AI Glossary A-F
Artificial Intelligence (AI) is rapidly transforming the way we live, work, and interact with the world. From enhancing everyday conveniences to driving complex innovations, AI's influence is pervasive and profound. As this technology continues to evolve, understanding its foundational language is crucial for anyone aiming to keep pace with its developments. This glossary is designed to demystify the key terms and concepts that underpin AI, providing a valuable resource for students, professionals, and enthusiasts alike. By familiarizing yourself with these terms, you'll be better equipped to engage with AI's growing role in various sectors.
As AI becomes increasingly integral to industries ranging from healthcare to finance, the ability to speak its language is more important than ever. Business leaders, developers, and innovators must all be able to articulate their visions and strategies within the context of AI's potential. This glossary not only aids in comprehension but also empowers you to participate in meaningful discussions about AI's applications and implications. By mastering these terms, you position yourself to leverage AI's transformative power effectively within your organization or personal projects.
Finally, as AI continues to shape the future, it is essential for everyone—not just tech experts—to be part of the conversation. This glossary aims to democratize access to AI knowledge, making it accessible to a wider audience. Whether you're a seasoned professional or a curious newcomer, understanding AI's vocabulary is a step toward harnessing its potential and contributing to its evolution. Dive into this glossary and embark on a journey of discovery that will illuminate the complexities and possibilities of artificial intelligence, equipping you to navigate and thrive in an AI-driven world.
3D Autoencoders
A specialized two-part neural network consisting of an encoder and a decoder. The encoder compresses data into a smaller representation, while the decoder reconstructs the original data from this representation.
3D GAN
An architectural framework within Generative Adversarial Networks (GANs) designed for generating three-dimensional shapes.
Activation Functions
Mathematical functions introducing non-linearity in neural networks, determining node outputs and facilitating complex computations.
Active Learning
A form of reinforcement learning where an algorithm actively interacts with users to refine performance by obtaining data labels.
Actor-Critic Model
A reinforcement learning structure with an "Actor" for action determination and a "Critic" for evaluating state-action quality.
Adapter
A framework facilitating transfer learning by integrating new layers into existing models for task adaptability without starting from scratch.
Adaptive Boosting (AdaBoost)
A boosting technique enhancing model accuracy by focusing on difficult-to-classify data points.
Adaptive Control Systems
Systems that automatically adjust control strategies based on changing conditions and feedback.
Adaptive learning
An educational method customizing content to individual student needs, optimizing personalized learning experiences.
Advanced Natural Language Processing
Techniques enhancing computer understanding and generation of human language, improving interaction and comprehension.
Advantage Actor Critic (A2C)
A reinforcement learning algorithm combining policy gradient and value function for refined learning via actor-critic components.
Adversarial Attack
An attempt to compromise a machine learning model using deceptive input data to induce errors or manipulate outputs.
Adversarial Machine Learning
A study area focusing on making machine learning models robust against deceptive or adversarial inputs.
Adversarial Training for Robustness
Training models with adversarial examples to enhance robustness against malicious inputs.
Affective Computing
Computing that recognizes, interprets, and processes human emotions, enabling empathetic human-computer interaction.
Affinity Propagation Clustering
A clustering technique determining optimal data grouping by passing messages between data points.
AI (Artificial Intelligence)
The development of systems capable of performing tasks requiring human intelligence, such as perception and decision-making.
AI Agent
An AI system designed for autonomous perception, decision-making, and action to achieve specific objectives.
AI algorithm
Programming instructions enabling machines to function independently, process data, and learn from it.
AI Explainability
Understanding and interpreting AI system decisions, promoting transparency and trust in complex models.
AI Governance
Policies and frameworks ensuring responsible AI development and deployment, aligning with ethical and legal standards.
AI Hallucination
Instances where AI generates incorrect or nonsensical outputs, often due to overgeneralization or data ambiguity.
AI Hardware
Components and technologies optimizing AI performance in various applications.
AI Model
A program trained on specific data to accomplish defined tasks, central to AI systems.
AI safety
Efforts to ensure AI systems operate safely, reliably, and in alignment with human values.
AI System
The infrastructure supporting AI model development and deployment, including data, hardware, and interfaces.
AI Writer
Software using AI for generating human-like written content, useful in content marketing.
ALBERT
A cloud-centric AI platform integrating digital marketing technologies.
Algorithm
A set of instructions guiding machines in task execution and decision-making.
Algorithmic Bias Detection
Methods for identifying and addressing bias in AI systems.
Algorithmic Fairness in AI
Efforts to ensure AI systems operate fairly and without bias.
AlphaGo Alpha Zero
An AI program by Google DeepMind playing the strategy game Go, showcasing narrow AI potential.
ANN Benchmarks
Evaluation tools for comparing artificial neural network performance.
Annotation
Labeling data for training AI models, enhancing understanding of patterns and contexts.
Anomaly Detection
Identifying unusual patterns in data to prevent anomalies like cyber attacks.
Anthropomorphism
Attributing human traits to AI systems, often leading to misconceptions about their capabilities.
Area Under the Curve (AUC)
A metric evaluating model performance by measuring the area under the ROC curve.
Artificial General Intelligence
AI with human-level intelligence, capable of broad task execution across contexts.
Artificial Narrow Intelligence (ANI)
AI designed for specific tasks, effective within predefined constraints.
Artificial Neural Network (ANN)
A machine learning element mimicking brain neural connections, foundational to deep learning.
Artificial Super Intelligence (ASI)
AI surpassing human intelligence, capable of performing tasks beyond human comprehension.
Association Rule Learning
A method for discovering relationships between variables in large datasets.
Associative Learning
Learning encoding data relationships, like Pavlov's Dog associating stimuli with responses.
Asynchronous Advantage Actor Critic (A3C)
A reinforcement learning algorithm leveraging policy and value functions for advanced learning.
Attention Map
A visual highlighting image sections relevant to specific target classes, offering insights into deep networks.
Attention Mechanisms
Model components prioritizing data segments during predictions, mimicking human cognitive focus.
Augmented Intelligence
Combining human and machine intelligence for optimal decision-making and data handling.
Augmented Reality Applications
Applications enhancing real-world experiences with computer-generated perceptual information.
Auto-associative Network
A neural network mapping input and output data for data reconstruction from partial patterns.
Autoencoder
A neural network variant for unsupervised learning, encoding and decoding data for efficient representation.
Autonomous Agents
Software components executing tasks in complex environments, making decisions to achieve goals.
AutoRegressive Integrated Moving Average (ARIMA)
A statistical model for time series forecasting using historical data patterns.
Backpropagation
An algorithm for refining neural networks by computing weight adjustments based on error gradients.
Batch
A set of examples used in one gradient update during model training.
Batch Gradient Descent
An optimization technique in machine learning known for its accuracy and stability.
Batch Size
The number of samples processed in a single iteration of model training, influencing optimization pace.
Bayes’s Theorem
A theorem calculating event probabilities based on prior knowledge of related conditions.
Bayesian Deep Learning
Combining Bayesian statistics with deep learning for improved predictions and decision-making.
Bayesian Inference Methods
Methods using Bayesian statistics for probabilistic inference and decision-making.
Bayesian Machine Learning
A methodology combining Bayesian statistics with machine learning for accurate predictions.
Bayesian Networks
Probabilistic models representing variable dependencies and causal relationships.
Behavioral Cloning in Reinforcement Learning
Techniques using imitation learning to replicate expert behavior in reinforcement learning.
BEIR
Benchmarking tool for evaluating Information Retrieval Systems across tasks and data types.
Bellman Equation
A dynamic programming equation facilitating optimal decision-making in sequential contexts.
Bias
Systematic errors or deviations in AI models, impacting accuracy and fairness.
Bias-Variance Tradeoff
A conflict in minimizing bias and variance, affecting model generalization.
Biased Random Walks in Graphs
Graph traversal methods using randomness to explore node connections.
Bidirectional Encoder Representations from Transformers (BERT)
A deep learning approach for natural language processing, capturing contextual text nuances.
Black box AI
AI models with non-transparent decision-making processes, raising trust and accountability concerns.
BLEU Score
A metric evaluating machine-translated text similarity to reference translations.
Boosting
A method reducing bias and variance in supervised learning, improving model accuracy.
Bootstrap aggregating
A method aggregating model versions trained on random data subsets for stability and accuracy.
Breadth-first Search
A graph search method exploring nodes level by level for goal discovery.
Breeding System
Generating new AI models by combining existing models or algorithms for improvement.
Brittleness
An AI system's tendency to fail on unfamiliar data, highlighting generalization challenges.
CAP Theorem
The CAP theorem, also known as Brewer's CAP theorem after computer scientist Eric Brewer, is a fundamental principle in theoretical computer science that profoundly influences the design and operation of distributed systems. It asserts that any distributed data store can provide only two guarantees: Consistency, Availability, and Partition Tolerance (often abbreviated as 'CAP'). This theorem is a guiding principle for developers and system architects, shaping decisions around data management and system architecture.
Capsule Network
A form of artificial neural networks that model intricate hierarchical relationships. Drawing inspiration from biological neural organization, they aim to emulate more closely the structure of human neural connections.
Catastrophic Forgetting
A problem that occurs when two similar game states yield dramatically divergent outcomes, causing confusion in the Q-function's learning process.
Causal Inference Models
Causal inference models are frameworks that help to understand the relationships between causes and effects using statistical data.
Chain of Thoughts
Chain-of-thought (CoT) is a prompt engineering technique that helps large Large Language Models (LLMs) break down complex problems into smaller, easier steps. Instead of giving a direct answer, the model walks through a sequence of thoughts or reasoning, like how a person might solve a problem step by step.
Chat Bot
A chatbot is a software application designed to simulate conversation with human users, especially over the internet. It utilizes techniques from the field of natural language processing (NLP) and sometimes machine learning (ML) to understand and respond to user queries. Chatbots can range from simple, rule-based systems that respond to specific keywords or phrases with pre-defined responses, to more sophisticated AI-driven bots capable of handling complex, nuanced, and context-dependent conversations.
Chunking
Chunking is the process of dividing a document into smaller, more manageable parts for efficient computational processing. A chunk can be characterized by its size (number of characters) or identified by analyzing the chunked text for natural segments, such as complete sentences or paragraphs.
Chunking vs Sections
In the context of Retrieval-Augmented Generation (RAG), the terms "content chunk" and "content section" are often used interchangeably to refer to small, manageable divisions of text for efficient computational processing and analysis.
Classification
The act of an AI model identifying patterns in the input data and then using those patterns to predict the category (or label) for new, unseen data points; it can be thought of as categorizing data. For example, an AI model trained on images of different types of animals (labeled as "cat," "dog," "bird") could then classify a new image and predict whether it shows a cat, dog, or bird. Classification is used heavily in e-commerce and powers many tagging systems, which categorize and organize products using keywords or labels. Classification is a supervised machine learning approach that categorizes data into predefined classes. Given an input, a classification model predicts the category or label the input belongs to. It's one of the most common tasks in machine learning and is used in many real-world applications, from email spam detection to medical diagnoses. The task of approximating a mapping function from input variables to discrete output variables, or, by extension, a class of Machine Learning algorithms that determine the classes to which specific instances belong.
Classification model
A type of model used in machine learning that is designed to take input data and sort it into different categories or classes. Sometimes referred to as classifiers.
Clustering
The act of an AI model identifying groupings and patterns in data without a human explicitly defining the criteria. Recommendation engines, such as Amazon and Netflix, commonly leverage clustering to provide personalized product and movie recommendations to users. Clustering allows us to identify natural groupings in data without needing a predefined target variable. It's like letting the data speak for itself. An unsupervised machine learning method in which patterns in the data are identified and evaluated, and data points are grouped accordingly into clusters based on their similarity. Sometimes referred to as clustering algorithms. In Machine Learning, the unsupervised task of grouping a set of objects so that objects within the same group (called a cluster) are more "similar" to each other than they are to those in other groups.
Cognitive Architectures
Cognitive architecture is a theory about the structure of the human mind and a computational model of that theory.
Cognitive Maps
In AI, a cognitive map is a structured model representing an environment or knowledge. It enables systems, particularly in robotics and autonomous technologies, to simulate real-world settings, predict outcomes, and make decisions based on understanding spatial relationships, much like humans do with their mental maps.
Cognitive Modeling
Cognitive modeling is a computer science field that simulates how people think and solve problems.
Cold-Start
A potential issue arising from the fact that a system cannot infer anything for users or items for which it has not gathered a sufficient amount of information yet.
Command Query Responsibility Segregation (CQRS)
Command Query Responsibility Segregation (CQRS) is a design pattern that separates the actions that change data (commands) from those that retrieve data (queries). Instead of using the same model for both, CQRS splits them into two distinct paths. By separating commands and queries, systems can handle each task more efficiently, leading to better performance, clearer code, and easier scaling as your application grows.
Completions
Completions are the output produced by AI in response to a given input or prompt. When a user inputs a prompt, the AI model processes it and generates text that logically follows or completes the given input. These completions are based on the patterns, structures, and information the model has learned during its training phase on vast datasets.
Complex Event Processing
Complex Event Processing (CEP) is a method for quickly analyzing and responding to streams of data in real time. Instead of processing data one piece at a time, CEP helps identify patterns and relationships between multiple events that occur within a short time period. It's like watching a series of events unfold and making quick decisions based on what's happening.
Computational Creativity
A multidisciplinary domain that aims to replicate human creativity through computational methods by combining approaches from fields including artificial intelligence, philosophy, cognitive psychology, and the arts.
Computer Vision
A subfield of artificial intelligence (AI) and computer science that focuses on enabling computers to interpret, understand, and process visual information from the world. It has a broad range of applications including object detection, image classification, facial recognition, and autonomous vehicles. The act of an AI model analyzing and interpreting visual data (e.g., images or videos) from cameras or sensors. AI models can leverage computer vision to identify defects in products using visual data, enabling efficient quality control and waste reduction in manufacturing. A field of AI that uses computers to process and analyze images, videos and other visual inputs. Common applications of computer vision include facial recognition, object recognition and medical imaging. A field of AI that allows computers to scan visual information such as images and video, identifying and classifying objects and people. The systems can react to what they see and take or recommend a particular action. The technology is being used to track wildlife for conservation and guide autonomous vehicles. There's been concern about its use in military operations and policing, where it's been shown to exhibit racial bias and to lack the precision needed to reliably identify a particular person. The field of Machine Learning that studies how to gain high-level understanding from images or videos.
Confidence Interval
A type of interval estimate that is likely to contain the true value of an unknown population parameter. The interval is associated with a confidence level that quantifies the level of confidence of this parameter being in the interval.
Conformity assessment
An analysis, often performed by an entity independent of a model developer, on an AI system to determine whether requirements, such as establishing a risk management system, data governance, record-keeping, transparency and cybersecurity practices have been met.
Confusion Matrix
A pivotal tool for evaluating model performance by identifying misclassified objects, providing insights into model accuracy.
Constitutional AI
A way of training an AI model so that it self-critiques and revises its responses to align with a predefined set of rules or principles (i.e., constitution) based on human principles such as avoiding harm, respecting preferences, and providing accurate information. The goal is for the model to be harmless and to self improve without the need for humans to label/identify harmful outputs. Anthropic's Claude is an example of a large language model (LLM) that's powered by constitutional AI.
Content Quality Filters
Content quality filters are mechanisms, criterias and tools to assess the quality of content and ensure minimum quality standards within a collection of content. Content quality filters can be used to filter out irrelevant, inappropriate or low quality content as part of content moderation for search engines and knowledge bases.
Contestability
The principle of ensuring AI systems and their decision-making processes can be questioned or challenged by humans. This ability to contest or challenge the outcomes, outputs and actions of AI systems depends on transparency and helps promote accountability within AI governance. Also called redress.
Context Length
Context length refers to the maximum number of tokens that can be processed by a Large Language Model (LLM) or other text processing model.
Context Window
A specified range of tokens (e.g., words) surrounding a target token that defines the scope of information considered when processing or analyzing individual elements within a sequence of data. In natural language processing (NLP), you can think of it as a window through which a computer looks to understand the meaning of words by considering the words nearby. If the window is big, it sees more words at once — helping it understand more complex sentences and larger pieces of information. The context window is a fundamental concept in AI, particularly in large language models (LLMs). It refers to the maximum amount of text, measured in tokens, an AI model can remember and use during a conversation when generating a response.
Contextual Bandit
An extension of the multi-armed bandit approach that considers contextual information to optimize decision-making in multi-action scenarios.
Continuous Learning
Continuous learning, in the context of artificial intelligence (AI), refers to the ability of a system to learn and adapt to new information and changes in its environment over time.
Continuous Profiling
Continuous profiling is a monitoring method that collects and analyzes relevant metrics regarding a system's or program's operations. For instance, profiling can inform engineers about CPU usage, memory utilization, frequency and duration of function calls, and I/O activity.
Convergence
Convergence in machine learning refers to the state during training where a model's loss stabilizes within a certain error range around its final value. It signifies that further training will not significantly enhance the model's performance.
Conversational AI
AI technologies like chatbots that engage in human-like conversations with users. Conversational AI refers to the technology that empowers machines, like chatbots, virtual assistants, and similar speech-based apps, to hold human-like conversations. Conversational AI is a specialized branch of artificial intelligence that enables computers to conduct conversations with humans in a way that feels natural and intuitive. This technology allows machines to engage in dialogues by interpreting and responding to spoken or written language, effectively mimicking human conversation through either text or voice. Conversational AI is a branch of artificial intelligence (AI) that focuses on enabling computers to engage in natural and human-like conversations with users. Conversational AI allows us to use everyday language when interacting with artificial intelligence. Using technologies like natural language processing, machine learning, and speech recognition, AI can understand questions and instructions, which helps it provide better responses. Engaging with AI becomes more natural and effortless, requiring no special training. In the past, you'd have to put awkwardly-worded terms into a search engine to find what you were looking for. With conversational AI, you can simply state the request as you would with another person.
Convolutional Neural Network
A Convolutional Neural Network (CNN) is a deep-learning model tailored for visual data like images, videos, and sometimes even audio files.
Corpus
The full set of data utilized to train an artificial intelligence model. It reviews this data to learn about a specific domain. A large collection of texts or data that a computer uses to find patterns, make predictions or generate specific outcomes. The corpus may include structured or unstructured data and cover a specific topic or a variety of topics.
Cross-Domain Transfer Learning
Cross-domain transfer learning (CDTL) is a machine learning technique that uses a model trained on one task to learn a related task in another domain.
Cross-Lingual Language Models
cross-lingual language model (XLM) is a type of artificial intelligence (AI) that can understand, interpret, and generate text in multiple languages.
Curse of Dimensionality in Machine Learning
The "curse of dimensionality" is a challenge in machine learning that occurs when we have too many characteristics (or "dimensions") to consider.
CycleGAN
An image-to-image translation methodology using unpaired datasets to learn mappings between the input and output images.
DALL-E
DALL-E is a multimodal model developed by OpenAI to create images from text prompts. It takes a simple written prompt, like "a cat wearing a superhero cape, flying through a city skyline at sunset," and turns it into a unique, visually creative image. DALL-E uses advanced deep-learning techniques to understand the meaning behind words and create matching visuals, even for imaginative or abstract ideas.
Data Annotation
The process of labeling and annotating data to facilitate supervised learning, enhancing a model's understanding of the inputs.
Data Augmentation
A technique where users artificially enrich the training set by adding modified copies of the same data. Data augmentation involves the skillful manipulation and expansion of your existing data. This practice is a cornerstone in machine learning and AI, as it amplifies the volume and diversity of training data for a model.
Data Bias
Data bias refers to the presence of systematic and unfair inaccuracies or prejudices in data that can lead to incorrect or discriminatory outcomes when used in AI models.
Data Catalog
A data catalog is a searchable inventory of available data assets, sources, datasets, and associated metadata.
Data Classification Techniques
Data classification techniques are methods for organizing data into categories to make it easier to find and use. These techniques can help with security, compliance, and risk management.
Data Imbalance
An uneven distribution of classes within a dataset. This can challenge model performance and accuracy.
Data Labeling
Data labeling is the process of assigning labels or tags to data in order to enrich that data. Data labeling is used to train neural networks and evaluate existing AI systems, such as for instance spam detection or sentiment analysis.
Data Modeling
Data modeling creates a blueprint representing an application's or a system's data structure. The data model is a diagram illustrating the relevant data entities, objects, relationships, and complex schemas for storage.
Data Pipeline
A data pipeline is a set of processes and tools for collecting, transforming, transporting, and enriching data from various sources. Data pipelines control the flow of data from source through transformation and processing components to the data's final storage location.
Data Poisoning
A type of adversarial attack involving the manipulation of training data by deliberately introducing contaminated samples to skew model behavior and outputs. An adversarial attack in which a malicious user injects false data into a model to manipulate the training process, thereby corrupting the learning algorithm. The goal is to introduce intentional errors into the training dataset, leading to compromised performance and resulting in undesired, misleading or harmful outputs.
Data provenance
A process that tracks and logs the history and origin of records in a dataset, encompassing the entire life cycle from its creation and collection to its transformation to its current state. It includes information about sources, processes, actors and methods used to ensure data integrity and quality. Data provenance is essential for data transparency and governance, and it promotes better understanding of the data and eventually the entire AI system.
Data Sanitization
Data sanitization is the process of ensuring that sensitive information is removed or masked in datasets to protect privacy and comply with data protection regulations. Sanitizing data involves identifying and appropriately handling personally identifiable information (PII) or other sensitive data to prevent unauthorized access or disclosure.
Data Warehouse
A data warehouse (or data lake) is a centralized repository for the storage of vast amounts of raw, unprocessed data. It is a flexible, scalable infrastructure that facilitates analytics, data processing, and data exploration across an organization.
Dataset
A dataset is a structured collection of data, usually with some sort of schema, that can be used for analysis, research, and large language or machine model training and evaluation.
Decision tree
A type of supervised learning model used in machine learning that represents decisions and their potential consequences in a branching structure. A category of Supervised Machine Learning algorithms where the data is iteratively split in respect to a given parameter or criteria.
Deep Blue
A chess-playing computer developed by IBM, better known for being the first computer chess-playing system to win both a chess game and a chess match against a reigning world champion under regular time controls.
Deep Deterministic Policy Gradient (DDPG)
A reinforcement learning algorithm employing deep neural networks to learn optimal policies in continuous action spaces to maximize the expected long-term reward.
Deep Learning
A machine learning approach that trains AI models using multi-layered neural networks (which simulate the decision-making capabilities of the human brain) in order to identify patterns and relationships and make predictions with high accuracy. Deep learning is the brain behind the AI revolution. It's a subset of the machine learning system that aims to mimic the human brain's structure, using artificial neural networks with multiple layers to process vast amounts of data. The application of Neural Networks to complex problems. The term 'Deep' refers to the multiple layers of hidden nodes that are used to form very large neural networks. A subfield of AI and machine learning that uses artificial neural networks. Deep learning is especially useful in fields where raw data needs to be processed, like image recognition, natural language processing and speech recognition. Deep Learning is a subfield of Machine Learning (ML) that focuses on developing neural networks with many layers, enabling them to learn complex patterns and representations from large amounts of data. Deep learning is an advanced form of AI that helps computers become really good at recognising complex patterns in data. It mimics the way our brain works by using what's called layered neural networks, where each layer is a pattern that then lets you make predictions based on the patterns you've learned before. It's really useful for things like image recognition, speech processing, and natural-language understanding.
Deep Q-Learning
A pivotal approach in reinforcement learning that employs deep neural networks to approximate the Q-function, which it uses to determine the optimal course of action.
Deep Q-Network (DQN)
A framework employing deep neural networks for Q-learning in reinforcement learning tasks.
Deep Reinforcement Learning
Deep Reinforcement Learning (DRL) is a field of AI where an agent learns how to make decisions by interacting with an environment and improving over time based on feedback or "rewards." DRL combines reinforcement learning (RL), a learning method through trial and error, with deep learning, which enables the agent to handle complex data like images or sensor readings. DRL can use deep neural networks to teach agents to navigate complicated tasks with high-dimensional inputs. It's widely used in applications like robotics and game playing, where traditional learning methods struggle due to the complexity and variability of the environment.
Deepfakes
Audio or visual content that has been altered or manipulated using artificial intelligence techniques. Deepfakes can be used to spread misinformation and disinformation.
Depth Estimation
The task of predicting the depth of objects within an image. This is essential for various computer vision applications like self-driving vehicles.
Depth-first Search
A way of searching a graph for a goal (answer) by assessing each node through the same branch (parent-child nodes), before searching the nodes at the next branch.
Differentiable Neural Computers
Advanced and typically recurrent neural network architectures enhanced with memory modules for complex learning and reasoning tasks.
Diffusion Model
A type of generative AI in which the AI model aims to learn the underlying structure of a dataset — as well as patterns and relationships between different pieces of information — by looking at how information or details move around using three main components: 1. Forward process: The model takes an image or sound and applies a series of steps to it, where each step adds a little bit of noise to the image or sound — distorting it more over time 2. Reverse process: The program then tries to guess what the original image or sound was by reversing the same process used to distort it 3. Sampling procedure: By repeating this process, the model improves in its ability to guess the original image or sound. This is currently the most popular option for image and video generation. The powerful image generator, Midjourney, operates via a combination of a diffusion model and large language model (LLM). A generative model used in image generation that works by iteratively refining a noise signal to transform it into a realistic image when prompted.
Digital Twins
A digital twin is a virtual model representing a physical system or process that helps users understand its behavior in various situations. The model gets real-time data from multiple sources associated with the system’s physical environment.
Dimensionality Reduction
A technique for simplifying datasets by reducing their feature dimensions while preserving critical information. Dimensionality reduction is a process used in data science and machine learning to reduce the number of variables, or "dimensions," in a dataset while retaining as much relevant information as possible. This reduction simplifies data analysis, visualization, and processing, especially in high-dimensional datasets. Techniques like Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE) identify patterns and relationships within data, projecting it onto fewer dimensions. By discarding less significant features, dimensionality reduction helps improve computational efficiency and mitigates overfitting, making it essential for managing complex data, particularly in fields like image and text analysis.
Discriminative model
A type of model used in machine learning that directly maps input features to class labels and analyzes for patterns that can help distinguish between different classes. It is often used for text classification tasks, like identifying the language of a piece of text or detecting spam. Examples are traditional neural networks, decision trees and random forest.
Discriminator
In a Generative Adversarial Network (GAN), the discriminator is like a detective. When it’s shown pictures (or other data), it has to guess which are real and which are fake. The “real” pictures are from a dataset, while the “fake” ones are created by the other part of the GAN, called the generator. The discriminator’s job is to get better at telling real from fake, while the generator tries to get better at creating fakes. This is the software version of continuously building a better mousetrap.
Disinformation
Audio or visual content that is intentionally manipulated or created to cause harm. Disinformation can spread through deepfakes created by those who have malicious intentions.
Document Classification
Document classification is the process of categorizing documents based on their content or attributes.
Double DQN
A DQN technique that uses double Q-learning to mitigate overestimation biases and improve Q-value approximations.
Dropout
A regularization technique involving the temporary exclusion of randomly selected nodes during neural network training.
Dynamic Time Warping (DTW)
A method for measuring similarity between time series data, commonly used in time series analysis and pattern recognition.
Edge Case
Data that falls just over the outer limits of the data used to train a machine learning system, thereby making it difficult for the system to identify and/or act upon. As an example, autonomous vehicles can be trained to recognise animals it may encounter, e.g., dogs, cats, horses – an ‘edge case’ could be a kangaroo, because this animal is unlikely to feature in the training data. Within AI, edge cases can be one of the main delays in moving a system from the lab to the real-world.
Edge Learning
A decentralized approach to machine learning where processing occurs on user devices, enhancing privacy and efficiency. This required model compression because of the complexity of AI programs.
Embedding Layer
An embedding layer converts complex data into numerical vectors that can be processed by neural networks. This post explains what embedding layers are, how they work and why they are so important in machine learning.
Embeddings
Embeddings are numerical representations of content, such as text, images or video, that are encoded with semantic relationships in a dense multi-dimensional space. Embeddings are one example of vectorization.
Emergent Behaviors
As large language models reach a certain scale, they sometimes start to display abilities that appear to have emerged from nowhere, in the sense that they were neither intended nor expected their trainers. Some examples include generating executable computer code, telling strange stories and identifying movies from a string of emojis as clues.
Ensemble Methods
In Statistics and Machine Learning, ensemble methods use multiple learning algorithms to obtain better predictive performance that could be obtained from any of the constituent learning algorithms alone. Unlike a statistical ensemble in statistical mechanics, which is usually infinite, a machine learning ensemble consists of only a concrete finite set of alternative models but typically allows for a much more flexible structure to exist among those alternatives.
Entropy
The measure of unpredictability or randomness in a set of data used in machine learning. A higher entropy signifies greater uncertainty in predicting outcomes. The average amount of information conveyed by a stochastic source of data.
Epoch
An iteration in the training process where the entire dataset is presented to a machine learning model. In the context of training Deep Learning models, one pass of the full training data set.
Expert System
A type of AI model in which the machine performs tasks based on predetermined expertise that has been hard-coded into them, coded by humans to simulate the judgment and behavior of a human expert. A form of rules-based AI that draws inferences from a knowledge base provided by human experts to replicate their decision-making abilities within a specific field, like medical diagnoses.
Explainability
A key approach to building responsible AI is model explainability, which refers to making AI models — and how they make certain decisions — transparent and easy to understand. The ability to describe or provide sufficient information about how an AI system generates a specific output or arrives at a decision in a specific context. Explainability is important in maintaining transparency and trust in AI.
Exploratory Data Analysis
Data discovery process techniques that take place before training a machine learning model to gain preliminary insights into a dataset, such as identifying patterns, outliers and anomalies and finding relationships among variables.
F-Score
A measure of a model’s accuracy considering both the precision and the recall to compute the score. More specifically, the F-Score is the harmonic average of the precision and recall, where it reaches its maximal value at 1 (perfect precision and recall) and minimum at 0.
Fairness
An attribute of an AI system that prioritizes relatively equal treatment of individuals or groups in its decisions and actions in a consistent, accurate and measurable manner. Every model must identify the appropriate standard of fairness that best applies, but most often it means the AI system's decisions should not adversely impact, whether directly or disparately, sensitive attributes like race, gender or religion.
FastText
A word embedding technique that represents words as bags of character N-grams, facilitating efficient language processing.
Feature Engineering
Feature engineering is the process of selecting, transforming, and creating relevant features in a dataset for the purpose of improving the downstream task performance of a system, such as a trained neural network.
Feature Extraction
The process of distilling relevant information from raw data to create meaningful features for machine learning.
Feature Learning
An ensemble of techniques meant to automatically discover the representations needed for feature detection or classification from raw data.
Feature Selection
The task of identifying and retaining the most crucial features while discarding less relevant ones in a dataset to use only relevant data and ignore noise.
Federated Learning
A collaborative machine learning paradigm where models are trained across distributed devices while preserving data privacy. A machine learning method that allows models to be trained on the local data of multiple edge devices. Only the updates of the local model, not the training data itself, are sent to a central location where they are aggregated to improve the global model — a process that is iterated until the global model is fully trained. This process enables better privacy and security controls for the individual user data.
Feedforward Neural Networks (FNN)
A Feedforward Neural Network (FNN) is a type of artificial neural network in which information flows in a single direction—from the input layer through hidden layers to the output layer—without loops or feedback. This straightforward structure is often used for pattern recognition tasks like image and speech classification. Compared to Convolutional Neural Networks (CNNs), which are specialized for processing grid-like data such as images using filters to capture spatial features, FNNs don’t handle spatial relationships as effectively. Unlike Recurrent Neural Networks (RNNs), which include feedback loops to manage sequence data (like text or time series), FNNs lack memory, making them better suited for static data.
Few Shot Learning
When training an AI model, giving it a few “shots” or examples in order to improve its performance on a specific task.
Fine Tuning LLM
Finetuning an LLM refers to algorithmically adjusting the parameters of that LLM in relation to a specific task-related dataset to further enhance the LLM’s performance on a particular task or domain.
Fitted Q Iteration (FQI)
An algorithm in reinforcement learning used to approximate the Q-function and solve optimal control problems.
Fitting
The process of adjusting the parameters of a model to best match observed data. Fitting involves minimizing the difference between the model’s predictions and the actual data points, typically achieved through optimization techniques like gradient descent. This process enables the model to generalize and make accurate predictions on new, unseen data.
Fog Computing
Fog computing is a decentralized computing infrastructure that brings data storage, processing, and applications closer to devices and users at the network’s edge rather than relying solely on centralized cloud servers. This approach reduces latency, improves response times, and enhances the efficiency of applications that require real-time data processing, such as autonomous vehicles, smart cities, and industrial IoT. By processing data closer to its source, fog computing minimizes the amount of information sent to the cloud, reducing bandwidth and increasing security. It essentially bridges edge devices and the cloud, making data management faster and more efficient.
Foundation Model
When an AI model is trained on a diverse set of unstructured data to create a general or “base model”, which can then be further fine-tuned (i.e., customized) to excel at a specific task. NOTE: Why unstructured data? Unstructured data (such as raw text from websites, books, and articles, or images from the internet) is more abundant and inherently diverse — providing a wealth of human knowledge, language, and visual information. This diversity is crucial for developing models with a broad understanding and the ability to generalize across a wide range of tasks. A foundational model is a pre-trained model, usually focusing on a single modality of content (i.e., image or text), that serves as a starting point for further training on more specific tasks and task-oriented data. A large-scale model that has been trained on extensive and diverse datasets to enable broad capabilities, such as language, vision, robotics, reasoning, search or human interaction, that can function as the base for use-specific applications. Also called general purpose AI model and frontier AI.
Fuzz Testing
Fuzz testing (or fuzzing) is a software testing technique that inputs large amounts of random or unexpected data ("fuzz") into a program to identify bugs, crashes, or vulnerabilities. By exposing how the system behaves under unexpected conditions, fuzz testing helps uncover edge cases, security flaws, and weaknesses that traditional testing might miss. It’s commonly used for improving software reliability and security, particularly in systems that process complex inputs like web services, file parsers, and APIs.
Fuzzy Logic Systems
Fuzzy logic systems (FLS) are a type of system that uses fuzzy logic to model uncertainty and ambiguity in real-world problems.