In computer science, an instance (also called a token in metalogic and metamathematics) is a specific occurrence of a software element that is based on a type definition. When such an occurrence is created, it is said to have been instantiated, and both the creation process and the resulting occurrence are referred to as instantiation. The concept is fundamental across many areas of computing, from object-oriented programming to artificial intelligence and cloud infrastructure.
The term derives from the broader philosophical and logical notion of an instance as a particular example of a general class or pattern. In computing, it provides a way to distinguish between a definition (such as a class, template, or model) and the concrete, usable realizations of that definition. Each instance typically has its own identity, state, and lifecycle, even when it shares structural characteristics with other instances of the same type.
Instances in Programming Languages
In object-oriented programming, an instance is an object created from a class type. The class defines the structure (attributes) and behavior (methods) that all its instances share, but each instance maintains its own identity and state. For example, a Car class might define properties like color and speed, and each individual car object created from that class is a distinct instance with its own values for those properties.
The concept also appears in procedural contexts. In the Simula language, which introduced many object-oriented ideas, each procedure call can be viewed as an instance of that procedure: an activation with its own parameters and local variables. This perspective treats a procedure definition as a template and each invocation as a concrete occurrence, which is useful for understanding recursion and call stacks.
In POSIX-oriented operating systems, a running process is an instance of a program. A program is a static file containing instructions, while a process is a dynamic execution of those instructions. Processes are instantiated via system calls such as fork() and exec(), and each executing process is an instance of the program from which it was instantiated. Multiple instances of the same program can run concurrently, each with its own memory space and execution state.
Instances in Artificial Intelligence and Machine Learning
In modern artificial intelligence and machine learning, the instance concept appears in several forms. In chat-based AI systems, which often rely on large language models, an assistant can be invoked across many independent conversation sessions, often called threads. Each thread has its own message history, and a specific execution of the assistant over that session may be represented as a run, which is an execution on a thread. This allows the same underlying model to serve many users simultaneously, with each conversation being a distinct instance of the assistant's behavior.
In deep learning, a training example or data point is sometimes called an instance of the dataset. Each instance is a specific realization of the data distribution that the model learns from. The distinction between a model definition and its trained instances is also important: a neural network architecture defines a class of possible models, and training produces a specific instance with particular learned weights.
Instances in Cloud Computing and Virtualization
In cloud computing and virtualization, an instance commonly refers to a provisioned virtual machine or virtual server with an allocated combination of compute, memory, network, and storage resources. Major cloud providers such as Amazon Web Services, Microsoft Azure, and Google Cloud offer various instance types optimized for different workloads, from general-purpose computing to GPU-accelerated training of AI models.
For example, AWS Trainium instances are specialized for training machine learning models, while Oracle Cloud and Alibaba Cloud provide their own instance families. The lifecycle of a cloud instance typically involves creation from a machine image or template, configuration, use, and eventual termination. This abstraction allows users to scale resources dynamically without managing physical hardware.
Instances in Computer Graphics
In computer graphics, instancing is a technique where a polygonal model is drawn multiple times with different transforms and parameters, improving performance by reusing shared geometry data. Instead of sending the full geometry for each object, the system sends the geometry once and then specifies multiple instances with different positions, rotations, scales, or other attributes. This is widely used in video games and simulations to render large numbers of similar objects, such as trees, particles, or crowd characters, efficiently.
Related Concepts and Distinctions
The instance concept is closely related to the type-token distinction in logic and linguistics. A type is a general category, while a token is a specific occurrence. In programming, a class or template is a type, and each object or process created from it is a token or instance. This distinction is also relevant in data augmentation, where new training instances are generated from existing ones by applying transformations.
In generative AI, each output produced by a model can be considered an instance of the model's behavior, conditioned on a specific input or prompt. The notion of instantiation is also used in transformer architectures, where each input sequence is processed as a specific instance of the model's input space. Understanding instances is essential for debugging, resource management, and reasoning about the behavior of complex software systems.