm that finds the optimal hyperplane to separate data into different classes.

K-means clustering partitions data into k clusters by minimizing variance within clusters, iteratively assigning points and updating centroids.

PCA is a dimensionality reduction technique that transforms data into a new coordinate system, with variance maximized along principal components.

A generative model learns joint probability distributions (e.g., Naive Bayes), while a discriminative model learns conditional probabilities (e.g., Logistic Regression).

The kernel trick allows SVM to perform non-linear classification by implicitly mapping data into a higher-dimensional space.

CNNs are neural networks designed for processing images, using convolutional layers for feature extraction and pooling layers for downsampling.

RNNs handle sequential data with hidden states, while LSTMs address the vanishing gradient problem with gates to control information flow.

GANs consist of a Generator that creates synthetic data and a Discriminator that evaluates its authenticity, training in a competitive manner.

Transfer Learning involves using a pre-trained model on a related problem, fine-tuning it on new data instead of training from scratch.

Q-learning is a reinforcement learning algorithm where an agent learns action values by maximizing cumulative rewards.

Autoencoders are neural networks for dimensionality reduction, with encoders compressing data into latent space and decoders reconstructing it.

The vanishing gradient problem occurs when gradients become too small during backpropagation. Solutions include using ReLU, LSTMs, or batch normalization.

Stochastic Gradient Descent updates parameters for each example, while Batch Gradient Descent updates them using the entire dataset.

Activation functions introduce non-linearity. Examples include Sigmoid, ReLU, Tanh, and Softmax.

Dropout is a regularization technique that prevents overfitting by randomly ignoring neurons during training.