AI Grammar Detection

AI Grammar Detection — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

National Parking Platform

The National Parking Platform is a digital platform in the United Kingdom providing interoperability between car park operators, parking apps, and other service providers. It enables all parking apps that support the system: RingGo, JustPark, PayByPhone, Apcoa Connect, AppyParking, and Caura to work at all participating car parks. It has been rolled out in 13 local authorities so far. It was first developed by the Department for Transport starting in 2019, and since May 2025 is controlled by the British Parking Association on a not-for-profit basis. == Participating local authorities == Buckinghamshire Cheshire West and Chester Coventry City East Hertfordshire East Suffolk Liverpool City Manchester City Oxfordshire County Peterborough City Stevenage Sutton Walsall Welwyn Hatfield
Read more →
Inverse consistency

In image registration, inverse consistency measures the consistency of mappings between images produced by a registration algorithm. The inverse consistency error, introduced by Christiansen and Johnson in 2001, quantifies the distance between the composition of the mappings from each image to the other, produced by the registration procedure, and the identity function, and is used as a regularisation constraint in the loss function of many registration algorithms to enforce consistent mappings. Inverse consistency is necessary for good image registration but it is not sufficient, since a mapping can be perfectly consistent but not register the images at all. == Definition == Image registration is the process of establishing a common coordinate system between two images, and given two images I 1 : Ω 1 → R I 2 : Ω 2 → R {\displaystyle {\begin{aligned}I_{1}:\Omega _{1}\to \mathbb {R} \\I_{2}:\Omega _{2}\to \mathbb {R} \end{aligned}}} registering a source image I 1 {\displaystyle I_{1}} to a target image I 2 {\displaystyle I_{2}} consists of determining a transformation f 1 : Ω 2 → Ω 1 {\displaystyle f_{1}:\Omega _{2}\to \Omega _{1}} that maps points from the target space to the source space. An ideal registration algorithm should not be sensitive to which image in the pair is used as source or target, and the registration operator should be antisymmetric such that the mappings f 1 : Ω 2 → Ω 1 f 2 : Ω 1 → Ω 2 {\displaystyle {\begin{aligned}f_{1}:\Omega _{2}\to \Omega _{1}\\f_{2}:\Omega _{1}\to \Omega _{2}\end{aligned}}} produced when registering I 1 {\displaystyle I_{1}} to I 2 {\displaystyle I_{2}} and I 2 {\displaystyle I_{2}} to I 1 {\displaystyle I_{1}} respectively should be the inverse of each other, i.e. f 2 = f 1 − 1 {\displaystyle f_{2}=f_{1}^{-1}} and f 1 = f 2 − 1 {\displaystyle f_{1}=f_{2}^{-1}} or, equivalently, f 2 ∘ f 1 = id Ω 2 {\displaystyle f_{2}\circ f_{1}=\operatorname {id} _{\Omega _{2}}} and f 1 ∘ f 2 = id Ω 1 {\displaystyle f_{1}\circ f_{2}=\operatorname {id} _{\Omega _{1}}} , where ∘ {\displaystyle \circ } denotes the function composition operator. Real algorithms are not perfect, and when swapping the role of source and target image in a registration problem the so obtained transformations are not the inverse of each other. Inverse consistency can be enforced by adding to the loss function of the registration a symmetric regularisation term that penalises inconsistent transformations ∫ Ω 2 ‖ f 2 ( f 1 ( x ) ) − x ‖ 2 d x + ∫ Ω 1 ‖ f 1 ( f 2 ( x ) ) − x ‖ 2 d x . {\displaystyle \int _{\Omega _{2}}\left\Vert f_{2}(f_{1}(x))-x\right\Vert ^{2}\mathrm {d} x+\int _{\Omega _{1}}\left\Vert f_{1}(f_{2}(x))-x\right\Vert ^{2}\mathrm {d} x.} Inverse consistency can be used as a quality metric to evaluate image registration results. The inverse consistency error ( I C E {\displaystyle ICE} ) measures the distance between the composition of the two transforms and the identity function, and it can be formulated in terms of both average ( I C E a {\displaystyle ICE_{a}} ) or maximum ( I C E m {\displaystyle ICE_{m}} ) over a region of interest Ω {\displaystyle \Omega } of the image: I C E a = 1 ∫ Ω d x ∫ Ω ‖ f 2 ( f 1 ( x ) ) − x ‖ d x I C E m = max x ∈ Ω ‖ f 2 ( f 1 ( x ) ) − x ‖ . {\displaystyle {\begin{aligned}ICE_{a}&={\frac {1}{\int _{\Omega }\mathrm {d} x}}\int _{\Omega }\left\Vert f_{2}(f_{1}(x))-x\right\Vert \mathrm {d} x\\ICE_{m}&=\max _{x\in \Omega }\left\Vert f_{2}(f_{1}(x))-x\right\Vert .\end{aligned}}} While inverse consistency is a necessary property of good registration algorithms, inverse consistency error alone is not a sufficient metric to evaluate the quality of image registration results, since a perfectly consistent mapping, with no other constraint, may be not even close to correctly register a pair of images.
Read more →
Weight initialization

In deep learning, weight initialization or parameter initialization describes the initial step in creating a neural network. A neural network contains trainable parameters that are modified during training: weight initialization is the pre-training step of assigning initial values to these parameters. The choice of weight initialization method affects the speed of convergence, the scale of neural activation within the network, the scale of gradient signals during backpropagation, and the quality of the final model. Proper initialization is necessary for avoiding issues such as vanishing and exploding gradients and activation function saturation. Note that even though this article is titled "weight initialization", both weights and biases are used in a neural network as trainable parameters, so this article describes how both of these are initialized. Similarly, trainable parameters in convolutional neural networks (CNNs) are called kernels and biases, and this article also describes these. == Constant initialization == We discuss the main methods of initialization in the context of a multilayer perceptron (MLP). Specific strategies for initializing other network architectures are discussed in later sections. For an MLP, there are only two kinds of trainable parameters, called weights and biases. Each layer l {\displaystyle l} contains a weight matrix W ( l ) ∈ R n l − 1 × n l {\displaystyle W^{(l)}\in \mathbb {R} ^{n_{l-1}\times n_{l}}} and a bias vector b ( l ) ∈ R n l {\displaystyle b^{(l)}\in \mathbb {R} ^{n_{l}}} , where n l {\displaystyle n_{l}} is the number of neurons in that layer. A weight initialization method is an algorithm for setting the initial values for W ( l ) , b ( l ) {\displaystyle W^{(l)},b^{(l)}} for each layer l {\displaystyle l} . The simplest form is zero initialization: W ( l ) = 0 , b ( l ) = 0 {\displaystyle W^{(l)}=0,b^{(l)}=0} Zero initialization is usually used for initializing biases, but it is not used for initializing weights, as it leads to symmetry in the network, causing all neurons to learn the same features. In this page, we assume b = 0 {\displaystyle b=0} unless otherwise stated. Recurrent neural networks typically use activation functions with bounded range, such as sigmoid and tanh, since unbounded activation may cause exploding values. (Le, Jaitly, Hinton, 2015) suggested initializing weights in the recurrent parts of the network to identity and zero bias, similar to the idea of residual connections and LSTM with no forget gate. In most cases, the biases are initialized to zero, though some situations can use a nonzero initialization. For example, in multiplicative units, such as the forget gate of LSTM, the bias can be initialized to 1 to allow good gradient signal through the gate. For neurons with ReLU activation, one can initialize the bias to a small positive value like 0.1, so that the gradient is likely nonzero at initialization, avoiding the dying ReLU problem. == Random initialization == Random initialization means sampling the weights from a normal distribution or a uniform distribution, usually independently. === LeCun initialization === LeCun initialization, popularized in (LeCun et al., 1998), is designed to preserve the variance of neural activations during the forward pass. It samples each entry in W ( l ) {\displaystyle W^{(l)}} independently from a distribution with mean 0 and variance 1 / n l − 1 {\displaystyle 1/n_{l-1}} . For example, if the distribution is a continuous uniform distribution, then the distribution is U ( ± 3 / n l − 1 ) {\displaystyle {\mathcal {U}}(\pm {\sqrt {3/n_{l-1}}})} . === Glorot initialization === Glorot initialization (or Xavier initialization) was proposed by Xavier Glorot and Yoshua Bengio. It was designed as a compromise between two goals: to preserve activation variance during the forward pass and to preserve gradient variance during the backward pass. For uniform initialization, it samples each entry in W ( l ) {\displaystyle W^{(l)}} independently and identically from U ( ± 6 / ( n l + 1 + n l − 1 ) ) {\displaystyle {\mathcal {U}}(\pm {\sqrt {6/(n_{l+1}+n_{l-1})}})} . In the context, n l − 1 {\displaystyle n_{l-1}} is also called the "fan-in", and n l + 1 {\displaystyle n_{l+1}} the "fan-out". When the fan-in and fan-out are equal, then Glorot initialization is the same as LeCun initialization. === He initialization === As Glorot initialization performs poorly for ReLU activation, He initialization (or Kaiming initialization) was proposed by Kaiming He et al. for networks with ReLU activation. It samples each entry in W ( l ) {\displaystyle W^{(l)}} from N ( 0 , 2 / n l − 1 ) {\displaystyle {\mathcal {N}}(0,2/n_{l-1})} . === Orthogonal initialization === (Saxe et al. 2013) proposed orthogonal initialization: initializing weight matrices as uniformly random (according to the Haar measure) semi-orthogonal matrices, multiplied by a factor that depends on the activation function of the layer. It was designed so that if one initializes a deep linear network this way, then its training time until convergence is independent of depth. Sampling a uniformly random semi-orthogonal matrix can be done by initializing X {\displaystyle X} by IID sampling its entries from a standard normal distribution, then calculate ( X X ⊤ ) − 1 / 2 X {\displaystyle \left(XX^{\top }\right)^{-1/2}X} or its transpose, depending on whether X {\displaystyle X} is tall or wide. For CNN kernels with odd widths and heights, orthogonal initialization is done this way: initialize the central point by a semi-orthogonal matrix, and fill the other entries with zero. As an illustration, a kernel K {\displaystyle K} of shape 3 × 3 × c × c ′ {\displaystyle 3\times 3\times c\times c'} is initialized by filling K [ 2 , 2 , : , : ] {\displaystyle K[2,2,:,:]} with the entries of a random semi-orthogonal matrix of shape c × c ′ {\displaystyle c\times c'} , and the other entries with zero. (Balduzzi et al., 2017) used it with stride 1 and zero-padding. This is sometimes called the Orthogonal Delta initialization. Related to this approach, unitary initialization proposes to parameterize the weight matrices to be unitary matrices, with the result that at initialization they are random unitary matrices (and throughout training, they remain unitary). This is found to improve long-sequence modelling in LSTM. Orthogonal initialization has been generalized to layer-sequential unit-variance (LSUV) initialization. It is a data-dependent initialization method, and can be used in convolutional neural networks. It first initializes weights of each convolution or fully connected layer with orthonormal matrices. Then, proceeding from the first to the last layer, it runs a forward pass on a random minibatch, and divides the layer's weights by the standard deviation of its output, so that its output has variance approximately 1. === Fixup initialization === In 2015, the introduction of residual connections allowed very deep neural networks to be trained, much deeper than the ~20 layers of the previous state of the art (such as the VGG-19). Residual connections gave rise to their own weight initialization problems and strategies. These are sometimes called "normalization-free" methods, since using residual connection could stabilize the training of a deep neural network so much that normalizations become unnecessary. Fixup initialization is designed specifically for networks with residual connections and without batch normalization, as follows: Initialize the classification layer and the last layer of each residual branch to 0. Initialize every other layer using a standard method (such as He initialization), and scale only the weight layers inside residual branches by L − 1 2 m − 2 {\displaystyle L^{-{\frac {1}{2m-2}}}} . Add a scalar multiplier (initialized at 1) in every branch and a scalar bias (initialized at 0) before each convolution, linear, and element-wise activation layer. Similarly, T-Fixup initialization is designed for Transformers without layer normalization. === Others === Instead of initializing all weights with random values on the order of O ( 1 / n ) {\displaystyle O(1/{\sqrt {n}})} , sparse initialization initialized only a small subset of the weights with larger random values, and the other weights zero, so that the total variance is still on the order of O ( 1 ) {\displaystyle O(1)} . Random walk initialization was designed for MLP so that during backpropagation, the L2 norm of gradient at each layer performs an unbiased random walk as one moves from the last layer to the first. Looks linear initialization was designed to allow the neural network to behave like a deep linear network at initialization, since W R e L U ( x ) − W R e L U ( − x ) = W x {\displaystyle W\;\mathrm {ReLU} (x)-W\;\mathrm {ReLU} (-x)=Wx} . It initializes a matrix W {\displaystyle W} of shape R n 2 × m {\displaystyle \mathbb {R} ^{{\frac {n}{2}}\times m}} by any method, such as orthogonal initialization, t
Read more →
Cognition Network Technology

Cognition Network Technology (CNT), also known as Definiens Cognition Network Technology, is an object-based image analysis method developed by Nobel laureate Gerd Binnig together with a team of researchers at Definiens AG in Munich, Germany. It serves for extracting information from images using a hierarchy of image objects (groups of pixels), as opposed to traditional pixel processing methods. To emulate the human mind's cognitive powers, Definiens used patented image segmentation and classification processes, and developed a method to render knowledge in a semantic network. CNT examines pixels not in isolation, but in context. It builds up a picture iteratively, recognizing groups of pixels as objects. It uses the color, shape, texture and size of objects as well as their context and relationships to draw conclusions and inferences, similar to human analysis. == History == In 1994 Professor Gerd Binnig founded Definiens. CNT was first available with the launch of the eCognition software in May 2000. In June 2010, Trimble Navigation Ltd (NASDAQ: TRMB) acquired Definiens business asset in earth sciences markets, including eCognition software, and also licensed Definiens' patented CNT. In 2014, Definiens was acquired by MedImmune, the global biologics research and development arm of AstraZeneca, for an initial consideration of $150 million. == Software == Definiens Tissue Studio Definiens Tissue Studio is a digital pathology image analysis software application based on CNT. The intended use of Definiens Tissue Studio is for biomarker translational research in formalin-fixed, paraffin-embedded tissue samples which have been treated with immunohistochemical staining assays, or hematoxylin and eosin (H&E). The central concept behind Definiens Tissue Studio is a user interface that facilitates machine learning from example digital histopathology images to derive an image analysis solution suitable for the measurement of biomarkers and/or histological features within pre-defined regions of interest on a cell-by-cell basis, and within sub-cellular compartments. The derived image analysis solution is then automatically applied to subsequent digital images to objectively measure defined sets of multiparametric image features. These data sets are used for further understanding the underlying biological processes that drive cancer and other diseases. Image processing and data analysis are performed either on a local desktop computer workstation, or on a server grid. eCognition The eCognition suite offers three components that can be used stand-alone or in combination to solve image analysis tasks. eCognition Developer is a development environment for object-based image analysis. It is used in earth sciences to develop rule sets (or applications) for the analysis of remote sensing data. eCognition Architect enables non-technical users to configure, calibrate and execute image analysis workflows created in eCognition Developer. eCognition Server software provides a processing environment for batch execution of image analysis jobs. eCognition software is utilized in numerous remote sensing and geospatial application scenarios and environments, using a variety of data types: Generic: Rapid Mapping, Change Detection, Object Recognition By environment: Diverse Landcover Mapping, Urban Analysis (i.e. impervious surface area analysis for taxation, property assessment for insurance, inventory of green infrastructure), Forestry (i.e. biomass measurement, species identification, firescar measurement), Agriculture (i.e. regional planning, precision farming, crisis response), Marine and Riparian (i.e. ecosystem evaluation, disaster management, harbor monitoring). Other: Defense, security, atmosphere and climate The online eCognition community was launched in July 2009 and had 2813 members as of July 9, 2010. Membership is distributed globally and user conferences are held regularly, the last having taken place in November 2009 in Munich, Germany. The bi-annual GEOBIA (Geographic Object-Based Image Analysis) conference is heavily attended by eCognition users, with the majority of presentations based on eCognition software.
Read more →
Zhura

Zhura ( ZUR-ə) is a free, web-based screenwriting software application for writing and formatting screenplays to the film industry standard, as well as other formats. Zhura allows users to collaborate on scripts in public or private groups and uses Creative Commons Licensing for all work in the public workspace. On March 29, 2010, Zhura announced its merger with Scripped. Scripped's CEO, Sunil Rajaraman, remains the company's Chief Executive Officer (CEO) as of 2022. The Zhura CEO was Eric MacDonald, a former Cascade Communications engineer. Scripped later closed on April 1, 2015 after a catastrophic, irrecoverable data loss. == Script editor == Screenplay Template – The script editor provides a built-in screenplay template which formats the document to a standard for scripts as recommended by the AMPAS. The screenplay document is composed of seven elements: scene, action, character, dialogue, parenthetical, transition, and shot (see image). Each element has a specific style to which the script editor conforms as you type.Script Formats – Other major script formats for stage play, sitcom, audio drama and comic book are also supported as well as the ability to switch between them.Auto-Complete – Characters, scene headings and custom transitions are “remembered” as they are written and “recalled” with tab-completion when a writer starts a new character, scene heading or transition, respectively.Multiple Editors – With a collaborative editing model comparable to Google Docs, two or more users can edit the same script simultaneously, regardless of having a different operating system or web browser. Import/Export – A screenplay written in another program can be imported into the script editor and automatically conformed to the screenplay template. The closer the original script has adhered to the standard format, the better it will appear when imported. Supported import/export formats include Text (.txt) Word (.doc) Rich Text (.rtf) and OpenDocument (.odt). Scripts can also be exported as a PDF file with additional options.Tracking Changes – Similar to the “tracking” feature in Microsoft Word, a user can review all changes made to a script in the revision history as well as highlight the contributions of each writer. Offline Mode – The Google Gears-based offline functionality is in the process of being updated and is not available for new subscribers, according to the company founders. == Community == Scripped supports typical social networking features such as discussion boards, comments, user profiles, public and private writing groups, internal web mail and instant messaging within the script editor. There is also the option to share scripts with others outside of Scripped by making scripts externally viewable. Scripped is made up entirely of user-generated scripts that other users can share, critique and edit, offering creative support to a community of writers. == Licensing of user-created work == There are three types of work-spaces on Scripped (personal, group and public) with unique copyright and licensing management for the work created in each area. Any work a user originates may be moved from the personal area to a public or group area at any time. Once another user edits a script, however, it cannot be moved into the originator’s personal area. Personal Workspace – Any script created or video uploaded in the user’s personal workspace remains copyrighted to that user. Until the user moves that script or video from their personal area into a group or public area, no other user shares a copyright or license to that work. Private Group Workspace – The copyright to any script created or video uploaded in a private group workspace is allocated by the individual members of the group, however they see fit. Public Workspace – Any script created or video uploaded in the public workspace is assigned a Creative Commons license by the originator of that work. The originator of a script may select one of four Creative Commons licenses before introducing that script to the public. The selection of the license is determined by what the author wants to allow others to do with the work. Below is a list of Creative Commons licenses available for all scripts and videos in the public workspace. Share Alike (BY-SA) This license lets others remix, tweak, and build upon your work even for commercial reasons, as long as they credit the original user and license their new creations under the identical terms. This license is often compared to open source software licenses. All new works based on the original user's will carry the same license, so any derivatives will also allow commercial use. No Derivatives (BY-ND) This license allows for redistribution, commercial and non-commercial, as long as it is passed along unchanged and in whole, with credit to the original user. Non-Commercial, No Derivatives (BY-NC-ND) This license is the most restrictive of the four licenses, allowing redistribution. This license is often called the "free advertising" license because it allows others to download the original user work and share them with others as long as they mention the original user and link back to them, but they can't change them in any way or use them commercially. Non-Commercial, Share Alike (BY-NC-SA) This license lets others remix, tweak, and build upon the original user's work non-commercially, as long as they credit the original user and license their new creations under the identical terms. Others can download and redistribute the original user's work just like the BY-NC-ND license, but they can also translate, make remixes, and produce new stories based on the original user's work. All new work based on the original user's work will carry the same license, so any derivatives will also be non-commercial in nature. == Events == In April 2008, Zhura partnered with Improv Asylum, a comedy troupe in Boston, Massachusetts to produce a live sketch comedy show called "You Wrote It, Live" entirely written by the public on Zhura. Another show was produced in June.
Read more →
History of natural language processing

The history of natural language processing describes the advances of natural language processing. There is some overlap with the history of machine translation, the history of speech recognition, and the history of artificial intelligence. == Early history == The history of machine translation dates back to the seventeenth century, when philosophers such as Leibniz and Descartes put forward proposals for codes which would relate words between languages. All of these proposals remained theoretical, and none resulted in the development of an actual machine. The first patents for "translating machines" were applied for in the mid-1930s. One proposal, by Georges Artsrouni, was simply an automatic bilingual dictionary using paper tape. The other proposal, by Peter Troyanskii, a Russian, was more detailed. Troyanskii’s proposal included both the bilingual dictionary and a method for dealing with grammatical roles between languages, based on Esperanto. == Logical period == In 1950, Alan Turing published his famous article "Computing Machinery and Intelligence" which proposed what is now called the Turing test as a criterion of intelligence. This criterion depends on the ability of a computer program to impersonate a human in a real-time written conversation with a human judge, sufficiently well that the judge is unable to distinguish reliably — on the basis of the conversational content alone — between the program and a real human. In 1957, Noam Chomsky’s Syntactic Structures revolutionized Linguistics with 'universal grammar', a rule-based system of syntactic structures. The Georgetown experiment in 1954 involved fully automatic translation of more than sixty Russian sentences into English. The authors claimed that within three or five years, machine translation would be a solved problem. However, real progress was much slower, and after the ALPAC report in 1966, which found that ten years long research had failed to fulfill the expectations, funding for machine translation was dramatically reduced. Little further research in machine translation was conducted until the late 1980s, when the first statistical machine translation systems were developed. Some notably successful NLP systems developed in the 1960s were SHRDLU, a natural language system working in restricted "blocks worlds" with restricted vocabularies. In 1969 Roger Schank introduced the conceptual dependency theory for natural language understanding. This model, partially influenced by the work of Sydney Lamb, was extensively used by Schank's students at Yale University, such as Robert Wilensky, Wendy Lehnert, and Janet Kolodner. In 1970, William A. Woods introduced the augmented transition network (ATN) to represent natural language input. Instead of phrase structure rules ATNs used an equivalent set of finite-state automata that were called recursively. ATNs and their more general format called "generalized ATNs" continued to be used for a number of years. During the 1970s many programmers began to write 'conceptual ontologies', which structured real-world information into computer-understandable data. Examples are MARGIE (Schank, 1975), SAM (Cullingford, 1978), PAM (Wilensky, 1978), TaleSpin (Meehan, 1976), QUALM (Lehnert, 1977), Politics (Carbonell, 1979), and Plot Units (Lehnert 1981). During this time, many chatterbots were written including PARRY, Racter, and Jabberwacky. == Statistical period == Up to the 1980s, most NLP systems were based on complex sets of hand-written rules. Starting in the late 1980s, however, there was a revolution in NLP with the introduction of machine learning algorithms for language processing. This was due both to the steady increase in computational power resulting from Moore's law and the gradual lessening of the dominance of Chomskyan theories of linguistics (e.g. transformational grammar), whose theoretical underpinnings discouraged the sort of corpus linguistics that underlies the machine-learning approach to language processing. Some of the earliest-used machine learning algorithms, such as decision trees, produced systems of hard if-then rules similar to existing hand-written rules. Increasingly, however, research has focused on statistical models, which make soft, probabilistic decisions based on attaching real-valued weights to the features making up the input data. The cache language models upon which many speech recognition systems now rely are examples of such statistical models. Such models are generally more robust when given unfamiliar input, especially input that contains errors (as is very common for real-world data), and produce more reliable results when integrated into a larger system comprising multiple subtasks. === Datasets === The emergence of statistical approaches was aided by both increase in computing power and the availability of large datasets. At that time, large multilingual corpora were starting to emerge. Notably, some were produced by the Parliament of Canada and the European Union as a result of laws calling for the translation of all governmental proceedings into all official languages of the corresponding systems of government. Many of the notable early successes occurred in the field of machine translation. In 1993, the IBM alignment models were used for statistical machine translation. Compared to previous machine translation systems, which were symbolic systems manually coded by computational linguists, these systems were statistical, which allowed them to automatically learn from large textual corpora. Though these systems do not work well in situations where only small corpora is available, so data-efficient methods continue to be an area of research and development. In 2001, a one-billion-word large text corpus, scraped from the Internet, referred to as "very very large" at the time, was used for word disambiguation. To take advantage of large, unlabelled datasets, algorithms were developed for unsupervised and self-supervised learning. Generally, this task is much more difficult than supervised learning, and typically produces less accurate results for a given amount of input data. However, there is an enormous amount of non-annotated data available (including, among other things, the entire content of the World Wide Web), which can often make up for the inferior results. == Neural period == Neural language models were developed in 1990s. In 1990, the Elman network, using a recurrent neural network, encoded each word in a training set as a vector, called a word embedding, and the whole vocabulary as a vector database, allowing it to perform such tasks as sequence-predictions that are beyond the power of a simple multilayer perceptron. A shortcoming of the static embeddings was that they didn't differentiate between multiple meanings of homonyms. Yoshua Bengio developed the first neural probabilistic language model in 2000. Novel algorithms, availability of larger datasets and higher processing power made possible training of larger and larger language models. Attention mechanism was introduced by Bahdanau et al. in 2014. This work laid the foundations for the famous "Attention Is All You Need" paper that introduced the Transformer architecture in 2017. The concept of large language model (LLM) emerged in late 2010s. LLM is a language model trained with self-supervised learning on vast amount of text. Earliest public LLMs had hundreds of millions of parameters, but this number quickly rose to billion and even trillions. In recent years, advancements in deep learning and large language models have significantly enhanced the capabilities of natural language processing, leading to widespread applications in areas such as healthcare, customer service, and content generation. == Software ==
Read more →
U-Net

U-Net is a convolutional neural network that was developed for image segmentation. The network is based on a fully convolutional neural network whose architecture was modified and extended to work with fewer training images and to yield more precise segmentation. Segmentation of a 512 × 512 image takes less than a second on a modern (2015) GPU using the U-Net architecture. The U-Net architecture has also been employed in diffusion models for iterative image denoising. This technology underlies many modern image generation models, such as DALL-E, Midjourney, and Stable Diffusion. U-Net is also being explored for language models. Tokenization is not a separate step, allowing the model to more easily understand spelling and concurrently vectorizing / tokenizing higher level concepts. == Description == The U-Net architecture stems from the so-called "fully convolutional network". The main idea is to supplement a usual contracting network by successive layers, where pooling operations are replaced by upsampling operators. Hence these layers increase the resolution of the output. A successive convolutional layer can then learn to assemble a precise output based on this information. One important modification in U-Net is that there are a large number of feature channels in the upsampling part, which allow the network to propagate context information to higher resolution layers. As a consequence, the expansive path is more or less symmetric to the contracting part, and yields a u-shaped architecture. The network only uses the valid part of each convolution without any fully connected layers. To predict the pixels in the border region of the image, the missing context is extrapolated by mirroring the input image. This tiling strategy is important to apply the network to large images, since otherwise the resolution would be limited by the GPU memory. Recently, there had also been an interest in receptive field based U-Net models for medical image segmentation. == Network architecture == The network consists of a contracting path and an expansive path, which gives it the u-shaped architecture. The contracting path is a typical convolutional network that consists of repeated application of convolutions, each followed by a rectified linear unit (ReLU) and a max pooling operation. During the contraction, the spatial information is reduced while feature information is increased. The expansive pathway combines the feature and spatial information through a sequence of up-convolutions and concatenations with high-resolution features from the contracting path. == Applications == There are many applications of U-Net in biomedical image segmentation, such as brain image segmentation (''BRATS'') and liver image segmentation ("siliver07") as well as protein binding site prediction. U-Net implementations have also found use in the physical sciences, for example in the analysis of micrographs of materials. Variations of the U-Net have also been applied for medical image reconstruction. Here are some variants and applications of U-Net as follows: Pixel-wise regression using U-Net and its application on pansharpening; 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation; TernausNet: U-Net with VGG11 Encoder Pre-Trained on ImageNet for Image Segmentation. Image-to-image translation to estimate fluorescent stains In binding site prediction of protein structure. == History == U-Net was created by Olaf Ronneberger, Philipp Fischer, Thomas Brox in 2015 and reported in the paper "U-Net: Convolutional Networks for Biomedical Image Segmentation". It is an improvement and development of FCN: Evan Shelhamer, Jonathan Long, Trevor Darrell (2014). "Fully convolutional networks for semantic segmentation".
Read more →
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
Read more →
Deep Learning Super Sampling

Deep Learning Super Sampling (DLSS) is a suite of real-time deep learning image enhancement and upscaling technologies developed by Nvidia that are available in a number of video games. The goal of these technologies is to allow the majority of the graphics pipeline to run at a lower resolution for increased performance, and then infer a higher resolution image from this that approximates the same level of detail as if the image had been rendered at this higher resolution. This allows for higher graphical settings or frame rates for a given output resolution, depending on user preference. All generations of DLSS are available on all RTX-branded cards from Nvidia in supported titles. However, the Frame Generation feature is only supported on RTX 40 series GPUs or newer and Multi Frame Generation is only available on 50 series GPUs. == History == Nvidia advertised DLSS as a key feature of GeForce RTX 20 series GPUs when they launched in September 2018. At that time, the results were limited to a few video games, namely Battlefield V, or Metro Exodus, because the algorithm had to be trained specifically on each game on which it was applied and the results were usually not as good as simple resolution upscaling. In 2019, Control shipped with ray tracing and an image processing algorithm that approximated DLSS, which did not use the Tensor Cores. In April 2020, Nvidia advertised and shipped an improved version of DLSS named DLSS 2 with driver version 445.75. DLSS 2.0 was available for a few existing games including Control and Wolfenstein: Youngblood, and would later be added to many newly released games and game engines such as Unreal Engine and Unity. This time Nvidia said that it used the Tensor Cores again, and that the AI did not need to be trained specifically on each game. Despite sharing the DLSS branding, the two iterations of DLSS differ significantly and are not backwards-compatible. In January 2025, Nvidia stated that there are over 540 games and apps supporting DLSS, and that over 80% of Nvidia RTX users activate DLSS. In March 2025, there were more than 100 games that support DLSS 4, according to Nvidia. By May 2025, over 125 games supported DLSS 4. The first video game console to use DLSS, the Nintendo Switch 2, was released on June 5, 2025. Nvidia announced DLSS 4.5 at CES 2026. In January 2026, Nvidia stated that over 250 games and applications support Multi Frame Generation. On March 16, 2026, at GTC 2026, Nvidia CEO Jensen Huang presented DLSS 5, a real-time AI model based on neural rendering that realistically enhances lighting and material surfaces at up to 4K resolution while retaining the developer's intended art style. It is planned to release in fall of 2026. In a blog post on its website, Nvidia has announced that DLSS 5 will be available in such games as Assassin's Creed Shadows, Delta Force, Hogwarts Legacy, Naraka: Bladepoint, Phantom Blade Zero, Resident Evil Requiem, Starfield, The Elder Scrolls IV: Oblivion Remastered, and more. On May 31, 2026, Nvidia announced an updated version of Ray Reconstruction for DLSS 4.5 in a blog post, scheduled for release on all RTX GPUs in August of the same year. They said it is designed to better embed spatial awareness into scenes and analyze engine data on movements and lighting conditions, resulting in a sharper, more stable, and less noisy image. === Release timeline === == Technology == === DLSS 1 === The first iteration of DLSS is a predominantly spatial image upscaler with two stages, both relying on convolutional auto-encoder neural networks. The first step is an image enhancement network which uses the current frame and motion vectors to perform edge enhancement, and spatial anti-aliasing. The second stage is an image upscaling step which uses the single raw, low-resolution frame to upscale the image to the desired output resolution. Using just a single frame for upscaling means the neural network itself must generate a large amount of new information to produce the high-resolution output, which can result in slight hallucinations such as leaves that differ in style to the source content. The neural networks are trained on a per-game basis by generating a "perfect frame" using traditional supersampling to 64 samples per pixel, as well as the motion vectors for each frame. The data collected must be as comprehensive as possible, including as many levels, times of day, graphical settings, resolutions, etc. as possible. This data is also augmented using common augmentations such as rotations, colour changes, and random noise to help generalize the test data. Training is performed on Nvidia's Saturn V supercomputer. This first iteration received a mixed response, with many criticizing the often soft appearance and artifacts along with glitches in certain situations; likely a side effect of the limited data from only using a single frame input to the neural networks which could not be trained to perform optimally in all scenarios and edge-cases. Nvidia also demonstrated the ability for the auto-encoder networks to learn the ability to recreate depth-of-field and motion blur, although this functionality has never been included in a publicly released product. === DLSS 2 === DLSS 2 is a temporal anti-aliasing upsampling (TAAU) implementation, using data from previous frames extensively through sub-pixel jittering to resolve fine detail and reduce aliasing. The data DLSS 2 collects includes: the raw low-resolution input, motion vectors, depth buffers, and exposure / brightness information. It can also be used as a simpler TAA implementation where the image is rendered at 100% resolution, rather than being upsampled by DLSS, Nvidia brands this as DLAA (Deep Learning Anti-Aliasing). TAA(U) is used in many modern video games and game engines; however, all previous implementations have used some form of manually written heuristics to prevent temporal artifacts such as ghosting and flickering. One example of this is neighborhood clamping which forcefully prevents samples collected in previous frames from deviating too much compared to nearby pixels in newer frames. This helps to identify and fix many temporal artifacts, but deliberately removing fine details in this way is analogous to applying a blur filter, and thus the final image can appear blurry when using this method. DLSS 2 uses a convolutional auto-encoder neural network trained to identify and fix temporal artifacts, instead of manually programmed heuristics as mentioned above. Because of this, DLSS 2 can generally resolve detail better than other TAA and TAAU implementations, while also removing most temporal artifacts. This is why DLSS 2 can sometimes produce a sharper image than rendering at higher, or even native resolutions using traditional TAA. However, no temporal solution is perfect, and artifacts (ghosting in particular) are still visible in some scenarios when using DLSS 2. Because temporal artifacts occur in most art styles and environments in broadly the same way, the neural network that powers DLSS 2 does not need to be retrained when being used in different games. Despite this, Nvidia does frequently ship new minor revisions of DLSS 2 with new titles, so this could suggest some minor training optimizations may be performed as games are released, although Nvidia does not provide changelogs for these minor revisions to confirm this. The main advancements compared to DLSS 1 include: Significantly improved detail retention, a generalized neural network that does not need to be re-trained per-game, and ~2x less overhead (~1–2 ms vs ~2–4 ms). It should also be noted that forms of TAAU such as DLSS 2 are not upscalers in the same sense as techniques such as ESRGAN or DLSS 1, which attempt to create new information from a low-resolution source; instead, TAAU works to recover data from previous frames, rather than creating new data. In practice, this means low resolution textures in games will still appear low-resolution when using current TAAU techniques. This is why Nvidia recommends game developers use higher resolution textures than they would normally for a given rendering resolution by applying a mip-map bias when DLSS 2 is enabled. === DLSS 3 === Augments DLSS 2 with improved image quality and the introduction of a new motion interpolation feature, called Frame Generation. The DLSS Frame Generation algorithm takes two rendered frames from the rendering pipeline and generates a new frame that smoothly transitions between them. For every frame rendered, one additional frame is generated. DLSS 3.0 makes use of a new generation Optical Flow Accelerator (OFA) included in the Ada Lovelace architecture of GeForce RTX 40 series GPUs and with that is exclusive to them. The new OFA is said to be faster and more accurate than the one already available in previous Turing and Ampere RTX GPUs. === DLSS 3.5 === DLSS 3.5 adds Ray Reconstruction, replacing multiple denoising algorithms with a single AI model trained o
Read more →
Sentence embedding

In natural language processing, a sentence embedding is a representation of a sentence as a vector of numbers which encodes meaningful semantic information. State of the art embeddings are based on the learned hidden layer representation of dedicated sentence transformer models. BERT pioneered an approach involving the use of a dedicated [CLS] token prepended to the beginning of each sentence inputted into the model; the final hidden state vector of this token encodes information about the sentence and can be fine-tuned for use in sentence classification tasks. In practice however, BERT's sentence embedding with the [CLS] token achieves poor performance, often worse than simply averaging non-contextual word embeddings. SBERT later achieved superior sentence embedding performance by fine tuning BERT's [CLS] token embeddings through the usage of a siamese neural network architecture on the SNLI dataset. Other approaches are loosely based on the idea of distributional semantics applied to sentences. Skip-Thought trains an encoder-decoder structure for the task of neighboring sentences predictions; this has been shown to achieve worse performance than approaches such as InferSent or SBERT. An alternative direction is to aggregate word embeddings, such as those returned by Word2vec, into sentence embeddings. The most straightforward approach is to simply compute the average of word vectors, known as continuous bag-of-words (CBOW). However, more elaborate solutions based on word vector quantization have also been proposed. One such approach is the vector of locally aggregated word embeddings (VLAWE), which demonstrated performance improvements in downstream text classification tasks. == Applications == In recent years, sentence embedding has seen a growing level of interest due to its applications in natural language queryable knowledge bases through the usage of vector indexing for semantic search. LangChain for instance utilizes sentence transformers for purposes of indexing documents. In particular, an indexing is generated by generating embeddings for chunks of documents and storing (document chunk, embedding) tuples. Then given a query in natural language, the embedding for the query can be generated. A top k similarity search algorithm is then used between the query embedding and the document chunk embeddings to retrieve the most relevant document chunks as context information for question answering tasks. This approach is also known formally as retrieval-augmented generation. Though not as predominant as BERTScore, sentence embeddings are commonly used for sentence similarity evaluation which sees common use for the task of optimizing a Large language model's generation parameters is often performed via comparing candidate sentences against reference sentences. By using the cosine-similarity of the sentence embeddings of candidate and reference sentences as the evaluation function, a grid-search algorithm can be utilized to automate hyperparameter optimization. == Evaluation == A way of testing sentence encodings is to apply them on Sentences Involving Compositional Knowledge (SICK) corpus for both entailment (SICK-E) and relatedness (SICK-R). In the best results are obtained using a BiLSTM network trained on the Stanford Natural Language Inference (SNLI) Corpus. The Pearson correlation coefficient for SICK-R is 0.885 and the result for SICK-E is 86.3. A slight improvement over previous scores is presented in: SICK-R: 0.888 and SICK-E: 87.8 using a concatenation of bidirectional Gated recurrent unit.
Read more →
LTX (text-to-video model)

LTX is a family of open source artificial intelligence video foundation models developed by Lightricks, and first released in November 2024. The latest models, LTX-2, create videos based on user prompts. They were preceded by LTX Video, which was released in 2024 as the company's first text-to-video model. LTX-2 is part of the LTX family of video generation models, which form the core technology, alongside LTX Studio, of the LTX ecosystem. == History == === Origins: LTX Video (2024–2025) === In November 2024 Lightricks publicly released its first text-to-video model, LTX Video. It was a 2-billion parameter model, available as open source. In May 2025 Lightricks launched LTXV-13b, a version with 13-billion parameters. Two months later, the model broke the 60 second barrier for generated video. === Release of LTX-2 (2025) === In October 2025 Lightricks announced its latest model, and renamed it LTX-2. The model was described as capable of generating synchronized audio and video at native 4K resolution and up to 50 frames per second (fps), using a variety of conditions and prompts, including text-to-video and image-to-video. Google highlighted the fact that LTX-2 was trained on its infrastructure, and saying it was "The first open source AI video generation model, powered by Google Cloud". Upon its release it was ranked in the top-3 models for image-to-video creation by Artificial Analysis, behind Kling 3.5 by Kling AI and Veo 3.1 by Google. Its text-to-image option was ranked 7th. In addition to its open-source release, Lightricks offers API access to LTX-2, allowing developers to generate videos from text and image prompts through a hosted service without running the model locally. === Open Source Release (2026) === In January 2026, Lightricks officially released the full open-source version of LTX-2, making the model’s complete codebase, weights, and associated tooling publicly available. In March 2026 the company released LTX-2.3, which was accompanied by a desktop video editor enabling the entire model to run locally on consumer hardware. == Technical features == === Advancements over LTX Video === LTX-2 builds upon the LTX Video architecture with several major improvements: Unified audio-video generation producing synchronized dialogue, ambience, and motion Native 4K rendering 50-fps output for cinematic motion Three operational modes (Fast, Pro, Ultra) More efficient diffusion pipelines enabling high fidelity on consumer GPUs === Core capabilities === Text-to-video generation Image-to-video generation Multimodal audiovisual synthesis High-resolution spatial and temporal coherence Configurable quality/performance settings Open-source distribution of weights and datasets == Reception == Initial reception to LTX-2 was broadly positive, with several technology and media outlets highlighting its open-source approach and multimodal capabilities. Open Source For You described LTX-2 as “one of the first AI video systems to combine 4K output, synchronized audio, and an open model release,” noting that it positioned Lightricks as a significant competitor to proprietary systems such as OpenAI's Sora and Google's Veo. IEA Green said that the model “could rewrite the AI filmmaking game,” emphasizing that its 50-fps rendering and unified audio-video generation made it suitable for professional studios and independent creators alike. AI News characterized LTX-2 as a “major step forward in the democratization of cinematic-quality video generation,” praising its consumer-grade hardware efficiency and multi-tier generation modes, while also noting ongoing challenges in long-form temporal stability. FinancialContent reported strong interest among creative agencies, attributing the attention to Lightricks’ decision to release model weights and datasets, which reviewers said enabled “a level of transparency not typically seen in commercial AI video models.” === Benchmarks and rankings === Upon release, LTX-2 ranked third for image-to-video creation in the Artificial Analysis benchmark, behind Kling 3.5 and Veo 3.1, while its text-to-video option ranked seventh. As of early 2026, it was the highest-ranked open-source model in the benchmark. === Limitations === Some early reviewers also pointed out quality limitations. The Ray3 technical review noted occasional inconsistencies in lip-sync and motion tracking during long scenes, though it stated these were “in line with the challenges faced by all current AI video diffusion models” and expected to improve with continued iteration. Like other diffusion-based video generators, LTX-2 can produce artifacts in complex multi-person scenes and may struggle with precise text rendering within generated video.
Read more →
XLNet

The XLNet was an autoregressive Transformer designed as an improvement over BERT, with 340M parameters and trained on 33 billion words. It was released on 19 June 2019, under the Apache 2.0 license. It achieved state-of-the-art results on a variety of natural language processing tasks, including language modeling, question answering, and natural language inference. == Architecture == The main idea of XLNet is to model language autoregressively like the GPT models, but allow for all possible permutations of a sentence. Concretely, consider the following sentence:My dog is cute.In standard autoregressive language modeling, the model would be tasked with predicting the probability of each word, conditioned on the previous words as its context: We factorize the joint probability of a sequence of words x 1 , … , x T {\displaystyle x_{1},\ldots ,x_{T}} using the chain rule: Pr ( x 1 , … , x T ) = Pr ( x 1 ) Pr ( x 2 | x 1 ) Pr ( x 3 | x 1 , x 2 ) … Pr ( x T | x 1 , … , x T − 1 ) . {\displaystyle \Pr(x_{1},\ldots ,x_{T})=\Pr(x_{1})\Pr(x_{2}|x_{1})\Pr(x_{3}|x_{1},x_{2})\ldots \Pr(x_{T}|x_{1},\ldots ,x_{T-1}).} For example, the sentence "My dog is cute" is factorized as: Pr ( My , dog , is , cute ) = Pr ( My ) Pr ( dog | My ) Pr ( is | My , dog ) Pr ( cute | My , dog , is ) . {\displaystyle \Pr({\text{My}},{\text{dog}},{\text{is}},{\text{cute}})=\Pr({\text{My}})\Pr({\text{dog}}|{\text{My}})\Pr({\text{is}}|{\text{My}},{\text{dog}})\Pr({\text{cute}}|{\text{My}},{\text{dog}},{\text{is}}).} Schematically, we can write it as → My → My dog → My dog is → My dog is cute . {\displaystyle {\texttt {}}{\texttt {}}{\texttt {}}{\texttt {}}\to {\text{My }}{\texttt {}}{\texttt {}}{\texttt {}}\to {\text{My dog }}{\texttt {}}{\texttt {}}\to {\text{My dog is }}{\texttt {}}\to {\text{My dog is cute}}.} However, for XLNet, the model is required to predict the words in a randomly generated order. Suppose we have sampled a randomly generated order 3241, then schematically, the model is required to perform the following prediction task: → is → dog is → dog is cute → My dog is cute {\displaystyle {\texttt {}}{\texttt {}}{\texttt {}}{\texttt {}}\to {\texttt {}}{\texttt {}}{\text{is }}{\texttt {}}\to {\texttt {}}{\text{dog is }}{\texttt {}}\to {\texttt {}}{\text{dog is cute}}\to {\text{My dog is cute}}} By considering all permutations, XLNet is able to capture longer-range dependencies and better model the bidirectional context of words. === Two-Stream Self-Attention === To implement permutation language modeling, XLNet uses a two-stream self-attention mechanism. The two streams are: Content stream: This stream encodes the content of each word, as in standard causally masked self-attention. Query stream: This stream encodes the content of each word in the context of what has gone before. In more detail, it is a masked cross-attention mechanism, where the queries are from the query stream, and the key-value pairs are from the content stream. The content stream uses the causal mask M causal = [ 0 − ∞ − ∞ … − ∞ 0 0 − ∞ … − ∞ 0 0 0 … − ∞ ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 … 0 ] {\displaystyle M_{\text{causal}}={\begin{bmatrix}0&-\infty &-\infty &\dots &-\infty \\0&0&-\infty &\dots &-\infty \\0&0&0&\dots &-\infty \\\vdots &\vdots &\vdots &\ddots &\vdots \\0&0&0&\dots &0\end{bmatrix}}} permuted by a random permutation matrix to P M causal P − 1 {\displaystyle PM_{\text{causal}}P^{-1}} . The query stream uses the cross-attention mask P ( M causal − ∞ I ) P − 1 {\displaystyle P(M_{\text{causal}}-\infty I)P^{-1}} , where the diagonal is subtracted away specifically to avoid the model "cheating" by looking at the content stream for what the current masked token is. Like the causal masking for GPT models, this two-stream masked architecture allows the model to train on all tokens in one forward pass. == Training == Two models were released: XLNet-Large, cased: 110M parameters, 24-layer, 1024-hidden, 16-heads XLNet-Base, cased: 340M parameters, 12-layer, 768-hidden, 12-heads. It was trained on a dataset that amounted to 32.89 billion tokens after tokenization with SentencePiece. The dataset was composed of BooksCorpus, and English Wikipedia, Giga5, ClueWeb 2012-B, and Common Crawl. It was trained on 512 TPU v3 chips, for 5.5 days. At the end of training, it still under-fitted the data, meaning it could have achieved lower loss with more training. It took 0.5 million steps with an Adam optimizer, linear learning rate decay, and a batch size of 8192.
Read more →
Truth discovery

Truth discovery (also known as truth finding) is the process of choosing the actual true value for a data item when different data sources provide conflicting information on it. Several algorithms have been proposed to tackle this problem, ranging from simple methods like majority voting to more complex ones able to estimate the trustworthiness of data sources. Truth discovery problems can be divided into two sub-classes: single-truth and multi-truth. In the first case only one true value is allowed for a data item (e.g birthday of a person, capital city of a country). While in the second case multiple true values are allowed (e.g. cast of a movie, authors of a book). Typically, truth discovery is the last step of a data integration pipeline, when the schemas of different data sources have been unified and the records referring to the same data item have been detected. == General principles == The abundance of data available on the web makes more and more probable to find that different sources provide (partially or completely) different values for the same data item. This, together with the fact that we are increasing our reliance on data to derive important decisions, motivates the need of developing good truth discovery algorithms. Many currently available methods rely on a voting strategy to define the true value of a data item. Nevertheless, recent studies, have shown that, if we rely only on majority voting, we could get wrong results even in 30% of the data items. The solution to this problem is to assess the trustworthiness of the sources and give more importance to votes coming from trusted sources. Ideally, supervised learning techniques could be exploited to assign a reliability score to sources after hand-crafted labeling of the provided values; unfortunately, this is not feasible since the number of needed labeled examples should be proportional to the number of sources, and in many applications the number of sources can be prohibitive. == Single-truth vs multi-truth discovery == Single-truth and multi-truth discovery are two very different problems. Single-truth discovery is characterized by the following properties: only one true value is allowed for each data item; different values provided for a given data item oppose to each other; values and sources can either be correct or erroneous. While in the multi-truth case the following properties hold: the truth is composed by a set of values; different values could provide a partial truth; claiming one value for a given data item does not imply opposing to all the other values; the number of true values for each data item is not known a priori. Multi-truth discovery has unique features that make the problem more complex and should be taken into consideration when developing truth-discovery solutions. The examples below point out the main differences of the two methods. Knowing that in both examples the truth is provided by source 1, in the single truth case (first table) we can say that sources 2 and 3 oppose to the truth and as a result provide wrong values. On the other hand, in the second case (second table), sources 2 and 3 are neither correct nor erroneous, they instead provide a subset of the true values and at the same time they do not oppose the truth. == Source trustworthiness == The vast majority of truth discovery methods are based on a voting approach: each source votes for a value of a certain data item and, at the end, the value with the highest vote is select as the true one. In the more sophisticated methods, votes do not have the same weight for all the data sources, more importance is indeed given to votes coming from trusted sources. Source trustworthiness usually is not known a priori but estimated with an iterative approach. At each step of the truth discovery algorithm the trustworthiness score of each data source is refined, improving the assessment of the true values that in turn leads to a better estimation of the trustworthiness of the sources. This process usually ends when all the values reach a convergence state. Source trustworthiness can be based on different metrics, such as accuracy of provided values, copying values from other sources and domain coverage. Detecting copying behaviors is very important, in fact, copy allows to spread false values easily making truth discovery very hard, since many sources would vote for the wrong values. Usually systems decrease the weight of votes associated to copied values or even don’t count them at all. == Single-truth methods == Most of the currently available truth discovery methods have been designed to work well only in the single-truth case. Below are reported some of the characteristics of the most relevant typologies of single-truth methods and how different systems model source trustworthiness. === Majority voting === Majority voting is the simplest method, the most popular value is selected as the true one. Majority voting is commonly used as a baseline when assessing the performances of more complex methods. === Web-link based === These methods estimate source trustworthiness exploiting a similar technique to the one used to measure authority of web pages based on web links. The vote assigned to a value is computed as the sum of the trustworthiness of the sources that provide that particular value, while the trustworthiness of a source is computed as the sum of the votes assigned to the values that the source provides. === Information-retrieval based === These methods estimate source trustworthiness using similarity measures typically used in information retrieval. Source trustworthiness is computed as the cosine similarity (or other similarity measures) between the set of values provided by the source and the set of values considered true (either selected in a probabilistic way or obtained from a ground truth). === Bayesian based === These methods use Bayesian inference to define the probability of a value being true conditioned on the values provided by all the sources. P ( v ∣ ψ ( o ) ) = P ( ψ ( o ) ∣ v ) ⋅ P ( v ) P ( ψ ( o ) ) {\displaystyle P(v\mid \psi (o))={\frac {P(\psi (o)\mid v)\cdot P(v)}{P(\psi (o))}}} where v {\displaystyle \textstyle v} is a value provided for a data item o {\displaystyle \textstyle o} and ψ ( o ) {\displaystyle \textstyle \psi (o)} is the set of the observed values provided by all the sources for that specific data item. The trustworthiness of a source is then computed based on the accuracy of the values that provides. Other more complex methods exploit Bayesian inference to detect copying behaviors and use these insights to better assess source trustworthiness. == Multi-truth methods == Due to its complexity, less attention has been devoted to the study of the multi-truth discovery Below are reported two typologies of multi-truth methods and their characteristics. === Bayesian based === These methods use Bayesian inference to define the probability of a group of values being true conditioned on the values provided by all the data sources. In this case, since there could be multiple true values for each data item, and sources can provide multiple values for a single data item, it is not possible to consider values individually. An alternative is to consider mappings and relations between set of provided values and sources providing them. The trustworthiness of a source is then computed based on the accuracy of the values that provides. More sophisticated methods also consider domain coverage and copying behaviors to better estimate source trustworthiness. === Probabilistic Graphical Models based === These methods use probabilistic graphical models to automatically define the set of true values of given data item and also to assess source quality without need of any supervision. == Applications == Many real-world applications can benefit from the use of truth discovery algorithms. Typical domains of application include: healthcare, crowd/social sensing, crowdsourcing aggregation, information extraction and knowledge base construction. Truth discovery algorithms could be also used to revolutionize the way in which web pages are ranked in search engines, going from current methods based on link analysis like PageRank, to procedures that rank web pages based on the accuracy of the information they provide.
Read more →
Stereo cameras

The stereo cameras approach is a method of distilling a noisy video signal into a coherent data set that a computer can begin to process into actionable symbolic objects, or abstractions. Stereo cameras is one of many approaches used in the broader fields of computer vision and machine vision. == Calculation == In this approach, two cameras with a known physical relationship (i.e. a common field of view the cameras can see, and how far apart their focal points sit in physical space) are correlated via software. By finding mappings of common pixel values, and calculating how far apart these common areas reside in pixel space, a rough depth map can be created. This is very similar to how the human brain uses stereoscopic information from the eyes to gain depth cue information, i.e. how far apart any given object in the scene is from the viewer. The camera attributes must be known, focal length and distance apart etc., and a calibration done. Once this is completed, the systems can be used to sense the distances of objects by triangulation. Finding the same singular physical point in the two left and right images is known as the correspondence problem. Correctly locating the point gives the computer the capability to calculate the distance that the robot or camera is from the object. On the BH2 Lunar Rover the cameras use five steps: a bayer array filter, photometric consistency dense matching algorithm, a Laplace of Gaussian (LoG) edge detection algorithm, a stereo matching algorithm and finally uniqueness constraint. == Uses == This type of stereoscopic image processing technique is used in applications such as 3D reconstruction, robotic control and sensing, crowd dynamics monitoring and off-planet terrestrial rovers; for example, in mobile robot navigation, tracking, gesture recognition, targeting, 3D surface visualization, immersive and interactive gaming. Although the Xbox Kinect sensor is also able to create a depth map of an image, it uses an infrared camera for this purpose, and does not use the dual-camera technique. Other approaches to stereoscopic sensing include time of flight sensors and ultrasound.
Read more →
Artificial Linguistic Internet Computer Entity

A.L.I.C.E. (Artificial Linguistic Internet Computer Entity), also referred to as Alicebot, or simply Alice, is a natural language processing chatbot—a program that engages in a conversation with a human by applying some heuristical pattern matching rules to the human's input. It was inspired by Joseph Weizenbaum's classical ELIZA program. It is one of the strongest programs of its type and has won the Loebner Prize, awarded to accomplished humanoid, talking robots, three times (in 2000, 2001, and 2004). The program is unable to pass the Turing test, as even the casual user will often expose its mechanistic aspects in short conversations. Alice was originally composed by Richard Wallace; it "came to life" on November 23, 1995. The program was rewritten in Java beginning in 1998. The current incarnation of the Java implementation is Program D. The program uses an XML Schema called AIML (Artificial Intelligence Markup Language) for specifying the heuristic conversation rules. Alice code has been reported to be available as open source. The AIML source is available from ALICE A.I. Foundation on Google Code and from the GitHub account of Richard Wallace. These AIML files can be run using an AIML interpreter like Program O or Program AB. == In popular culture == Spike Jonze has cited ALICE as the inspiration for his academy award-winning film Her, in which a human falls in love with a chatbot. In a New Yorker article titled “Can Humans Fall in Love with Bots?” Jonze said “that the idea originated from a program he tried about a decade ago called the ALICE bot, which engages in friendly conversation.” The Los Angeles Times reported:Though the film’s premise evokes comparisons to Siri, Jonze said he actually had the idea well before the Apple digital assistant came along, after using a program called Alicebot about ten years ago. As geek nostalgists will recall, that intriguing if at times crude software (it flunked the industry-standard Turing Test) would attempt to engage users in everyday chatter based on a database of prior conversations. Jonze liked it, and decided to apply a film genre to it. “I thought about that idea, and what if you had a real relationship with it?” Jonze told reporters. “And I used that as a way to write a relationship movie and a love story.”
Read more →