Kinect is a discontinued line of motion sensing input devices produced by Microsoft and first released in 2010. The devices generally contain RGB cameras, and infrared projectors and detectors that map depth through either structured light or time of flight calculations, which can in turn be used to perform real-time gesture recognition and body skeletal detection, among other capabilities. They also contain microphones that can be used for speech recognition and voice control. Kinect was originally developed as a motion controller peripheral for Xbox video game consoles, distinguished from competitors (such as Nintendo's Wii Remote and Sony's PlayStation Move) by not requiring physical controllers. The first-generation Kinect was based on technology from Israeli company PrimeSense, and unveiled at E3 2009 as a peripheral for Xbox 360 codenamed "Project Natal". It was first released on November 4, 2010, and would go on to sell eight million units in its first 60 days of availability. The majority of the games developed for Kinect were casual, family-oriented titles, which helped to attract new audiences to Xbox 360, but did not result in wide adoption by the console's existing, overall userbase. As part of the 2013 unveiling of Xbox 360's successor, Xbox One, Microsoft unveiled a second-generation version of Kinect with improved tracking capabilities. Microsoft also announced that Kinect would be a required component of the console, and that it would not function unless the peripheral is connected. The requirement proved controversial among users and critics due to privacy concerns, prompting Microsoft to backtrack on the decision. However, Microsoft still bundled the new Kinect with Xbox One consoles upon their launch in November 2013. A market for Kinect-based games still did not emerge after the Xbox One's launch; Microsoft would later offer Xbox One hardware bundles without Kinect included, and later revisions of the console removed the dedicated ports used to connect it (requiring a powered USB adapter instead). Microsoft ended production of Kinect for Xbox One in October 2017. Kinect has also been used as part of non-game applications in academic and commercial environments, as it was cheaper and more robust than other depth-sensing technologies at the time. While Microsoft initially objected to such applications, it later released software development kits (SDKs) for the development of Microsoft Windows applications that use Kinect. In 2020, Microsoft released Azure Kinect as a continuation of the technology integrated with the Microsoft Azure cloud computing platform. Part of the Kinect technology was also used within Microsoft's HoloLens project. Microsoft discontinued the Azure Kinect developer kits in October 2023. == History == === Development === The origins of the Kinect started around 2005, at a point where technology vendors were starting to develop depth-sensing cameras. Microsoft had been interested in a 3D camera for the Xbox line earlier but because the technology had not been refined, had placed it in the "Boneyard", a collection of possible technology they could not immediately work on. In 2005, Israeli company PrimeSense was founded by mathematicians and engineers to develop the "next big thing" for video games, incorporating cameras that were capable of mapping a human body in front of them and sensing hand motions. They showed off their system at the 2006 Game Developers Conference, where Microsoft's Alex Kipman, the general manager of hardware incubation, saw the potential in PrimeSense's technology for the Xbox system. Microsoft began discussions with PrimeSense about what would need to be done to make their product more consumer-friendly: not only improvements in the capabilities of depth-sensing cameras, but a reduction in size and cost, and a means to manufacture the units at scale was required. PrimeSense spent the next few years working at these improvements. Nintendo released the Wii in November 2006. The Wii's central feature was the Wii Remote, a handheld device that was detected by the Wii through a motion sensor bar mounted onto a television screen to enable motion controlled games. Microsoft felt pressure from the Wii, and began looking into depth-sensing in more detail with PrimeSense's hardware, but could not get to the level of motion tracking they desired. While they could determine hand gestures, and sense the general shape of a body, they could not do skeletal tracking. A separate path within Microsoft looked to create an equivalent of the Wii Remote, considering that this type of unit may become standardized similar to how two-thumbstick controllers became a standard feature. However, it was still ultimately Microsoft's goal to remove any device between the player and the Xbox. Kudo Tsunoda and Darren Bennett joined Microsoft in 2008, and began working with Kipman on a new approach to depth-sensing aided by machine learning to improve skeletal tracking. They internally demonstrated this and established where they believed the technology could be in a few years, which led to the strong interest to fund further development of the technology; this has also occurred at a time that Microsoft executives wanted to abandon the Wii-like motion tracking approach, and favored the depth-sensing solution to present a product that went beyond the Wii's capabilities. The project was greenlit by late 2008 with work started in 2009. The project was codenamed "Project Natal" after the Brazilian city Natal, Kipman's birthplace. Additionally, Kipman recognized the Latin origins of the word "natal" to mean "to be born", reflecting the new types of audiences they hoped to draw with the technology. Much of the initial work was related to ethnographic research to see how video game players' home environments were laid out, lit, and how those with Wiis used the system to plan how Kinect units would be used. The Microsoft team discovered from this research that the up-and-down angle of the depth-sensing camera would either need to be adjusted manually, or would require an expensive motor to move automatically. Upper management at Microsoft opted to include the motor despite the increased cost to avoid breaking game immersion. Kinect project work also involved packaging the system for mass production and optimizing its performance. Hardware development took around 22 months. During hardware development, Microsoft engaged with software developers to use Kinect. Microsoft wanted to make games that would be playable by families since Kinect could sense multiple bodies in front of it. One of the first internal titles developed for the device was the pack-in game Kinect Adventures developed by Good Science Studio that was part of Microsoft Studios. One of the game modes of Kinect Adventures was "Reflex Ridge", based on the Japanese Brain Wall game where players attempt to contort their bodies in a short time to match cutouts of a wall moving at them. This type of game was a key example of the type of interactivity they wanted with Kinect, and its development helped feed into the hardware improvements. Another development was Project Milo, a prototype game developed by Lionhead Studios led by Peter Molyneux where the player could interact with a virtual avatar through motion controls and voice recognition. Lionhead had developed the project based on original capabilities of the Kinect, but according to Molyneux, Microsoft had found that a consumer-grade version of the Kinect would cost thousands of dollars, so they scaled back the device and refocused the role of games for the Kinect to be more casual games as seen on the Wii. As a result, Project Milo no longer fit Microsoft's portfolio and was cancelled. Nearing the planned release, there was a problem of widespread testing of Kinect in various room types and different bodies accounting for age, gender, and race among other factors, while keeping the details of the unit confidential. Microsoft engaged in a company-wide program offering employees to take home Kinect units to test them. Microsoft also brought other non-gaming divisions, including its Microsoft Research, Microsoft Windows, and Bing teams to help complete the system. Microsoft established its own large-scale manufacturing facility to bulk product Kinect units and test them. === Introduction === Kinect was first announced to the public as "Project Natal" on June 1, 2009, during Microsoft's press conference at E3 2009; film director Steven Spielberg joined Microsoft's Don Mattrick to introduce the technology and its potential. Three demos were presented during the conference—Microsoft's Ricochet and Paint Party, and Lionhead Studios' Milo & Kate created by Peter Molyneux—while a Project Natal-enabled version of Criterion Games' Burnout Paradise was shown during the E3 exhibition. By E3 2009, the skeletal mapping technology was capable of simultaneously tracking four people, with a feature extraction of 4
Nouvelle AI
Nouvelle artificial intelligence (Nouvelle AI) is an approach to artificial intelligence pioneered in the 1980s by Rodney Brooks, who was then part of MIT artificial intelligence laboratory. Nouvelle AI differs from classical AI by aiming to produce robots with intelligence levels similar to insects. Researchers believe that intelligence can emerge organically from simple behaviors as these intelligences interacted with the "real world", instead of using the constructed worlds which symbolic AIs typically needed to have programmed into them. == Motivation == The differences between nouvelle AI and symbolic AI are apparent in early robots Shakey and Freddy. These robots contained an internal model (or "representation") of their micro-worlds consisting of symbolic descriptions. As a result, this structure of symbols had to be renewed as the robot moved or the world changed. Shakey's planning programs assessed the program structure and broke it down into the necessary steps to complete the desired action. This level of computation required a large amount time to process, so Shakey typically performed its tasks very slowly. Symbolic AI researchers had long been plagued by the problem of updating, searching, and otherwise manipulating the symbolic worlds inside their AIs. A nouvelle system refers continuously to its sensors rather than to an internal model of the world. It processes the external world information it needs from the senses when it is required. As Brooks puts it, "the world is its own best model--always exactly up to date and complete in every detail." A central idea of nouvelle AI is that simple behaviors combine to form more complex behaviors over time. For example, simple behaviors can include elements like "move forward" and "avoid obstacles." A robot using nouvelle AI with simple behaviors like collision avoidance and moving toward a moving object could possibly come together to produce a more complex behavior like chasing a moving object. === The frame problem === The frame problem describes an issue with using first-order logic (FOL) to express facts about a robot in the world. Representing the state of a robot with traditional FOL requires the use of many axioms (symbolic language) to imply that things about an environment do not change arbitrarily. Nouvelle AI seeks to sidestep the frame problem by dispensing with filling the AI or robot with volumes of symbolic language and instead letting more complex behaviors emerge by combining simpler behavioral elements. === Embodiment === The goal of traditional AI was to build intelligences without bodies, which would only have been able to interact with the world via keyboard, screen, or printer. However, nouvelle AI attempts to build embodied intelligence situated in the real world. Brooks quotes approvingly from the brief sketches that Turing gave in 1948 and 1950 of the "situated" approach. Turing wrote of equipping a machine "with the best sense organs that money can buy" and teaching it "to understand and speak English" by a process that would "follow the normal teaching of a child." This approach was contrasted to the others where they focused on abstract activities such as playing chess. == Brooks' robots == === Insectoid robots === Brooks focused on building robots that acted like simple insects while simultaneously working to remove some traditional AI characteristics. He created insect-like robots, named Allen and Herbert after cognitive science and AI pioneers Allen Newell and Herbert A. Simon. Brooks's insectoid robots contained no internal models of the world. Herbert, for example, discarded a high volume of the information received from its sensors and never stored information for more than two seconds. ==== Allen ==== Allen had a ring of twelve ultrasonic sonars as its primary sensors and three independent behavior-producing modules. These modules were programmed to avoid both stationary and moving objects. With only this module activated, Allen stayed in the middle of a room until an object approached and then it ran away while avoiding obstacles in its way. ==== Herbert ==== Herbert used infrared sensors to avoid obstacles and a laser system to collect 3D data over a distance of about 12 feet. Herbert also carried a number of simple sensors in its "hand." The robot's testing ground was the real world environment of the busy offices and workspaces of the MIT AI lab where it searched for empty soda cans and carried them away, a seemingly goal-oriented activity that emerged as a result of 15 simple behavior units combining. As a parallel, Simon noted that an ant's complicated path is due to the structure of its environment rather than the depth of its thought processes. ==== Other insectoid robots ==== Other robots by Brooks' team were Genghis and Squirt. Genghis had six legs and was able to walk over rough terrain and follow a human. Squirt's behavior modules had it stay in dark corners until it heard a noise, then it would begin to follow the source of the noise. Brooks agreed that the level of nouvelle AI had come near the complexity of a real insect, which raised a question about whether or not insect level-behavior was and is a reasonable goal for nouvelle AI. === Humanoid robots === Brooks' own recent work has taken the opposite direction to that proposed by Von Neumann in the quotations "theorists who select the human nervous system as their model are unrealistically picking 'the most complicated object under the sun,' and that there is little advantage in selecting instead the ant, since any nervous system at all exhibits exceptional complexity." ==== Cog ==== In the 1990s, Brooks decided to pursue the goal of human-level intelligence and, with Lynn Andrea Stein, built a humanoid robot called Cog. Cog is a robot with an extensive collection of sensors, a face, and arms (among other features) that allow it to interact with the world and gather information and experience so as to assemble intelligence organically in the manner described above by Turing. The team believed that Cog would be able to learn and able to find a correlation between the sensory information it received and its actions, and to learn common sense knowledge on its own. As of 2003, all development of the project had ceased.
Self-supervised learning
Self-supervised learning (SSL) is a paradigm in machine learning where a model is trained on a task using the data itself to generate supervisory signals, rather than relying on externally-provided labels. In the context of neural networks, self-supervised learning aims to leverage inherent structures or relationships within the input data to create meaningful training signals. SSL tasks are designed so that solving them requires capturing essential features or relationships in the data. The input data is typically augmented or transformed in a way that creates pairs of related samples, where one sample serves as the input, and the other is used to formulate the supervisory signal. This augmentation can involve introducing noise, cropping, rotation, or other transformations. Self-supervised learning more closely imitates the way humans learn to classify objects. During SSL, the model learns in two steps. First, the task is solved based on an auxiliary or pretext classification task using pseudo-labels, which help to initialize the model parameters. Next, the actual task is performed with supervised or unsupervised learning. Self-supervised learning has produced promising results in recent years, and has found practical application in fields such as audio processing, and is being used by Facebook and others for speech recognition. == Pseudo-labels == Pseudo-labels are automatically generated labels that a model assigns to unlabeled data based on its own predictions. They are widely used in self-supervised and semi-supervised learning, where ground-truth annotations are limited or unavailable. By treating predicted labels as surrogate ground truth, learning algorithms can make use of large quantities of unlabeled data in the training process. Pseudo-labeling also plays an important role in systems that must adapt to concept drift, where the statistical properties of the data change over time. In these scenarios, the model may detect that an incoming instance deviates from previously learned behavior. The system then generates a classification result for that instance, and this predicted class is used as a pseudo-label for updating or retraining model components that are becoming outdated. This approach enables continuous adaptation in dynamic environments without requiring manual annotation. In many adaptive learning pipelines, pseudo-labels are chosen when the classifier produces sufficiently confident predictions, reducing the risk of propagating errors. These pseudo-labeled instances are then incorporated into training to refresh or evolve the model's understanding of emerging data patterns, particularly when existing components show signs of “aging” due to drift or distributional shifts. This strategy reduces reliance on manual labeling while helping maintain long-term model performance. == Types == === Autoassociative self-supervised learning === Autoassociative self-supervised learning is a specific category of self-supervised learning where a neural network is trained to reproduce or reconstruct its own input data. In other words, the model is tasked with learning a representation of the data that captures its essential features or structure, allowing it to regenerate the original input. The term "autoassociative" comes from the fact that the model is essentially associating the input data with itself. This is often achieved using autoencoders, which are a type of neural network architecture used for representation learning. Autoencoders consist of an encoder network that maps the input data to a lower-dimensional representation (latent space), and a decoder network that reconstructs the input from this representation. The training process involves presenting the model with input data and requiring it to reconstruct the same data as closely as possible. The loss function used during training typically penalizes the difference between the original input and the reconstructed output (e.g. mean squared error). By minimizing this reconstruction error, the autoencoder learns a meaningful representation of the data in its latent space. === Contrastive self-supervised learning === For a binary classification task, training data can be divided into positive examples and negative examples. Positive examples are those that match the target. For example, if training a classifier to identify birds, the positive training data would include images that contain birds. Negative examples would be images that do not. Contrastive self-supervised learning uses both positive and negative examples. The loss function in contrastive learning is used to minimize the distance between positive sample pairs, while maximizing the distance between negative sample pairs. An early example uses a pair of 1-dimensional convolutional neural networks to process a pair of images and maximize their agreement. Contrastive Language-Image Pre-training (CLIP) allows joint pretraining of a text encoder and an image encoder, such that a matching image-text pair have image encoding vector and text encoding vector that span a small angle (having a large cosine similarity). InfoNCE (Noise-Contrastive Estimation) is a method to optimize two models jointly, based on Noise Contrastive Estimation (NCE). Given a set X = { x 1 , … x N } {\displaystyle X=\left\{x_{1},\ldots x_{N}\right\}} of N {\displaystyle N} random samples containing one positive sample from p ( x t + k ∣ c t ) {\displaystyle p\left(x_{t+k}\mid c_{t}\right)} and N − 1 {\displaystyle N-1} negative samples from the 'proposal' distribution p ( x t + k ) {\displaystyle p\left(x_{t+k}\right)} , it minimizes the following loss function: L N = − E X [ log f k ( x t + k , c t ) ∑ x j ∈ X f k ( x j , c t ) ] {\displaystyle {\mathcal {L}}_{\mathrm {N} }=-\mathbb {E} _{X}\left[\log {\frac {f_{k}\left(x_{t+k},c_{t}\right)}{\sum _{x_{j}\in X}f_{k}\left(x_{j},c_{t}\right)}}\right]} === Non-contrastive self-supervised learning === Non-contrastive self-supervised learning (NCSSL) uses only positive examples. Counterintuitively, NCSSL converges on a useful local minimum rather than reaching a trivial solution, with zero loss. For the example of binary classification, it would trivially learn to classify each example as positive. Effective NCSSL requires an extra predictor on the online side that does not back-propagate on the target side. === Joint-Embedding and Predictive Architectures === A major class of self-supervised learning moves beyond contrastive pairs, instead maximizing the agreement between views while preventing collapse through statistical constraints. Rooted in Deep Canonical Correlation Analysis (Deep CCA), this approach includes Joint-Embedding Architectures (JEA) like Barlow Twins and VICReg, which enforce covariance constraints to learn invariant representations without negative sampling. Deep Latent Variable Path Modelling (DLVPM) generalizes this to multimodal systems, using path models to enforce correlation and orthogonality across diverse data types. In 2022 Yann LeCun introduced Joint-Embedding Predictive Architectures (JEPA) as a step towards decision making, reasoning, and autonomous human intelligence in machines, including self-improvement through autonomous learning. Founded in representation learning, LeCun included the concept of a “world model” in JEPA which aims to enable machines to replicate human intellect by providing machines with a concept for the world in which they exist. Unlike autoencoders, JEPAs operate entirely in latent space, avoiding pixel-level noise to focus on semantic structure. Rather than just learning invariance, JEPAs learn by predicting masked latent representations from visible context. JEPA has been applied to domains such as image analysis, audio processing, and motion in images and video. == Comparison with other forms of machine learning == SSL belongs to supervised learning methods insofar as the goal is to generate a classified output from the input. At the same time, however, it does not require the explicit use of labeled input-output pairs. Instead, correlations, metadata embedded in the data, or domain knowledge present in the input are implicitly and autonomously extracted from the data. These supervisory signals, extracted from the data, can then be used for training. SSL is similar to unsupervised learning in that it does not require labels in the sample data. Unlike unsupervised learning, however, learning is not done using inherent data structures. Semi-supervised learning combines supervised and unsupervised learning, requiring only a small portion of the learning data be labeled. In transfer learning, a model designed for one task is reused on a different task. Training an autoencoder intrinsically constitutes a self-supervised process, because the output pattern needs to become an optimal reconstruction of the input pattern itself. However, in current jargon, the term 'self-supervised' often refers to tasks based on a pretext-task training setup
Attention (machine learning)
In machine learning, attention is a method that determines the importance of each component in a sequence relative to the other components in that sequence. In natural language processing, importance is represented by "soft" weights assigned to each word in a sentence. More generally, attention encodes vectors called token embeddings across a fixed-width sequence that can range from tens to millions of tokens in size. Unlike "hard" weights, which are computed during the backwards training pass, "soft" weights exist only in the forward pass and therefore change with every step of the input. Earlier designs implemented the attention mechanism in a serial recurrent neural network (RNN) language translation system, but a more recent design, namely the transformer, removed the slower sequential RNN and relied more heavily on the faster parallel attention scheme. Inspired by ideas about attention in humans, the attention mechanism was developed to address the weaknesses of using information from the hidden layers of recurrent neural networks. Recurrent neural networks favor information contained in words at the end of a sentence and thus deemed more recent, thereby tending to attenuate the significance and associated predictive weight assigned to information earlier in the sentence. Attention allows a token equal access to any part of a sentence directly, rather than only through the previous state. == History == Additional surveys of the attention mechanism in deep learning are provided by Niu et al. and Soydaner. The major breakthrough came with self-attention, where each element in the input sequence attends to all others, enabling the model to capture global dependencies. This idea was central to the Transformer architecture, which replaced recurrence with attention mechanisms. As a result, Transformers became the foundation for models like BERT, T5 and generative pre-trained transformers (GPT). == Overview == The modern era of machine attention was revitalized by grafting an attention mechanism (Fig 1. orange) to an Encoder-Decoder. Figure 2 shows the internal step-by-step operation of the attention block (A) in Fig 1. === Interpreting attention weights === In translating between languages, alignment is the process of matching words from the source sentence to words of the translated sentence. Networks that perform verbatim translation without regard to word order would show the highest scores along the (dominant) diagonal of the matrix. The off-diagonal dominance shows that the attention mechanism is more nuanced. Consider an example of translating I love you to French. On the first pass through the decoder, 94% of the attention weight is on the first English word I, so the network offers the word je. On the second pass of the decoder, 88% of the attention weight is on the third English word you, so it offers t'. On the last pass, 95% of the attention weight is on the second English word love, so it offers aime. In the I love you example, the second word love is aligned with the third word aime. Stacking soft row vectors together for je, t', and aime yields an alignment matrix: Sometimes, alignment can be multiple-to-multiple. For example, the English phrase look it up corresponds to cherchez-le. Thus, "soft" attention weights work better than "hard" attention weights (setting one attention weight to 1, and the others to 0), as we would like the model to make a context vector consisting of a weighted sum of the hidden vectors, rather than "the best one", as there may not be a best hidden vector. == Variants == Many variants of attention implement soft weights, such as fast weight programmers, or fast weight controllers (1992). A "slow" neural network outputs the "fast" weights of another neural network through outer products. The slow network learns by gradient descent. It was later renamed as "linearized self-attention". Bahdanau-style attention, also referred to as additive attention, Luong-style attention, which is known as multiplicative attention, Early attention mechanisms similar to modern self-attention were proposed using recurrent neural networks. However, the highly parallelizable self-attention was introduced in 2017 and successfully used in the Transformer model, positional attention and factorized positional attention. For convolutional neural networks, attention mechanisms can be distinguished by the dimension on which they operate, namely: spatial attention, channel attention, or combinations. These variants recombine the encoder-side inputs to redistribute those effects to each target output. Often, a correlation-style matrix of dot products provides the re-weighting coefficients. In the figures below, W is the matrix of context attention weights, similar to the formula in Overview section above. == Optimizations == === Flash attention === The size of the attention matrix is proportional to the square of the number of input tokens. Therefore, when the input is long, calculating the attention matrix requires a lot of GPU memory. Flash attention is an implementation that reduces the memory needs and increases efficiency without sacrificing accuracy. It achieves this by partitioning the attention computation into smaller blocks that fit into the GPU's faster on-chip memory, reducing the need to store large intermediate matrices and thus lowering memory usage while increasing computational efficiency. === FlexAttention === FlexAttention is an attention kernel developed by Meta that allows users to modify attention scores prior to softmax and dynamically chooses the optimal attention algorithm. == Applications == Attention is widely used in natural language processing, computer vision, and speech recognition. In NLP, it improves context understanding in tasks like question answering and summarization. In vision, visual attention helps models focus on relevant image regions, enhancing object detection and image captioning. === Attention maps as explanations for vision transformers === From the original paper on vision transformers (ViT), visualizing attention scores as a heat map (called saliency maps or attention maps) has become an important and routine way to inspect the decision making process of ViT models. One can compute the attention maps with respect to any attention head at any layer, while the deeper layers tend to show more semantically meaningful visualization. Attention rollout is a recursive algorithm to combine attention scores across all layers, by computing the dot product of successive attention maps. Because vision transformers are typically trained in a self-supervised manner, attention maps are generally not class-sensitive. When a classification head is attached to the ViT backbone, class-discriminative attention maps (CDAM) combines attention maps and gradients with respect to the class [CLS] token. Some class-sensitive interpretability methods originally developed for convolutional neural networks can be also applied to ViT, such as GradCAM, which back-propagates the gradients to the outputs of the final attention layer. Using attention as basis of explanation for the transformers in language and vision is not without debate. While some pioneering papers analyzed and framed attention scores as explanations, higher attention scores do not always correlate with greater impact on model performances. == Mathematical representation == === Standard scaled dot-product attention === For matrices: Q ∈ R m × d k , K ∈ R n × d k {\displaystyle Q\in \mathbb {R} ^{m\times d_{k}},K\in \mathbb {R} ^{n\times d_{k}}} and V ∈ R n × d v {\displaystyle V\in \mathbb {R} ^{n\times d_{v}}} , the scaled dot-product, or QKV attention, is defined as: Attention ( Q , K , V ) = softmax ( Q K T d k ) V ∈ R m × d v {\displaystyle {\text{Attention}}(Q,K,V)={\text{softmax}}\left({\frac {QK^{T}}{\sqrt {d_{k}}}}\right)V\in \mathbb {R} ^{m\times d_{v}}} where T {\displaystyle {}^{T}} denotes transpose and the softmax function is applied independently to every row of its argument. The matrix Q {\displaystyle Q} contains m {\displaystyle m} queries, while matrices K , V {\displaystyle K,V} jointly contain an unordered set of n {\displaystyle n} key-value pairs. Value vectors in matrix V {\displaystyle V} are weighted using the weights resulting from the softmax operation, so that the rows of the m {\displaystyle m} -by- d v {\displaystyle d_{v}} output matrix are confined to the convex hull of the points in R d v {\displaystyle \mathbb {R} ^{d_{v}}} given by the rows of V {\displaystyle V} . To understand the permutation invariance and permutation equivariance properties of QKV attention, let A ∈ R m × m {\displaystyle A\in \mathbb {R} ^{m\times m}} and B ∈ R n × n {\displaystyle B\in \mathbb {R} ^{n\times n}} be permutation matrices; and D ∈ R m × n {\displaystyle D\in \mathbb {R} ^{m\times n}} an arbitrary matrix. The softmax function is permutation equivariant in the sense that: softmax ( A D B ) = A softmax ( D ) B {\displays
Artificial intelligence of things
Artificial Intelligence of Things (AIoT) is the combination of artificial intelligence (AI) technologies with the Internet of things (IoT) infrastructure to create systems capable of sensing, learning, and acting on data without continuous human intervention. While IoT focuses on connectivity and sensor data collection, AI enables IoT devices to analyse data in real time and produce actionable outputs, including automated decisions at the edge. == Applications == === Manufacturing and predictive maintenance === Manufacturing accounts for the largest share of AIoT adoption by industry vertical. A common application is predictive maintenance, where sensors measuring vibration, temperature, current draw, and acoustic emissions feed machine learning models trained to detect signatures that precede equipment failure. These systems can flag developing faults weeks or months in advance, and in more advanced deployments can autonomously adjust machine parameters such as motor speed or cooling cycles to delay or prevent failure. === Other industries === In healthcare, AIoT enables remote patient monitoring through wearable devices that collect vital signs and apply AI models to detect anomalies or predict deterioration. In logistics, GPS and telematics sensors combined with AI models support real-time route optimisation, vehicle maintenance prediction, and fuel cost forecasting. Smart building systems use occupancy, temperature, and energy sensors with AI to dynamically adjust HVAC and lighting, reducing energy consumption. == Architecture == AIoT systems typically operate across three layers: a device layer of sensors and actuators that collect data, a connectivity layer that transmits data via protocols such as MQTT or HTTP, and a compute layer where AI models process the data either in the cloud or at the edge. The trend toward edge-based processing, where inference runs on low-cost processors near the data source rather than in a centralised cloud, has accelerated as hardware costs have fallen and applications increasingly require sub-second response times. == Market == Market sizing estimates for AIoT vary significantly depending on scope and definition. Fortune Business Insights valued the AIoT market at USD 35.65 billion in 2023, projecting growth to USD 253.86 billion by 2030 at a compound annual growth rate of 32.4%. Grand View Research estimated the broader market at USD 171.4 billion in 2024 with a CAGR of 31.7% through 2030, reflecting a wider definition that includes AI-integrated hardware components. North America accounted for approximately 40% of global market share in 2024, with the Asia-Pacific region projected as the fastest-growing market.
NationBuilder
NationBuilder is a Los Angeles-based technology start-up that develops content management and customer relationship management (CRM) software. Although the company initially targeted political campaigns and nonprofit organizations, it later expanded its marketing efforts to include other people and organizations trying to build an online following, such as artists, musicians and restaurants. The software uses voter data such as names, addresses and other information, such as previous voting records in the case of political campaigns, to allow users to centralize, build and manage campaigns by integrating various communication tools like websites, newsletters, text messaging and social media channels under one platform. Among other features, the software enables users to quickly create websites, build databases through registrations, send targeted newsletters, analyse data from multiple sources and leverage micro-donations. The software's appeal towards political campaigns comes from the combination of a number of previously separate campaigning services, channels and data sources into a single platform that was presented as a facile solution for non-technical users and which enabled political campaigners to quickly deploy campaigns by convincing numerous people to donate. == History == NationBuilder was founded in 2009 in Los Angeles by Jim Gilliam and launched in 2011. In 2012 Joe Green joined NationBuilder as co-founder and president. He left that role 11 months later in February 2013. Gilliam was previously a movie-maker who co-founded Brave New Films with Robert Greenwald and had sought funding for his films through crowd-sourcing. Green, who studied organizing at Harvard and was Mark Zuckerberg's roommate, is also the co-founder of the Causes Facebook app; he left NationBuilder in 2013. Since its founding, the company has helped campaigns raise $1.2 billion. In 2012, NationBuilder announced that 1,000 subscribers have used its software to amass 2.5 million supporters and raise $12 million in campaign donations. In 2015 it has helped raise $264 million, recruit over one million volunteers and coordinate some 129,000 events. By 2016, the company said its software was used by about 40 percent of all contested elections at the state and national level in the U.S., which included 3,000 political campaigns. Using such software is easier in the U.S. than Europe, where comprehensive data protection and privacy laws are in effect since 2018. The Scottish National Party was the first political party to use NationBuilder, harvesting vast amounts of data pertaining to voter activity via websites such as Facebook and Twitter. This revelation prompted outrage over privacy concerns. Guy Herbert of the No2ID campaign called the use of such data harvesting tools by the SNP "utterly hypocritical". == Funding == Investors in NationBuilder include Chris Hughes - the Facebook co-founder, Sean Parker - first president of Facebook and co-founder of Napster and Causes, Dan Senor - the former Republican foreign-policy adviser and Ben Horowitz, co-founder of Andreessen Horowitz. In 2012, it has raised $6.3 million in funding from a number of investors. == Notable implementations == The software is reported to have played a role in some public elections in Europe, the US and New Zealand, as well as non-profit initiatives, and political parties in Australia. Notable users include Bernie Sanders, Mitch McConnell, Andrew Yang, Theresa May, Amnesty International, the NAACP and Donald Trump. === France === La République En Marche used NationBuilder in their campaign for the 2017 National Assembly. === New Zealand === NationBuilder's services are used by New Zealand political parties, including in the campaigns of both the National and Labour parties in the 2017 general election. === United Kingdom === Despite stricter data protection and privacy laws in the UK and EU, NationBuilder was used to significant impact in a number of UK elections, most notably in the 2016 campaign for withdrawal of the United Kingdom from the European Union. The company later made a public announcement that both sides in that campaign had used its software. === United States === NationBuilder was used in the Donald Trump presidential campaign to advance his election efforts and eventually win the 2016 presidential race. Jill Stein of the Green Party, Republican Rick Santorum, and independent supporters of various candidates all used NationBuilder during their 2016 runs for president. During the 2018 US election cycle, political entities paid more than $1 million for the use of NationBuilder. Among the entities paying the most were Donald J. Trump for President, Prosperity Action and the Republican Party of Tennessee.
Cross-validation (statistics)
Cross-validation, sometimes called rotation estimation or out-of-sample testing, is any of various similar model validation techniques for assessing how the results of a statistical analysis will generalize to an independent data set. Cross-validation includes resampling and sample splitting methods that use different portions of the data to test and train a model on different iterations. It is often used in settings where the goal is prediction, and one wants to estimate how accurately a predictive model will perform in practice. It can also be used to assess the quality of a fitted model and the stability of its parameters. In a prediction problem, a model is usually given a dataset of known data on which training is run (training dataset), and a dataset of unknown data (or first seen data) against which the model is tested (called the validation dataset or testing set). The goal of cross-validation is to test the model's ability to predict new data that was not used in estimating it, in order to flag problems like overfitting or selection bias and to give an insight on how the model will generalize to an independent dataset (i.e., an unknown dataset, for instance from a real problem). One round of cross-validation involves partitioning a sample of data into complementary subsets, performing the analysis on one subset (called the training set), and validating the analysis on the other subset (called the validation set or testing set). To reduce variability, in most methods multiple rounds of cross-validation are performed using different partitions, and the validation results are combined (e.g. averaged) over the rounds to give an estimate of the model's predictive performance. In summary, cross-validation combines (averages) measures of fitness in prediction to derive a more accurate estimate of model prediction performance. == Motivation == Assume a model with one or more unknown parameters, and a data set to which the model can be fit (the training data set). The fitting process optimizes the model parameters to make the model fit the training data as well as possible. If an independent sample of validation data is taken from the same population as the training data, it will generally turn out that the model does not fit the validation data as well as it fits the training data. The size of this difference is likely to be large especially when the size of the training data set is small, or when the number of parameters in the model is large. Cross-validation is a way to estimate the size of this effect. === Example: linear regression === In linear regression, there exist real response values y 1 , … , y n {\textstyle y_{1},\ldots ,y_{n}} , and n p-dimensional vector covariates x1, ..., xn. The components of the vector xi are denoted xi1, ..., xip. If least squares is used to fit a function in the form of a hyperplane ŷ = a + βTx to the data (xi, yi) 1 ≤ i ≤ n, then the fit can be assessed using the mean squared error (MSE). The MSE for given estimated parameter values a and β on the training set (xi, yi) 1 ≤ i ≤ n is defined as: MSE = 1 n ∑ i = 1 n ( y i − y ^ i ) 2 = 1 n ∑ i = 1 n ( y i − a − β T x i ) 2 = 1 n ∑ i = 1 n ( y i − a − β 1 x i 1 − ⋯ − β p x i p ) 2 {\displaystyle {\begin{aligned}{\text{MSE}}&={\frac {1}{n}}\sum _{i=1}^{n}(y_{i}-{\hat {y}}_{i})^{2}={\frac {1}{n}}\sum _{i=1}^{n}(y_{i}-a-{\boldsymbol {\beta }}^{T}\mathbf {x} _{i})^{2}\\&={\frac {1}{n}}\sum _{i=1}^{n}(y_{i}-a-\beta _{1}x_{i1}-\dots -\beta _{p}x_{ip})^{2}\end{aligned}}} If the model is correctly specified, it can be shown under mild assumptions that the expected value of the MSE for the training set is (n − p − 1)/(n + p + 1) < 1 times the expected value of the MSE for the validation set (the expected value is taken over the distribution of training sets). Thus, a fitted model and computed MSE on the training set will result in an optimistically biased assessment of how well the model will fit an independent data set. This biased estimate is called the in-sample estimate of the fit, whereas the cross-validation estimate is an out-of-sample estimate. Since in linear regression it is possible to directly compute the factor (n − p − 1)/(n + p + 1) by which the training MSE underestimates the validation MSE under the assumption that the model specification is valid, cross-validation can be used for checking whether the model has been overfitted, in which case the MSE in the validation set will substantially exceed its anticipated value. (Cross-validation in the context of linear regression is also useful in that it can be used to select an optimally regularized cost function.) === General case === In most other regression procedures (e.g. logistic regression), there is no simple formula to compute the expected out-of-sample fit. Cross-validation is, thus, a generally applicable way to predict the performance of a model on unavailable data using numerical computation in place of theoretical analysis. == Types == Two types of cross-validation can be distinguished: exhaustive and non-exhaustive cross-validation. === Exhaustive cross-validation === Exhaustive cross-validation methods are cross-validation methods which learn and test on all possible ways to divide the original sample into a training and a validation set. ==== Leave-p-out cross-validation ==== Leave-p-out cross-validation (LpO CV) involves using p observations as the validation set and the remaining observations as the training set. This is repeated on all ways to cut the original sample on a validation set of p observations and a training set. LpO cross-validation require training and validating the model C p n {\displaystyle C_{p}^{n}} times, where n is the number of observations in the original sample, and where C p n {\displaystyle C_{p}^{n}} is the binomial coefficient. For p > 1 and for even moderately large n, LpO CV can become computationally infeasible. For example, with n = 100 and p = 30, C 30 100 ≈ 3 × 10 25 . {\displaystyle C_{30}^{100}\approx 3\times 10^{25}.} A variant of LpO cross-validation with p=2 known as leave-pair-out cross-validation has been recommended as a nearly unbiased method for estimating the area under ROC curve of binary classifiers. ==== Leave-one-out cross-validation ==== Leave-one-out cross-validation (LOOCV) is a particular case of leave-p-out cross-validation with p = 1. The process looks similar to jackknife; however, with cross-validation one computes a statistic on the left-out sample(s), while with jackknifing one computes a statistic from the kept samples only. LOO cross-validation requires less computation time than LpO cross-validation because there are only C 1 n = n {\displaystyle C_{1}^{n}=n} passes rather than C p n {\displaystyle C_{p}^{n}} . However, n {\displaystyle n} passes may still require quite a large computation time, in which case other approaches such as k-fold cross validation may be more appropriate. Pseudo-code algorithm: Input: x, {vector of length N with x-values of incoming points} y, {vector of length N with y-values of the expected result} interpolate( x_in, y_in, x_out ), { returns the estimation for point x_out after the model is trained with x_in-y_in pairs} Output: err, {estimate for the prediction error} Steps: err ← 0 for i ← 1, ..., N do // define the cross-validation subsets x_in ← (x[1], ..., x[i − 1], x[i + 1], ..., x[N]) y_in ← (y[1], ..., y[i − 1], y[i + 1], ..., y[N]) x_out ← x[i] y_out ← interpolate(x_in, y_in, x_out) err ← err + (y[i] − y_out)^2 end for err ← err/N === Non-exhaustive cross-validation === Non-exhaustive cross validation methods do not compute all ways of splitting the original sample. These methods are approximations of leave-p-out cross-validation. ==== k-fold cross-validation ==== In k-fold cross-validation, the original sample is randomly partitioned into k equal sized subsamples, often referred to as "folds". Of the k subsamples, a single subsample is retained as the validation data for testing the model, and the remaining k − 1 subsamples are used as training data. The cross-validation process is then repeated k times, with each of the k subsamples used exactly once as the validation data. The k results can then be averaged to produce a single estimation. The advantage of this method over repeated random sub-sampling (see below) is that all observations are used for both training and validation, and each observation is used for validation exactly once. 10-fold cross-validation is commonly used, but in general k remains an unfixed parameter. For example, setting k = 2 results in 2-fold cross-validation. In 2-fold cross-validation, the dataset is randomly shuffled into two sets d0 and d1, so that both sets are equal size (this is usually implemented by shuffling the data array and then splitting it in two). We then train on d0 and validate on d1, followed by training on d1 and validating on d0. When k = n (the number of observations), k-fold cross-validation is equivalent to leave-one-out cr