The Enterprise Objects Framework, or simply EOF, was introduced by NeXT in 1994 as a pioneering object-relational mapping product for its NeXTSTEP and OpenStep development platforms. EOF abstracts the process of interacting with a relational database by mapping database rows to Java or Objective-C objects. This largely relieves developers from writing low-level SQL code. EOF enjoyed some niche success in the mid-1990s among financial institutions who were attracted to the rapid application development advantages of NeXT's object-oriented platform. Since Apple Inc's merger with NeXT in 1996, EOF has evolved into a fully integrated part of WebObjects, an application server also originally from NeXT. Many of the core concepts of EOF re-emerged as part of Core Data, which further abstracts the underlying data formats to allow it to be based on non-SQL stores. == History == In the early 1990s NeXT Computer recognized that connecting to databases was essential to most businesses and yet also potentially complex. Every data source has a different data-access language (or API), driving up the costs to learn and use each vendor's product. The NeXT engineers wanted to apply the advantages of object-oriented programming, by getting objects to "talk" to relational databases. As the two technologies are very different, the solution was to create an abstraction layer, insulating developers from writing the low-level procedural code (SQL) specific to each data source. The first attempt came in 1992 with the release of Database Kit (DBKit), which wrapped an object-oriented framework around any database. Unfortunately, NEXTSTEP at the time was not powerful enough and DBKit had serious design flaws. NeXT's second attempt came in 1994 with the Enterprise Objects Framework (EOF) version 1, a complete rewrite that was far more modular and OpenStep compatible. EOF 1.0 was the first product released by NeXT using the Foundation Kit and introduced autoreleased objects to the developer community. The development team at the time was only four people: Jack Greenfield, Rich Williamson, Linus Upson and Dan Willhite. EOF 2.0, released in late 1995, further refined the architecture, introducing the editing context. At that point, the development team consisted of Dan Willhite, Craig Federighi, Eric Noyau and Charly Kleissner. EOF achieved a modest level of popularity in the financial programming community in the mid-1990s, but it would come into its own with the emergence of the World Wide Web and the concept of web applications. It was clear that EOF could help companies plug their legacy databases into the Web without any rewriting of that data. With the addition of frameworks to do state management, load balancing and dynamic HTML generation, NeXT was able to launch the first object-oriented Web application server, WebObjects, in 1996, with EOF at its core. In 2000, Apple Inc. (which had merged with NeXT) officially dropped EOF as a standalone product, meaning that developers would be unable to use it to create desktop applications for the forthcoming Mac OS X. It would, however, continue to be an integral part of a major new release of WebObjects. WebObjects 5, released in 2001, was significant for the fact that its frameworks had been ported from their native Objective-C programming language to the Java language. Critics of this change argue that most of the power of EOF was a side effect of its Objective-C roots, and that EOF lost the beauty or simplicity it once had. Third-party tools, such as EOGenerator, help fill the deficiencies introduced by Java (mainly due to the loss of categories). The Objective-C code base was re-introduced with some modifications to desktop application developers as Core Data, part of Apple's Cocoa API, with the release of Mac OS X Tiger in April 2005. == How EOF works == Enterprise Objects provides tools and frameworks for object-relational mapping. The technology specializes in providing mechanisms to retrieve data from various data sources, such as relational databases via JDBC and JNDI directories, and mechanisms to commit data back to those data sources. These mechanisms are designed in a layered, abstract approach that allows developers to think about data retrieval and commitment at a higher level than a specific data source or data source vendor. Central to this mapping is a model file (an "EOModel") that you build with a visual tool — either EOModeler, or the EOModeler plug-in to Xcode. The mapping works as follows: Database tables are mapped to classes. Database columns are mapped to class attributes. Database rows are mapped to objects (or class instances). You can build data models based on existing data sources or you can build data models from scratch, which you then use to create data structures (tables, columns, joins) in a data source. The result is that database records can be transposed into Java objects. The advantage of using data models is that applications are isolated from the idiosyncrasies of the data sources they access. This separation of an application's business logic from database logic allows developers to change the database an application accesses without needing to change the application. EOF provides a level of database transparency not seen in other tools and allows the same model to be used to access different vendor databases and even allows relationships across different vendor databases without changing source code. Its power comes from exposing the underlying data sources as managed graphs of persistent objects. In simple terms, this means that it organizes the application's model layer into a set of defined in-memory data objects. It then tracks changes to these objects and can reverse those changes on demand, such as when a user performs an undo command. Then, when it is time to save changes to the application's data, it archives the objects to the underlying data sources. === Using Inheritance === In designing Enterprise Objects developers can leverage the object-oriented feature known as inheritance. A Customer object and an Employee object, for example, might both inherit certain characteristics from a more generic Person object, such as name, address, and phone number. While this kind of thinking is inherent in object-oriented design, relational databases have no explicit support for inheritance. However, using Enterprise Objects, you can build data models that reflect object hierarchies. That is, you can design database tables to support inheritance by also designing enterprise objects that map to multiple tables or particular views of a database table. == Enterprise Objects (EOs) == An Enterprise Object is analogous to what is often known in object-oriented programming as a business object — a class which models a physical or conceptual object in the business domain (e.g. a customer, an order, an item, etc.). What makes an EO different from other objects is that its instance data maps to a data store. Typically, an enterprise object contains key-value pairs that represent a row in a relational database. The key is basically the column name, and the value is what was in that row in the database. So it can be said that an EO's properties persist beyond the life of any particular running application. More precisely, an Enterprise Object is an instance of a class that implements the com.webobjects.eocontrol.EOEnterpriseObject interface. An Enterprise Object has a corresponding model (called an EOModel) that defines the mapping between the class's object model and the database schema. However, an enterprise object doesn't explicitly know about its model. This level of abstraction means that database vendors can be switched without it affecting the developer's code. This gives Enterprise Objects a high degree of reusability. == EOF and Core Data == Despite their common origins, the two technologies diverged, with each technology retaining a subset of the features of the original Objective-C code base, while adding some new features. === Features Supported Only by EOF === EOF supports custom SQL; shared editing contexts; nested editing contexts; and pre-fetching and batch faulting of relationships, all features of the original Objective-C implementation not supported by Core Data. Core Data also does not provide the equivalent of an EOModelGroup—the NSManagedObjectModel class provides methods for merging models from existing models, and for retrieving merged models from bundles. === Features Supported Only by Core Data === Core Data supports fetched properties; multiple configurations within a managed object model; local stores; and store aggregation (the data for a given entity may be spread across multiple stores); customization and localization of property names and validation warnings; and the use of predicates for property validation. These features of the original Objective-C implementation are not supported by the Java implementation.
DevOps toolchain
A DevOps toolchain is a set or combination of tools that aid in the delivery, development, and management of software applications throughout the systems development life cycle, as coordinated by an organization that uses DevOps practices. Generally, DevOps tools fit into one or more activities, which supports specific DevOps initiatives: Plan, Create, Verify, Package, Release, Configure, Monitor, and Version Control. == Toolchains == In software, a toolchain is the set of programming tools that is used to perform a complex software development task or to create a software product, which is typically another computer program or a set of related programs. In general, the tools forming a toolchain are executed consecutively so the output or resulting environment state of each tool becomes the input or starting environment for the next one, but the term is also used when referring to a set of related tools that are not necessarily executed consecutively. As DevOps is a set of practices that emphasizes the collaboration and communication of both software developers and other information technology (IT) professionals, while automating the process of software delivery and infrastructure changes, its implementation can include the definition of the series of tools used at various stages of the lifecycle; because DevOps is a cultural shift and collaboration between development and operations, there is no one product that can be considered a single DevOps tool. Instead a collection of tools, potentially from a variety of vendors, are used in one or more stages of the lifecycle. == Stages of DevOps == === Plan === Plan consists of two elements: "define" and "plan". This activity refers to the business value and application requirements. Specifically "Plan" activities include: Production metrics, objects and feedback Requirements Business metrics Update release metrics Release plan, timing and business case Security policy and requirement A combination of the IT personnel will be involved in these activities: business application owners, software development, software architects, continual release management, security officers and the organization responsible for managing the production of IT infrastructure. === Create === Create consists of the building, coding, and configuring of the software development process. The specific activities are: Design of the software and configuration Coding including code quality and performance Software build and build performance Release candidate Tools and vendors in this category often overlap with other categories. Because DevOps is about breaking down silos, this is reflective in the activities and product solutions. === Verify === Verify is directly associated with ensuring the quality of the software release; activities designed to ensure code quality is maintained and the highest quality is deployed to production. The main activities in this are: Acceptance testing Regression testing Security and vulnerability analysis Performance Configuration testing Solutions for verify-related activities generally fall under four main categories: Test automation, Static analysis, Test Lab, and Security. === Package === Package refers to the activities involved once the release is ready for deployment, often also referred to as staging or Preproduction / "preprod". This often includes tasks and activities such as: Approval/preapprovals Package configuration Triggered releases Release staging and holding === Release === Release related activities include schedule, orchestration, provisioning and deploying software into production and targeted environment. The specific Release activities include: Release coordination Deploying and promoting applications Fallbacks and recovery Scheduled/timed releases Solutions that cover this aspect of the toolchain include application release automation, deployment automation and release management. === Configure === Configure activities fall under the operation side of DevOps. Once software is deployed, there may be additional IT infrastructure provisioning and configuration activities required. Specific activities including: Infrastructure storage, database and network provisioning and configuring Application provision and configuration. The main types of solutions that facilitate these activities are continuous configuration automation, configuration management, and infrastructure as code tools. === Monitor === Monitoring is an important link in a DevOps toolchain. It allows IT organization to identify specific issues of specific releases and to understand the impact on end-users. A summary of Monitor related activities are: Performance of IT infrastructure End-user response and experience Production metrics and statistics Information from monitoring activities often impacts Plan activities required for changes and for new release cycles. === Version Control === Version Control is an important link in a DevOps toolchain and a component of software configuration management. Version Control is the management of changes to documents, computer programs, large web sites, and other collections of information. A summary of Version Control related activities are: Non-linear development Distributed development Compatibility with existent systems and protocols Toolkit-based design Information from Version Control often supports Release activities required for changes and for new release cycles.
Artificial intelligence in government
Artificial intelligence (AI) has a range of uses in government. It can be used to further public policy objectives (in areas such as emergency services, health and welfare), as well as assist the public to interact with the government (through the use of virtual assistants, for example). According to the Harvard Business Review, "Applications of artificial intelligence to the public sector are broad and growing, with early experiments taking place around the world." Hila Mehr from the Ash Center for Democratic Governance and Innovation at Harvard University notes that AI in government is not new, with postal services using machine methods in the late 1990s to recognise handwriting on envelopes to automatically route letters. The use of AI in government comes with significant benefits, including efficiencies resulting in cost savings (for instance by reducing the number of front office staff) and reducing the opportunities for corruption. However, it also carries risks (described below). == Uses of AI in government == The potential uses of AI in government are wide and varied, with Deloitte considering that "Cognitive technologies could eventually revolutionize every facet of government operations". Mehr suggests that six types of government problems are appropriate for AI applications: Resource allocation—such as where administrative support is required to complete tasks more quickly. Large datasets—where these are too large for employees to work efficiently and multiple datasets could be combined to provide greater insights. Expert shortage—including where basic questions could be answered and niche issues can be learned. Predictable scenario—historical data makes the situation predictable. Procedural tasks refer to repetitive tasks in which the answers to inputs or outputs are binary. Diverse data—where data takes various forms (such as visual and linguistic) and needs to be summarized regularly. Mehr states that "While applications of AI in government work have not kept pace with the rapid expansion of AI in the private sector, the potential use cases in the public sector mirror common applications in the private sector." Potential and actual uses of AI in government can be divided into three broad categories: those that contribute to public policy objectives, those that assist public interactions with the government, and other uses. === Contributing to public policy objectives === There are a range of examples of where AI can contribute to public policy objectives. These include: Receiving benefits at job loss, retirement, bereavement and child birth almost immediately, in an automated way (thus without requiring any actions from citizens at all) Social insurance service provision Classifying emergency calls based on their urgency (like the system used by the Cincinnati Fire Department in the United States) Detecting and preventing the spread of diseases Assisting public servants in making welfare payments and immigration decisions Adjudicating bail hearings Triaging health care cases Monitoring social media for public feedback on policies Monitoring social media to identify emergency situations Identifying fraudulent benefits claims Predicting a crime and recommending optimal police presence Predicting traffic congestion and car accidents Anticipating road maintenance requirements Identifying breaches of health regulations Providing personalised education to students Marking exam papers Assisting with defence and national security (see Artificial intelligence § Military and Applications of artificial intelligence § Other fields in which AI methods are implemented respectively) Artificial Intelligence in China has been used to drive both political and economic markets. In 2019, Shanghai’s government rolled out 100 billion yuan to assist in funding enterprises that used AI to introduce 22 new policy agendas. Shanghai invested in these enterprises to attract top international talent in order to set up the Shanghai Municipal Big Data Center. City Brain AI is an urban management platform made by Alibaba. China uses City Brain AI to maintain a significant share of capital investment through public and state owned enterprises. The synergy between public and private sectors are more than capital-driven with City Brain AI. The blend of both public and private shareholding is only made out to be through the role of provincial and sub-provincial governments. Both hold control over the direction that City Brain AI makes both socially and economically. === Assisting public interactions with government === AI can be used to assist members of the public to interact with government and access government services, for example by: Answering questions using virtual assistants or chatbots (see below) Directing requests to the appropriate area within government Filling out forms Assisting with searching documents (e.g. IP Australia's trade mark search) Scheduling appointments Various governments, including those of Australia and Estonia, have implemented virtual assistants to aid citizens in navigating services, with applications ranging from tax inquiries to life-event registrations. === Gerrymandering === Gerrymandering is a method of influencing political process by drawing map boundaries in favor of incumbent parties. Academic researchers Wendy Tam Cho and Bruce Cain have proposed partially automating the map-drawing process with an AI system to reduce partisan gerrymandering. Even with this AI system, the process may still be manipulated to favor partisan interests, so the researchers emphasized the importance of transparency and human involvement. === Other uses === Other uses of AI in government include: Translation Language interpretation pioneered by the European Commission's Directorate General for Interpretation and Florika Fink-Hooijer. Drafting documents == Potential benefits == AI offers potential efficiencies and cost savings for the government. For example, Deloitte has estimated that automation could save US Government employees between 96.7 million to 1.2 billion hours a year, resulting in potential savings of between $3.3 billion to $41.1 billion a year. The Harvard Business Review has stated that while this may lead a government to reduce employee numbers, "Governments could instead choose to invest in the quality of its services. They can re-employ workers' time towards more rewarding work that requires lateral thinking, empathy, and creativity—all things at which humans continue to outperform even the most sophisticated AI program." == Risks == Risks associated with the use of AI in government include AI becoming susceptible to bias, a lack of transparency in how an AI application may make decisions, and the accountability for any such decisions. For example, a 2026 lawsuit alleged that the U.S. Department of Government Efficiency used ChatGPT to flag and cancel federal humanities grants, including projects on Jewish history and Israeli culture, over some objections from NEH officials, illustrating how automated decision-making could affect funding outcomes.
Transaction data
Transaction data or transaction information is a category of data describing transactions. Transaction data/information gather variables generally referring to reference data or master data – e.g. dates, times, time zones, currencies. Typical transactions are: Financial transactions about orders, invoices, payments; Work transactions about plans, activity records; Logistic transactions about deliveries, storage records, travel records, etc.. == Management == Recording and storing transactions is called records management. The record of the transaction is stored in a place where the retention can be guaranteed and where data is archived or removed following a retention period. Formats of recorded transactions can be digital data in databases and spreadsheets, or handwritten texts in physical documents like former bankbooks. Transaction processing systems are application software that generate transactions and manage transaction data/information, e.g. SAP and Oracle Financials. == Data warehousing == Transaction data can be summarised in a data warehouse, which helps accessibility and analysis of the data.
Birkhoff algorithm
Birkhoff's algorithm (also called Birkhoff-von-Neumann algorithm) is an algorithm for decomposing a bistochastic matrix into a convex combination of permutation matrices. It was published by Garrett Birkhoff in 1946. It has many applications. One such application is for the problem of fair random assignment: given a randomized allocation of items, Birkhoff's algorithm can decompose it into a lottery on deterministic allocations. == Terminology == A bistochastic matrix (also called: doubly-stochastic) is a matrix in which all elements are greater than or equal to 0 and the sum of the elements in each row and column equals 1. An example is the following 3-by-3 matrix: ( 0.2 0.3 0.5 0.6 0.2 0.2 0.2 0.5 0.3 ) {\displaystyle {\begin{pmatrix}0.2&0.3&0.5\\0.6&0.2&0.2\\0.2&0.5&0.3\end{pmatrix}}} A permutation matrix is a special case of a bistochastic matrix, in which each element is either 0 or 1 (so there is exactly one "1" in each row and each column). An example is the following 3-by-3 matrix: ( 0 1 0 0 0 1 1 0 0 ) {\displaystyle {\begin{pmatrix}0&1&0\\0&0&1\\1&0&0\end{pmatrix}}} A Birkhoff decomposition (also called: Birkhoff-von-Neumann decomposition) of a bistochastic matrix is a presentation of it as a sum of permutation matrices with non-negative weights. For example, the above matrix can be presented as the following sum: 0.2 ( 0 1 0 0 0 1 1 0 0 ) + 0.2 ( 1 0 0 0 1 0 0 0 1 ) + 0.1 ( 0 1 0 1 0 0 0 0 1 ) + 0.5 ( 0 0 1 1 0 0 0 1 0 ) {\displaystyle 0.2{\begin{pmatrix}0&1&0\\0&0&1\\1&0&0\end{pmatrix}}+0.2{\begin{pmatrix}1&0&0\\0&1&0\\0&0&1\end{pmatrix}}+0.1{\begin{pmatrix}0&1&0\\1&0&0\\0&0&1\end{pmatrix}}+0.5{\begin{pmatrix}0&0&1\\1&0&0\\0&1&0\end{pmatrix}}} Birkhoff's algorithm receives as input a bistochastic matrix and returns as output a Birkhoff decomposition. == Tools == A permutation set of an n-by-n matrix X is a set of n entries of X containing exactly one entry from each row and from each column. A theorem by Dénes Kőnig says that: Every bistochastic matrix has a permutation-set in which all entries are positive.The positivity graph of an n-by-n matrix X is a bipartite graph with 2n vertices, in which the vertices on one side are n rows and the vertices on the other side are the n columns, and there is an edge between a row and a column if the entry at that row and column is positive. A permutation set with positive entries is equivalent to a perfect matching in the positivity graph. A perfect matching in a bipartite graph can be found in polynomial time, e.g. using any algorithm for maximum cardinality matching. Kőnig's theorem is equivalent to the following:The positivity graph of any bistochastic matrix admits a perfect matching.A matrix is called scaled-bistochastic if all elements are non-negative, and the sum of each row and column equals c, where c is some positive constant. In other words, it is c times a bistochastic matrix. Since the positivity graph is not affected by scaling:The positivity graph of any scaled-bistochastic matrix admits a perfect matching. == Algorithm == Birkhoff's algorithm is a greedy algorithm: it greedily finds perfect matchings and removes them from the fractional matching. It works as follows. Let i = 1. Construct the positivity graph GX of X. Find a perfect matching in GX, corresponding to a positive permutation set in X. Let z[i] > 0 be the smallest entry in the permutation set. Let P[i] be a permutation matrix with 1 in the positive permutation set. Let X := X − z[i] P[i]. If X contains nonzero elements, Let i = i + 1 and go back to step 2. Otherwise, return the sum: z[1] P[1] + ... + z[2] P[2] + ... + z[i] P[i]. The algorithm is correct because, after step 6, the sum in each row and each column drops by z[i]. Therefore, the matrix X remains scaled-bistochastic. Therefore, in step 3, a perfect matching always exists. == Run-time complexity == By the selection of z[i] in step 4, in each iteration at least one element of X becomes 0. Therefore, the algorithm must end after at most n2 steps. However, the last step must simultaneously make n elements 0, so the algorithm ends after at most n2 − n + 1 steps, which implies O ( n 2 ) {\displaystyle O(n^{2})} . In 1960, Joshnson, Dulmage and Mendelsohn showed that Birkhoff's algorithm actually ends after at most n2 − 2n + 2 steps, which is tight in general (that is, in some cases n2 − 2n + 2 permutation matrices may be required). == Application in fair division == In the fair random assignment problem, there are n objects and n people with different preferences over the objects. It is required to give an object to each person. To attain fairness, the allocation is randomized: for each (person, object) pair, a probability is calculated, such that the sum of probabilities for each person and for each object is 1. The probabilistic-serial procedure can compute the probabilities such that each agent, looking at the matrix of probabilities, prefers his row of probabilities over the rows of all other people (this property is called envy-freeness). This raises the question of how to implement this randomized allocation in practice? One cannot just randomize for each object separately, since this may result in allocations in which some people get many objects while other people get no objects. Here, Birkhoff's algorithm is useful. The matrix of probabilities, calculated by the probabilistic-serial algorithm, is bistochastic. Birkhoff's algorithm can decompose it into a convex combination of permutation matrices. Each permutation matrix represents a deterministic assignment, in which every agent receives exactly one object. The coefficient of each such matrix is interpreted as a probability; based on the calculated probabilities, it is possible to pick one assignment at random and implement it. == Extensions == The problem of computing the Birkhoff decomposition with the minimum number of terms has been shown to be NP-hard, but some heuristics for computing it are known. This theorem can be extended for the general stochastic matrix with deterministic transition matrices. Budish, Che, Kojima and Milgrom generalize Birkhoff's algorithm to non-square matrices, with some constraints on the feasible assignments. They also present a decomposition algorithm that minimizes the variance in the expected values. Vazirani generalizes Birkhoff's algorithm to non-bipartite graphs. Valls et al. showed that it is possible to obtain an ϵ {\displaystyle \epsilon } -approximate decomposition with O ( log ( 1 / ϵ 2 ) ) {\displaystyle O(\log(1/\epsilon ^{2}))} permutations.
Neural radiance field
A neural radiance field (NeRF) is a neural field for reconstructing a three-dimensional representation of a scene from two-dimensional images. The NeRF model enables downstream applications of novel view synthesis, scene geometry reconstruction, and obtaining the reflectance properties of the scene. Additional scene properties such as camera poses may also be jointly learned. First introduced in 2020, it has since gained significant attention for its potential applications in computer graphics and content creation. == Algorithm == The NeRF algorithm represents a scene as a radiance field parametrized by a deep neural network (DNN). The network predicts a volume density and view-dependent emitted radiance given the spatial location ( x , y , z ) {\displaystyle (x,y,z)} and viewing direction in Euler angles ( θ , Φ ) {\displaystyle (\theta ,\Phi )} of the camera. By sampling many points along camera rays, traditional volume rendering techniques can produce an image. === Data collection === A NeRF needs to be retrained for each unique scene. The first step is to collect images of the scene from different angles and their respective camera pose. These images are standard 2D images and do not require a specialized camera or software. Any camera is able to generate datasets, provided the settings and capture method meet the requirements for SfM (Structure from Motion). This requires tracking of the camera position and orientation, often through some combination of SLAM, GPS, or inertial estimation. Researchers often use synthetic data to evaluate NeRF and related techniques. For such data, images (rendered through traditional non-learned methods) and respective camera poses are reproducible and error-free. === Training === For each sparse viewpoint (image and camera pose) provided, camera rays are marched through the scene, generating a set of 3D points with a given radiance direction (into the camera). For these points, volume density and emitted radiance are predicted using the multi-layer perceptron (MLP). An image is then generated through classical volume rendering. Because this process is fully differentiable, the error between the predicted image and the original image can be minimized with gradient descent over multiple viewpoints, encouraging the MLP to develop a coherent model of the scene. == Variations and improvements == Early versions of NeRF were slow to optimize and required that all input views were taken with the same camera in the same lighting conditions. These performed best when limited to orbiting around individual objects, such as a drum set, plants or small toys. Since the original paper in 2020, many improvements have been made to the NeRF algorithm, with variations for special use cases. === Fourier feature mapping === In 2020, shortly after the release of NeRF, the addition of Fourier Feature Mapping improved training speed and image accuracy. Deep neural networks struggle to learn high frequency functions in low dimensional domains; a phenomenon known as spectral bias. To overcome this shortcoming, points are mapped to a higher dimensional feature space before being fed into the MLP. γ ( v ) = [ a 1 cos ( 2 π B 1 T v ) a 1 sin ( 2 π B 1 T v ) ⋮ a m cos ( 2 π B m T v ) a m sin ( 2 π B m T v ) ] {\displaystyle \gamma (\mathrm {v} )={\begin{bmatrix}a_{1}\cos(2{\pi }{\mathrm {B} }_{1}^{T}\mathrm {v} )\\a_{1}\sin(2\pi {\mathrm {B} }_{1}^{T}\mathrm {v} )\\\vdots \\a_{m}\cos(2{\pi }{\mathrm {B} }_{m}^{T}\mathrm {v} )\\a_{m}\sin(2{\pi }{\mathrm {B} }_{m}^{T}\mathrm {v} )\end{bmatrix}}} Where v {\displaystyle \mathrm {v} } is the input point, B i {\displaystyle \mathrm {B} _{i}} are the frequency vectors, and a i {\displaystyle a_{i}} are coefficients. This allows for rapid convergence to high frequency functions, such as pixels in a detailed image. === Bundle-adjusting neural radiance fields === One limitation of NeRFs is the requirement of knowing accurate camera poses to train the model. Often times, pose estimation methods are not completely accurate, nor is the camera pose even possible to know. These imperfections result in artifacts and suboptimal convergence. So, a method was developed to optimize the camera pose along with the volumetric function itself. Called Bundle-Adjusting Neural Radiance Field (BARF), the technique uses a dynamic low-pass filter (DLPF) to go from coarse to fine adjustment, minimizing error by finding the geometric transformation to the desired image. This corrects imperfect camera poses and greatly improves the quality of NeRF renders. === Multiscale representation === Conventional NeRFs struggle to represent detail at all viewing distances, producing blurry images up close and overly aliased images from distant views. In 2021, researchers introduced a technique to improve the sharpness of details at different viewing scales known as mip-NeRF (comes from mipmap). Rather than sampling a single ray per pixel, the technique fits a gaussian to the conical frustum cast by the camera. This improvement effectively anti-aliases across all viewing scales. mip-NeRF also reduces overall image error and is faster to converge at about half the size of ray-based NeRF. === Learned initializations === In 2021, researchers applied meta-learning to assign initial weights to the MLP. This rapidly speeds up convergence by effectively giving the network a head start in gradient descent. Meta-learning also allowed the MLP to learn an underlying representation of certain scene types. For example, given a dataset of famous tourist landmarks, an initialized NeRF could partially reconstruct a scene given one image. === NeRF in the wild === Conventional NeRFs are vulnerable to slight variations in input images (objects, lighting) often resulting in ghosting and artifacts. As a result, NeRFs struggle to represent dynamic scenes, such as bustling city streets with changes in lighting and dynamic objects. In 2021, researchers at Google developed a new method for accounting for these variations, named NeRF in the Wild (NeRF-W). This method splits the neural network (MLP) into three separate models. The main MLP is retained to encode the static volumetric radiance. However, it operates in sequence with a separate MLP for appearance embedding (changes in lighting, camera properties) and an MLP for transient embedding (changes in scene objects). This allows the NeRF to be trained on diverse photo collections, such as those taken by mobile phones at different times of day. === Relighting === In 2021, researchers added more outputs to the MLP at the heart of NeRFs. The output now included: volume density, surface normal, material parameters, distance to the first surface intersection (in any direction), and visibility of the external environment in any direction. The inclusion of these new parameters lets the MLP learn material properties, rather than pure radiance values. This facilitates a more complex rendering pipeline, calculating direct and global illumination, specular highlights, and shadows. As a result, the NeRF can render the scene under any lighting conditions with no re-training. === Plenoctrees === Although NeRFs had reached high levels of fidelity, their costly compute time made them useless for many applications requiring real-time rendering, such as VR/AR and interactive content. Introduced in 2021, Plenoctrees (plenoptic octrees) enabled real-time rendering of pre-trained NeRFs through division of the volumetric radiance function into an octree. Rather than assigning a radiance direction into the camera, viewing direction is taken out of the network input and spherical radiance is predicted for each region. This makes rendering over 3000x faster than conventional NeRFs. === Sparse Neural Radiance Grid === Similar to Plenoctrees, this method enabled real-time rendering of pretrained NeRFs. To avoid querying the large MLP for each point, this method bakes NeRFs into Sparse Neural Radiance Grids (SNeRG). A SNeRG is a sparse voxel grid containing opacity and color, with learned feature vectors to encode view-dependent information. A lightweight, more efficient MLP is then used to produce view-dependent residuals to modify the color and opacity. To enable this compressive baking, small changes to the NeRF architecture were made, such as running the MLP once per pixel rather than for each point along the ray. These improvements make SNeRG extremely efficient, outperforming Plenoctrees. === Instant NeRFs === In 2022, researchers at Nvidia enabled real-time training of NeRFs through a technique known as Instant Neural Graphics Primitives. An innovative input encoding reduces computation, enabling real-time training of a NeRF, an improvement orders of magnitude above previous methods. The speedup stems from the use of spatial hash functions, which have O ( 1 ) {\displaystyle O(1)} access times, and parallelized architectures which run fast on modern GPUs. == Related techniques == === Plenoxels === Plen
Aidoc
Aidoc Medical is an Israeli technology company that develops computer-aided simple triage and notification systems. Aidoc has obtained U.S. Food and Drug Administration and CE mark approval for its stroke, pulmonary embolism, cervical fracture, intracranial hemorrhage, intra-abdominal free gas, and incidental pulmonary embolism algorithms. Aidoc algorithms are in use in more than 900 hospitals and imaging centers, including Montefiore Nyack Hospital, LifeBridge Health, LucidHealth, Yale New Haven Hospital, Cedars-Sinai Medical Center, University of Rochester Medical Center, and Sheba Medical Center. == History == Aidoc was founded in 2016 by Elad Walach as the CEO, Michael Braginsky as the CTO and Guy Reiner as the VP. In April 2017, the company raised $7M, led by TLV Partners, and in April 2019, the company raised another $27M, led by Square Peg capital. There have been several additional rounds of funding as well, bringing Aidoc's total investment to $370M as of July 2025. In August 2018, Aidoc gained FDA clearance for its intracranial hemorrhage system, and in May 2019 it received clearance for the pulmonary embolism system. In January 2020, the system for detecting large-vessel occlusions (LVOs) in head CTA examinations obtained FDA clearance. In October 2024, it was reported that Aidoc is working with NVIDIA to develop a framework for deployment and integration of artificial intelligence tools in healthcare. The Blueprint for Resilient Integration and Deployment of Guided Excellence (BRIDGE) is a guideline to facilitate AI adoption in the healthcare industry. == Products and market == Aidoc has developed a suite of artificial intelligence products that flag both time-sensitive and time-consuming (for the radiologist) abnormalities across the body. The algorithms are developed with large quantities of data to provide diagnostic aid for a broad set of pathologies. The company offers an array of algorithms that span across the body, including for intracranial hemorrhage, spine fractures (C, T & L), free air in the abdomen, pulmonary embolism, and more. It developed "Always-on AI", a term coined by Elad Walach that refers to a type of artificial intelligence that is "Always-on—constantly running in the background and automatically analyzing medical imaging data, identifying urgent findings, and sparing radiologists from "drowning" in vast amounts of irrelevant data. Aidoc's solutions cover medical conditions prevalent in all settings (ED/inpatient/outpatient), including level 1 trauma centers, outpatient imaging centers, teleradiology groups and, are set up in over 200 medical centers worldwide. Notable customers include the University of Rochester Medical Center and Global Diagnostics Australia. Aidoc announced in 2024 that its new Clinical AI Reasoning Engine (CARE1) had been submitted for FDA approval. In September 2025 Aidoc received a "Breakthrough Device Designation" from the FDA for a new multi-triage solution that spans numerous acute findings in CT scans. Aidoc's CARE1 foundation model was the basis of the workflow on which the designation was made, enabling simultaneous coverage of multiple pathologies. This new designation allows parallel FDA review of multiple indications under a single submission. In April 2026, Aidoc raised million in a Series E funding round led by Growth Equity at Goldman Sachs Alternatives, with participation from General Catalyst and NVentures. The financing brought the company's total funding to over million. == Clinical Research == A clinical study on Aidoc’ accuracy of deep convolutional neural networks for the detection of pulmonary embolism (PE) on CT pulmonary angiograms (CTPAs) was performed by the University Hospital of Basel and presented at the European Congress of Radiology, showing that the Aidoc algorithm reached 93% sensitivity and 95% specificity. Clinical research has also been performed to test the diagnostic performance of Aidoc's deep learning-based triage system for the flagging of acute findings in abdominal computed tomography (CT) examinations. Overall, the algorithm achieved 93% sensitivity (91/98, 7 false negatives) and 97% specificity (93/96, 3 false-positive) in the detection of acute abdominal findings. Additional clinical research on Aidoc's Intracranial hemorrhage algorithm accuracy was presented at the European Congress of Radiology by Antwerp University Hospital, evaluating the use of its deep learning algorithm for the detection of intracranial hemorrhage on non-contrast enhanced CT of the brain. The University of Washington completed a study on the accuracy of Aidoc's intracranial hemorrhage algorithm.