AI Reference

A mathematical function applied to the output of a neuron in a neural network. It introduces non-linearity, enabling the network to learn complex patterns. Common examples include ReLU, sigmoid, and tanh.

A step-by-step set of instructions or rules designed to solve a specific problem or perform a computation. In AI, algorithms drive how models learn from data and make predictions.

A hypothetical form of AI that possesses the ability to understand, learn, and apply intelligence across a wide range of tasks at a level comparable to a human being, rather than being limited to a specific domain.

The broad field of computer science focused on creating systems capable of performing tasks that typically require human intelligence, such as learning, reasoning, problem-solving, perception, and language understanding.

A technique in neural networks that allows a model to focus on the most relevant parts of an input sequence when producing an output. It is the core innovation behind the Transformer architecture.

The primary algorithm used to train neural networks. It calculates the gradient of the loss function with respect to each weight by propagating errors backward through the network, allowing weights to be updated to minimize error.

In machine learning, bias can refer to (1) a parameter in a neuron that shifts the activation function, or (2) systematic errors in a model's predictions caused by flawed assumptions, unrepresentative training data, or prejudiced labeling.

A software application designed to simulate conversation with human users, especially over the internet. Modern chatbots are often powered by large language models (LLMs) and can handle complex, multi-turn dialogues.

A type of supervised learning task where the model predicts which category or class an input belongs to. Examples include spam detection (spam vs. not spam) and image recognition (cat vs. dog).

A field of AI that enables computers to interpret and understand visual information from the world, such as images and videos. Applications include facial recognition, object detection, and medical image analysis.

A type of deep neural network specifically designed for processing structured grid data like images. CNNs use convolutional layers to automatically detect features such as edges, textures, and shapes.

A technique used to increase the size and diversity of a training dataset by applying transformations like rotation, flipping, cropping, or color adjustment to existing data, helping to reduce overfitting.

A structured collection of data used to train, validate, and test machine learning models. Datasets can contain text, images, numbers, audio, or any other type of information relevant to the learning task.

A subset of machine learning that uses neural networks with many layers (hence 'deep') to model complex patterns in large amounts of data. It powers most modern AI breakthroughs in vision, language, and more.

A class of generative AI models that learn to create data (often images) by gradually removing noise from a random signal. Models like Stable Diffusion and DALL-E use this approach to generate high-quality images from text prompts.

A dense vector representation of data (such as words, sentences, or images) in a continuous vector space. Embeddings capture semantic meaning, allowing models to understand relationships between items.

One complete pass through the entire training dataset during the training of a machine learning model. Models are typically trained over multiple epochs to improve performance.

An individual measurable property or characteristic of the data being observed. In a dataset of houses, features might include square footage, number of bedrooms, and location.

The process of using domain knowledge to select, transform, or create new input variables (features) from raw data to improve the performance of a machine learning model.

The process of taking a pre-trained model and further training it on a smaller, task-specific dataset. This allows the model to adapt its general knowledge to perform well on a particular task.

A framework consisting of two neural networks—a generator and a discriminator—that compete against each other. The generator creates synthetic data, while the discriminator tries to distinguish it from real data, driving both to improve.

A category of AI systems designed to create new content—such as text, images, music, code, or video—rather than simply analyzing or classifying existing data. Examples include ChatGPT, DALL-E, and Midjourney.

A specialized processor originally designed for rendering graphics but now widely used to accelerate the training and inference of deep learning models due to its ability to perform many parallel computations.

An optimization algorithm used to minimize the loss function of a model by iteratively adjusting parameters in the direction of steepest descent. It is the fundamental optimization method behind most neural network training.

A phenomenon where an AI model, particularly a large language model, generates information that sounds plausible but is factually incorrect or fabricated. This is a known challenge with current generative AI systems.

A configuration setting used to control the training process of a model, set before training begins. Examples include learning rate, batch size, and number of layers. Unlike parameters, these are not learned from data.

The process of using a trained machine learning model to make predictions or generate outputs on new, unseen data. This is the deployment phase, as opposed to the training phase.

A type of AI model trained on massive amounts of text data that can understand, generate, and manipulate human language. Examples include GPT-4, Claude, Gemini, and LLaMA.

A mathematical function that measures how far a model's predictions are from the actual target values. The goal of training is to minimize this function. Common examples include cross-entropy loss and mean squared error.

A subset of AI that focuses on building systems that learn from data to improve their performance on a task without being explicitly programmed. The three main types are supervised, unsupervised, and reinforcement learning.

In AI, a model is a mathematical representation learned from data that can make predictions or decisions. It consists of an architecture (structure) and parameters (weights) tuned during training.

A branch of AI focused on enabling computers to understand, interpret, and generate human language. NLP powers applications like chatbots, translation, sentiment analysis, and text summarization.

A computing system inspired by the biological neural networks in the human brain. It consists of interconnected nodes (neurons) organized in layers that process information and learn patterns from data.

A condition where a machine learning model learns the training data too well—including its noise and outliers—resulting in poor performance on new, unseen data. Regularization and more data can help prevent it.

A value inside a model that is learned from training data, such as the weights and biases in a neural network. Large language models can have billions or even trillions of parameters.

The input text or instruction given to a generative AI model to elicit a desired response. Prompt design (or 'prompt engineering') significantly affects the quality and relevance of the model's output.

The practice of designing and refining input prompts to guide AI models toward producing more accurate, relevant, or useful outputs. It is a key skill for effectively working with LLMs.

A type of neural network designed for sequential data, where the output from previous steps is fed back into the network. RNNs are used for tasks like language modeling and time-series prediction, though Transformers have largely replaced them.

A type of supervised learning task where the model predicts a continuous numerical value rather than a category. Examples include predicting house prices, stock values, or temperature.

A type of machine learning where an agent learns to make decisions by taking actions in an environment and receiving rewards or penalties. It is used in game playing, robotics, and training LLMs via RLHF.

A training technique where a model is fine-tuned using feedback from human evaluators to better align its outputs with human preferences and values. It was a key technique in training ChatGPT.

A technique that enhances LLM responses by first retrieving relevant information from an external knowledge base and then using it as context for generation. This helps reduce hallucinations and keep responses current.

A machine learning approach where the model is trained on labeled data—input-output pairs where the correct answer is provided. The model learns to map inputs to outputs and can then predict labels for new data.

A multi-dimensional array of numbers that serves as the fundamental data structure in deep learning frameworks like TensorFlow and PyTorch. Scalars, vectors, and matrices are all special cases of tensors.

The basic unit of text that a language model processes. A token can be a word, part of a word, or a character, depending on the tokenization method. For example, 'unhappiness' might be split into 'un', 'happiness'.

The process of feeding data to a machine learning algorithm so it can learn patterns and adjust its internal parameters. Training involves iterating over data, computing loss, and updating weights via backpropagation.

A technique where a model trained on one task is repurposed as the starting point for a model on a different but related task. This dramatically reduces training time and data requirements.

A neural network architecture introduced in the 2017 paper 'Attention Is All You Need.' It relies entirely on attention mechanisms and has become the foundation for most modern LLMs and many computer vision models.

A condition where a model is too simple to capture the underlying patterns in the data, resulting in poor performance on both training and test data. It can be addressed by using a more complex model or training longer.

A machine learning approach where the model is trained on unlabeled data and must discover hidden patterns or structures on its own. Common techniques include clustering, dimensionality reduction, and anomaly detection.

A type of generative model that learns to encode data into a latent space and then decode it back, allowing it to generate new data samples. VAEs are used in image generation, anomaly detection, and drug discovery.

A numerical value in a neural network that determines the strength of the connection between neurons. Weights are adjusted during training to minimize prediction errors and are the core learned parameters of a model.

The ability of a model to perform a task it was not explicitly trained on, by leveraging its general knowledge. For example, an LLM can classify sentiment without specific sentiment training by understanding the concept from its pre-training data.