Agile development is a way of creating software where teams work flexibly and cooperate with the client. They divide work into short cycles (iterations) and respond flexibly to changes. The goal is to deliver valuable results more efficiently.
An algorithm is a list of steps or rules that a computer follows to solve a specific problem or perform a task.
also see Artificial intelligence
Annotation for computer vision involves labelling or describing images so that a computer model can understand the content. This way, the model learns to recognise and classify objects, which is crucial for its ability to "see" and understand visual information.
Automation means replacing human tasks and processes with machines or AI systems with the aim of increasing efficiency, speed and accuracy in various fields.
Autonomous systems are systems that are capable of operating independently and automatically without human intervention.
Bayesian networks are a graphical model that uses probabilistic methods to model and analyse relationships between various events or variables. They are used for decision-making under uncertainty.
also see Big data (extended)
Blockchain is a special kind of distributed decentralised database storing a continually growing number of records that are protected against unauthorised modification both from outside and from the peer-to-peer network nodes themselves.
A bounding box is a rectangle that surrounds an object in an image. It helps computers identify and locate objects in visual data.
A chatbot is software designed to communicate with users via text or voice messages. It can answer questions, perform tasks or simulate conversation with human users. They are often used to improve customer support, automate tasks and interact with users online.
Smart cities are cities that use modern technologies and data to improve the quality of life of their residents and the efficiency of city services. This technology includes the Internet of Things (IoT), artificial intelligence, big data and other digital tools to optimise transport, energy consumption, public safety, waste management and other urban functions.
Smart buildings use advanced technologies and systems to improve their functionality, efficiency and comfort. These technologies include automation of lighting, heating, air conditioning, security and other systems in the building. Smart buildings are often equipped with sensors and control systems that allow optimisation of energy consumption, increase the safety and comfort of residents and reduce maintenance costs.
also see Machine vision
A dashboard is a part of the user interface where overviews of information such as analyses, trends and reports are located.
Data mining is the process of finding patterns and information in large datasets using methods of machine learning, statistics and other techniques.
A database is a structured set of data organised and stored to allow efficient searching, access and management of information. In artificial intelligence, a database can be used to store data for model training, such as text data, images or time series.
A dataset is a collection of data. It can be anything from images to numbers. It is the foundation that computers use to learn and build models.
also see Deep learning (extended)
Digitisation refers to the process of converting physical data and information into a digital form that is easily processable and storable on computers. This process is crucial for data processing and analysis in the field of artificial intelligence.
Recommendation systems are technologies that analyse data and provide recommendations to users based on their behaviour or preferences.
Edge AI is artificial intelligence technology deployed on the device itself, where data is generated, enabling its processing and decision-making in real time. This helps not only with latency but also with efficiency and accuracy.
Edge computing represents a type of architecture adapted for Edge AI, which moves computation and data storage closer to the data source. This brings improved response and saves the network.
End-to-End learning is an approach in which the model is trained directly on the input and output data without explicitly using intermediate steps or representations. This approach allows the model to automatically estimate the best way to process data and create complex relationships between inputs and outputs.
Ethernet is a technology used for computer networks that allows devices such as computers, servers, routers, switches and other devices to communicate with each other via a wired connection. Ethernet has become one of the most widespread standards for connecting devices in local networks.
AI ethics deals with the discussion of the moral and societal impacts of creating and using artificial intelligence and autonomous systems.
FIFO (First In, First Out) is a principle of data or object processing in which the item that was inserted into the system first is removed first. It works just like a queue of people — whoever comes first is first in line. FIFO is used, for example, in data queues, buffers, network communication or task management.
Fine-tuning means continuing the training of a machine learning model on specific data or tasks after its initial training. The purpose is to improve the model's performance for specific tasks and conditions.
A framework is like a ready-made building set for creating software. It provides pre-built tools and rules that make application development easier. Programmers can use them as a foundation and save time and effort when creating programs.
Generative models are AI models that focus on creating new data based on patterns and distributions learned from the training data.
GPT is a machine learning model focused on generating text based on the data used during training. For example, GPT-4 is one of the largest and most powerful text generators developed by OpenAI.
also see Graphics card
A graphics card is specialised hardware designed for fast graphics processing. In the context of AI applications, it plays a key role thanks to parallel processing, when it handles many tasks at the same time. This significantly speeds up machine learning and data processing. The graphics card provides the computational power needed for complex operations of neural networks, thereby improving the performance and efficiency of AI applications.
Hallucination in this context refers to the situation where a machine learning model generates information that is not true or does not exist at all. This can happen if the model is poorly trained or if the training data is not representative of the real scenarios with which the model will work.
Hardware refers to the physical components of a computer that you can see and touch, such as the monitor, keyboard and processor.
Deep learning is a specific method of machine learning using artificial neural networks with multiple layers for data analysis, which is the basis of many successful applications in AI.
An ID is a property of an object that helps us distinguish objects from each other and thus guarantee their uniqueness. When a new object appears, it is automatically assigned an ID (most often a number). If the computer needs to return to this object, it has to know its ID.
Image Processing deals with the analysis, modification and interpretation of digital images. In artificial intelligence, it includes techniques that allow computers to work with visual data, such as object detection, image segmentation and image data processing.
Inference refers to the process of deriving new information or conclusions based on existing data and knowledge. In artificial intelligence, this process involves making decisions and predictions based on training models and available data.
Integration connects different systems or elements so that they work together. It is like assembling a puzzle: different pieces (data or programs) are linked and work together for a better result.
The Internet of Things represents a network of physical devices and objects that are connected to the internet and can communicate and share data with each other. In connection with artificial intelligence, data from devices is often used for model training, analysis and support for automation or prediction.
also see Internet of Things
An IP camera is a digital video camera that sends and receives data over a network or the internet. Unlike analogue cameras, which transmit images through a coaxial cable, IP cameras use network infrastructure (Ethernet or Wi-Fi) and can be connected directly to the network.
Calibration involves refining analysis thanks to data from real operation. For example, in the case of traffic analysis, using video recordings of the road, we can refine the analysis even in poor lighting conditions.
Conversational artificial intelligence focuses on the ability of systems to communicate with people in natural language. It includes chatbots, virtual assistants and other technologies that enable smooth interaction through text or voice.
also see Local network
A large language model is a type of artificial intelligence trained on a huge amount of text data. The goal of an LLM is to understand natural language and generate meaningful text based on the input query.
LIFO (Last In, First Out) is a principle of data organisation in which the item that was inserted last is processed or removed first. It corresponds to the behaviour of a stack — newly added data lies on top and is taken first when removed. LIFO is used, for example, in the stack memory of programs, when processing recursion or in some types of data structures.
also see Large language model
A local network is a group of connected computers in one place, for example in a home or office. Using a local network, devices can share data, printers or an internet connection.
also see Machine learning (extended)
A model is a program that learns from experience and data to perform tasks without explicit instructions. For example, it recognises images or predicts trends. It is like an artificial mind that improves itself.
Model drift occurs when the performance of an AI model gradually decreases due to changes in the training data or environment. This requires regular updating and retraining of the model.
Narrow artificial intelligence refers to AI systems that are designed to solve specific tasks or specific domains, without having general intelligence. These systems cannot perform tasks outside their specialisation and focus, for example, on image recognition, autonomous driving or interaction through chatbots.
Instruction tuning refers to the process of optimising and refining instructions for a machine learning model so that the user better understands the output information. This can include modifying the rules or instructions that affect the model's behaviour.
Neural networks form a model inspired by the human brain, consisting of artificial neurons that communicate and process information together.
Robustness refers to the ability of an artificial intelligent system to maintain its performance and accuracy in various situations or under deviations from the training data. Robust systems are able to operate efficiently even in unpredictable conditions or environmental changes.
Optimisation involves the process of finding the best solution for a specific task. In artificial intelligence, it is often used to fine-tune model parameters, select features or find optimal hyperparameters to achieve maximum performance.
An undertrained model is a model that, due to a small dataset, does not understand the relationship between data well, and therefore often starts to predict poorly and thus has unnecessarily high loss and low accuracy.
Impact assessment focuses on the analysis and evaluation of the impact of a particular measure, policy or technology on the environment, society or economy. In the context of artificial intelligence, it serves to assess the consequences that the implementation of AI systems may bring.
Predictive analysis deals with the use of data, statistics and machine learning techniques to predict future events or trends.
Pre-training is the process in which a machine learning model is trained on a dataset before its final fine-tuning for a specific task. This pre-training helps us obtain information that we can subsequently use when finalising the dataset.
A pre-training model is a model created from a dataset that is not yet in its final form. This model can reveal shortcomings in the given dataset.
Industry 4.0 represents the concept of digitisation and automation of industrial processes using modern technologies such as artificial intelligence, the Internet of Things (IoT) and others. Its goal is to improve efficiency, productivity and innovation in the industrial sector.
Pruning is a data compression technique in machine learning and search algorithms that increases the speed and efficiency of the model by removing those parts of the model that are not key and are redundant for its correct functioning.
Overfitting occurs when a machine model achieves high accuracy on training data but achieves low accuracy on new, unseen data. This phenomenon indicates that the model has overly adapted to the training data and is unable to generalise to new situations. Overfitting represents one of the main challenges in machine learning, which requires measures such as regularisation and model validation.
Python is a programming language, increasingly popular for wide use. It excels particularly in the areas of artificial intelligence and machine learning.
Regularisation is a technique in machine learning used to limit persistent overfitting of the model. This is achieved by adding additional conditions or penalties to the model parameters during training. Regularisation helps the model generalise better to new data and prevents excessive adaptation to the training data.
Risk tolerance refers to the extent to which an organisation is willing to accept risks associated with the deployment or use of artificial intelligence. This tolerance can influence decisions regarding the acceptability of a type of AI system and the way of managing risks associated with its operation.
Robotics is a field focused on the design, development and operation of robots. In the context of artificial intelligence, robotics is often used to create autonomous robots that are able to perform various tasks and interact with the physical environment.
Scaling is the process of expanding or adjusting the performance of a system so that it can handle a larger volume of data, users or requests. It can be vertical (increasing the performance of a single server) or horizontal (adding more servers or services). The goal of scaling is to maintain the speed, stability and availability of the system even under increasing load.
also see Smart buildings (extended)
also see Smart cities (extended)
Software is a set of instructions that controls the behaviour of a computer or device. It is the intangible part of a computer that allows you to perform various tasks, e.g. write text, watch films or work with images.
Data stacking is a way of organisation in which individual pieces of information are stored or stacked on top of each other in layers (so-called stack). This principle is used, for example, in the processing of time-series data from sensors, in combining the outputs of multiple models in machine learning, or in storing in a stack in programming. Stacking allows efficient work with data in sequence and their fast processing according to the principle "first down comes last up" (LIFO) or by time layers.
Machine learning deals with AI processes that allow computers to learn from data and experience and improve their performance without explicit programming.
Machine vision deals with the ability of computers to understand and interpret visual information from digital images or videos. This field includes techniques for object recognition, image segmentation, face detection and other tasks related to visual perception.
A tracker is a technology that allows us to distinguish several objects that are located in the same scene from each other. The result of this identification is that we can assign a unique identifier (ID) to each object. The tracker receives information from the model — namely, where and what objects are located. Based on this information, the tracker realises that, for example, object number 1 is moving to the left and object number 2 to the right. In this way, it distinguishes these two objects.
Model training is the process in which a computer program (model) improves its abilities by learning from data. Similarly to how a child learns, the model adapts and improves its skills based on experience.
Artificial intelligence is a field of computer science that deals with creating systems and algorithms that are capable of performing tasks that usually require human intelligence.
Artificial general intelligence is a concept of AI which means creating a system with intelligence equivalent to human intelligence, capable of adapting to various tasks and environments.
An update focuses on fixing bugs, improving security and more general updates of an existing version. Its goal is to keep software or hardware current and efficient without the need for a major shift to a new version.
An upgrade means moving to a newer version with improved features and capabilities. It is like "moving up a level" for better performance or new features.
Big data refers to huge volumes of data that are too large, complex or rapidly changing to be efficiently processed by traditional database tools and techniques. Big data includes structured and unstructured data from various sources, such as social media, sensors, transaction systems and others.
Video stream is watching video over the internet without the need for downloading to a computer.
VLA (Vision-Language-Action) models are artificial intelligence systems that connect the ability to see (image processing) and understand text directly with physical action, typically controlling robots.
Zero Trust is a security approach that does not trust any user or device automatically. Every access is always verified — based on identity, device, context and other factors. The goal is to minimise the risk of unauthorised access and prevent the spread of threats within the system.