Concepts, made simple.

Attention is a mechanism that lets a model focus on the most relevant parts of input data when making predictions.

Backpropagation is the algorithm that trains neural networks by calculating how much each weight contributed to the error and adjusting them to reduce it.

A Convolutional Neural Network is a type of AI that learns to recognize patterns in images, like edges, shapes, or objects, by scanning them in small chunks.

Deep Learning is a type of machine learning that uses many-layered neural networks to automatically learn patterns from vast amounts of data.

Gradient Descent is an optimization algorithm that iteratively adjusts parameters to minimize a loss function by moving in the direction of steepest descent.

A Large Language Model is an AI trained on vast text data to predict and generate human-like text.

A loss function measures how wrong a model's predictions are, guiding it to improve.

Machine Learning is a way for computers to learn from data without being explicitly programmed for every task.

A neural network is a computer system inspired by the brain that learns patterns from data to make decisions or predictions.

Pretraining is the initial, broad learning phase where an AI model is trained on a huge amount of general data to understand language patterns before being fine-tuned for specific tasks.

Self-attention lets a model weigh the importance of every word in a sentence when processing each word, capturing relationships regardless of distance.

A Transformer is a neural network architecture that processes all parts of a sequence simultaneously using self-attention, revolutionizing AI tasks like language translation and text generation.

Batch Normalization speeds up and stabilizes training of deep neural networks by normalizing each layer's inputs to have zero mean and unit variance.

Embedding is a way to turn words, images, or other data into a list of numbers that captures their meaning and relationships.

A GAN pits two neural networks against each other to generate new, realistic data from scratch.

Inference is when a trained AI uses its knowledge to make predictions or decisions on new data.

Latent space is a compressed, abstract representation of data where similar items cluster together, used by AI to understand and generate new content.

LSTM is a type of neural network that remembers information for long periods, solving the problem of forgetting in standard RNNs.

Overfitting is when a model learns the training data too perfectly, including its noise, and fails to generalize to new data.

Parameters are the internal numbers an AI learns from training data that determine how it processes inputs and generates outputs.

Regularization prevents AI from memorizing training data too well, helping it perform better on new, unseen data.

Reinforcement Learning is a type of machine learning where an agent learns to make decisions by taking actions in an environment to maximize a reward signal.

A token is the smallest unit of text that an AI model processes, like a word or part of a word.

Tokenization is the process of breaking text into smaller pieces, called tokens, that a language model can understand.

Transfer learning reuses a model trained on one task to jump-start learning on a different but related task.

Chain of Thought is a technique where AI models break down complex problems into intermediate reasoning steps, like showing their work in math class.

A context window is the amount of text an AI model can 'see' at once to understand and generate responses.

Cosine similarity measures how similar two vectors are by calculating the cosine of the angle between them, ignoring magnitude.

A diffusion model is an AI that learns to generate data, like images, by gradually adding noise to training data and then learning to reverse that process to create new, clean data from random noise.

Dropout randomly turns off some neurons during training to prevent overfitting and make neural networks more robust.

An epoch is one complete pass of the entire training dataset through a machine learning model.

Fine-tuning is the process of taking a pre-trained AI model and further training it on a smaller, specific dataset to adapt it for a particular task.

Hallucination is when an AI makes up confident-sounding but false information.

Mixture of Experts is a machine learning technique that uses multiple specialized sub-models (experts) and a gating network to route each input to the most relevant expert, improving efficiency and performance.

Prompt engineering is the practice of designing and refining input prompts to get the best possible output from an AI language model.

Quantization reduces the precision of numbers in a neural network to make it faster and smaller, like rounding 3.14159 to 3.14.

RAG (Retrieval-Augmented Generation) is a technique that lets AI models look up external information before answering, making their responses more accurate and up-to-date.

A Recurrent Neural Network (RNN) is a type of artificial neural network that uses loops to remember previous inputs, making it ideal for processing sequences like sentences or stock prices.

RLHF (Reinforcement Learning from Human Feedback) is a technique that trains AI models to produce outputs aligned with human preferences by using human ratings as a reward signal.

Softmax turns a list of numbers into probabilities that sum to 1, making it easy to pick the most likely option.

A vector database stores and retrieves data as mathematical vectors, enabling fast similarity searches for AI applications like recommendation systems and semantic search.

Few-shot learning is an AI technique where a model learns to recognize new concepts from just a handful of examples, like a child learning a new animal from a few photos.

Knowledge Distillation is a technique where a small, fast model learns to mimic a large, accurate teacher model, compressing its knowledge into a more efficient form.

Multimodal AI processes and combines multiple types of data, like text, images, and sound, to understand or generate information more comprehensively.

An autoencoder is a neural network that learns to compress data into a smaller representation and then reconstruct it back, like a lossy zip file for images or sounds.

Underfitting happens when a model is too simple to learn the patterns in the data, leading to poor performance on both training and new data.

Zero-shot learning is a machine learning technique where a model can recognize objects or concepts it has never seen during training, by leveraging auxiliary information like descriptions or attributes.

Temperature controls how creative or random an AI's responses are.

An activation function is a mathematical gate that decides whether a neuron should 'fire' or stay silent, enabling neural networks to learn complex patterns.

An AI agent is a software program that can perceive its environment, make decisions, and take actions to achieve specific goals on its own.

An API is a messenger that lets two software programs talk to each other.

Bayes Theorem updates the probability of a hypothesis based on new evidence.

Beam Search is a heuristic search algorithm that explores a graph by expanding the most promising nodes in a limited set, balancing exploration and efficiency.

Bias-Variance Tradeoff is the balancing act between a model's error from oversimplification (bias) and its error from being too sensitive to training data (variance).

Big O Notation describes how the runtime or space requirements of an algorithm grow as the input size increases.

A confusion matrix is a table that shows how well a classification model performs by comparing its predictions to the actual correct answers.

Correlation measures how two variables move together, but doesn't mean one causes the other.

Cross-validation is a method to evaluate how well a machine learning model generalizes to new data by splitting the data into multiple subsets and testing the model on each subset.

A decision tree is a flowchart-like model that makes predictions by splitting data into branches based on yes/no questions.

Ensemble Learning combines multiple models to make better predictions than any single model alone.

The F1 Score is a single metric that balances precision and recall to measure a model's accuracy, especially when classes are imbalanced.

Feature engineering is the process of transforming raw data into inputs that make machine learning models work better.

Flash Attention is a faster, memory-efficient algorithm for computing attention in transformers that avoids storing large attention matrices.

Gradient Boosting builds a strong predictive model by sequentially adding weak models, each correcting the errors of the previous one.

A gradient is the direction and rate of steepest increase of a function, used in AI to guide learning.

A hash table is a data structure that stores key-value pairs and uses a hash function to compute an index, enabling fast lookups, insertions, and deletions in average constant time.

Hyperparameters are settings you choose before training an AI model that control how it learns, like the knobs on a stove that you set before cooking.

Idempotency means you can do the same operation over and over, but the result is always the same as doing it once.

K-Means Clustering is an algorithm that automatically groups similar data points into clusters based on their features.

LoRA is a technique that fine-tunes large AI models efficiently by training only small, low-rank matrices instead of the entire model.

Model Context Protocol is a standard way for AI models to securely access and share context from external tools and data sources.

The normal distribution is a bell-shaped curve that shows how data tends to cluster around an average, with most values near the middle and fewer at the extremes.

Normalization organizes data in a database to reduce redundancy and improve integrity.

A p-value tells you how likely your data would be if there were no real effect, helping you decide if your results are statistically significant.

Precision measures how many of your positive predictions were correct, while Recall measures how many of the actual positives you caught.

Principal Component Analysis is a technique that reduces the number of variables in a dataset while keeping as much of the original information as possible.

Random Forest is an AI method that combines many decision trees to make more accurate predictions by averaging their results.

Recursion is a programming technique where a function calls itself to solve a problem by breaking it into smaller, similar subproblems.

Regression finds the best-fit line or curve that predicts a numeric outcome from one or more input variables.

ReLU is a simple activation function that outputs the input directly if it's positive, and zero otherwise, helping neural networks learn efficiently.

The sigmoid function squashes any input number into a value between 0 and 1, often used to turn scores into probabilities.

Standard deviation measures how spread out numbers are from the average.

Standardization rescales data to have a mean of 0 and a standard deviation of 1, making different features comparable.

A vector is a mathematical object that has both magnitude (length) and direction, often visualized as an arrow.

Vision Transformer (ViT) applies the Transformer architecture, originally for language, directly to image patches for classification.