AI Assistant Roblox

AI Assistant Roblox — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  Depop

  Depop Limited is a social e-commerce company based in London, with additional offices in Milan and New York City. The company allows users to buy and sell items, which are mostly used and vintage pieces of clothing. == History == Depop was founded in 2011 by entrepreneur Simon Beckerman at an Italian technological incubator and business start-up centre, H-Farm. Beckerman came up with the original outline of the application during his time working on PIG, a fashion magazine based in Italy that he co-founded. The idea was to create a platform where products shown in the magazine could be purchased by users online. This idea turned into a concept similar to a flea market but on the internet, where people could sell their items while also being in control of advertising, public relations, and the creative process behind their accounts. While being financially supported by H-Farm, Beckerman worked within a team to create and lay out the Depop application while exposing it to numerous investors. In 2013, Beckerman became a member of the company's board to help improve the application and business while concurrently ceding his role of CEO. Maria Raga, Depop's co-founder and former CEO, took on the role of vice president of operations in 2014, and in 2016, she became chief executive. According to Raga, the main goal while developing Depop was to become the next Airbnb or Spotify, but to make an impact on fashion. Paolo Barberis and Nana Bianca were two of the first investors in the platform in 2012 with a seed investment. Its headquarters were moved to London in 2012. Depop expanded and opened additional offices in Milan and New York City. Beckerman raised €1 million in funding in October 2013 from Red Circle Investment and brought on Faroese Runar Reistrup as new CEO. In 2015, Depop secured another investment of $8 million from Balderton Capital and HV Capital. In March 2016, former CEO, Runar Reistrup, stated that Depop's growth was achieved through word of mouth. During his time as CEO, this growth involved taking Depop as a startup and working to raise funds to eventually amass a significant user base within the United States. In June 2019, Depop raised $62 million in Series C from General Atlantic to fund its expansion. Previous investors HV Capital, Balderton Capital, Creandum, Octopus Ventures, TempoCap and Sebastian Siemiatkowski also participated. During this time, Depop held workshops and conversations as part of their Depop Live NY events, and the company also opened a London store through their partnership with Selfridges. In 2020, Depop's gross merchandise sales and revenue both more than doubled to $650 million and $70 million respectively. This may be attributed to Depop's responsiveness to user trends, its lack of issues regarding inventory management, and the increase in users looking to resell. As of 2024, Depop has over 35 million users, according to their website. Depop is popular for Gen Z and young millennials, it is the 10th most-visited shopping platform for Gen Z consumers in the US, and, in a poll conducted by The Strategist in 2019, Depop was voted by teenagers as their favorite resale website. === Acquisition by Etsy === In June 2021, Depop was acquired by Etsy for $1.6 billion in cash, making it Etsy's most expensive acquisition; however, Depop continues to operate as a standalone brand independent from Etsy. This means that in addition to Depop keeping its existing team, the company retained its London location. At the time of acquisition, Etsy CEO Josh Silverman’s goal was to counteract the influx of buyers starting to go back to physical shops for their purchases. He saw Depop for its potential as a platform supporting a variety of products and creating a greater community of users. According to Silverman, Depop may expand and improve its services for its significant Gen Z user base. For Etsy, this acquisition maintains the company's foothold in the clothing industry and allows the company to expand its customer base to a younger demographic; at the same time, Depop is now able to make use of Etsy's company operations. When Maria Raga relinquished her position as Depop's CEO in 2022, Etsy assigned the role to Kruti Patel Goyal, who was Etsy's former chief product officer and a leader there for eleven years. When Goyal was appointed president and chief growth officer for Etsy in May, Peter Semple, former chief marketing officer, was assigned CEO of Depop officially on August 1st. === Acquisition by eBay === In February 2026, Etsy announced a proposed sale of Depop to eBay for $1.2 billion that was estimated to close within the year. == Business model == === Selling === Depop operates as a marketplace and social platform, where users can follow friends and other influencers to view their buying and selling activities. Through the platform, users are able to sell branded and designer items, as well as vintage pieces. Depop users are also encouraged by the platform to use other social networking services such as Instagram to promote their shop profiles. Celebrities have resold their own items on Depop, with some donating proceeds to charitable causes. Depop's user interface is modeled after that of Instagram. According to Depop, users who list and sell items provide their own photos with item descriptions. Users also note their designer items' authenticity and if they include any labels, tags, and receipts. These listings will appear in users' feeds. The platform's "Explore" page features items picked out by Depop staff. According to Depop, purchases are made via Apple Pay, Google Pay, credit and debit cards, and Klarna. Depop payments stay in-app, allowing for the company to mediate disputes and process refunds. Depop payments allow sellers to directly receive their payments in their bank account. To get paid by Depop, a seller has to add a bank account and verify their identification by uploading an ID. On July 18, 2024, Depop CEO Kruti Patel Goyal announced the removal of selling fees for US sellers, while maintaining a payment processing fee. This policy adjustment aimed to enhance seller revenue and support the growth of the second-hand market. === Buying === A Depop transaction includes the agreed sale price of the item, shipping fees, VAT or other applicable taxes and duties, and the marketplace fee for buyers in the U.S. or U.K. For international deliveries, packages may be subject to import taxes, customs duties, or fees, payable upon arrival or at checkout if Depop collects the tax on behalf of the buyer. For domestic purchases, relevant taxes may be collected by the seller or charged by the platform at checkout, ensuring no additional taxes are due upon delivery. For users in Australia, the United Kingdom, and the United States, Depop allows users to receive a full refund if their item does not arrive, arrives damaged, or is considerably different from the original when the issue is reported within 30 days. === Competitors === As of June 2021, Depop's competitors include Vinted, a platform founded by Milda Mitkute and Justas Janauskas in 2008 and valued at €3.5 billion, as well as the U.S. resale site Poshmark, valued at $3.5 billion. Additional competitors include Grailed, a peer-to-peer e-commerce site founded in 2014 that is recognized for its high-end second-hand menswear and streetwear, and Vestiaire Collection, a European resale app established in 2009 which specializes in authenticated pre-owned luxury items. The popularity of Depop has negatively impacted traditional second-hand stores, which can struggle to compete due to high labor costs and quality demands. There is an oversupply of clothes with the rise of fast fashion; this has taken a toll on the revenue aspect of the second-hand clothing industry. == Criticism == In November 2019, Business of Fashion reported that users within the Depop app were receiving sexually suggestive messages. In February 2020, Jessica Hamilton, a Depop buyer, reported that she found many scammers on the platform. She noticed this issue after she attempted to purchase a Nintendo Switch from a seller who would suspiciously only accept payment through a direct bank transfer without buyer protection. Hamilton blamed the company for its lack of action and relaxed security measures compared to other e-commerce sites, which made the platform especially susceptible to hackers. Without a clear strategy for managing scams, Depop lost some users' trust because of its negligence. In October 2020, some Depop buyers were tricked into paying sellers directly to bypass Depop's buyer protections, and the Depop sellers then sold those users' information on the dark web. In response, Depop claimed that it would improve security through mandatory password updates and multi-factor authentication. Users have criticized Depop for belatedly taking action against this issue.

  Read more →
• 

  Weak supervision

  Weak supervision (also known as semi-supervised learning) is a paradigm in machine learning, the relevance and notability of which increased with the advent of large language models due to the large amount of data required to train them. It is characterized by using a combination of a small amount of human-labeled data (exclusively used in more expensive and time-consuming supervised learning paradigm), followed by a large amount of unlabeled data (used exclusively in unsupervised learning paradigm). In other words, the desired output values are provided only for a subset of the training data. The remaining data is unlabeled or imprecisely labeled. Intuitively, it can be seen as an exam and labeled data as sample problems that the teacher solves for the class as an aid in solving another set of problems. In the transductive setting, these unsolved problems act as exam questions. In the inductive setting, they become practice problems of the sort that will make up the exam. == Problem == The acquisition of labeled data for a learning problem often requires a skilled human agent (e.g. to transcribe an audio segment) or a physical experiment (e.g. determining the 3D structure of a protein or determining whether there is oil at a particular location). The cost associated with the labeling process thus may render large, fully labeled training sets infeasible, whereas acquisition of unlabeled data is relatively inexpensive. In such situations, semi-supervised learning can be of great practical value. Semi-supervised learning is also of theoretical interest in machine learning and as a model for human learning. == Technique == More formally, semi-supervised learning assumes a set of l {\displaystyle l} independently identically distributed examples x 1 , … , x l ∈ X {\displaystyle x_{1},\dots ,x_{l}\in X} with corresponding labels y 1 , … , y l ∈ Y {\displaystyle y_{1},\dots ,y_{l}\in Y} and u {\displaystyle u} unlabeled examples x l + 1 , … , x l + u ∈ X {\displaystyle x_{l+1},\dots ,x_{l+u}\in X} are processed. Semi-supervised learning combines this information to surpass the classification performance that can be obtained either by discarding the unlabeled data and doing supervised learning or by discarding the labels and doing unsupervised learning. Semi-supervised learning may refer to either transductive learning or inductive learning. The goal of transductive learning is to infer the correct labels for the given unlabeled data x l + 1 , … , x l + u {\displaystyle x_{l+1},\dots ,x_{l+u}} only. The goal of inductive learning is to infer the correct mapping from X {\displaystyle X} to Y {\displaystyle Y} . It is unnecessary (and, according to Vapnik's principle, imprudent) to perform transductive learning by way of inferring a classification rule over the entire input space; however, in practice, algorithms formally designed for transduction or induction are often used interchangeably. == Assumptions == In order to make any use of unlabeled data, some relationship to the underlying distribution of data must exist. Semi-supervised learning algorithms make use of at least one of the following assumptions: === Continuity / smoothness assumption === Points that are close to each other are more likely to share a label. This is also generally assumed in supervised learning and yields a preference for geometrically simple decision boundaries. In the case of semi-supervised learning, the smoothness assumption additionally yields a preference for decision boundaries in low-density regions, so few points are close to each other but in different classes. === Cluster assumption === The data tend to form discrete clusters, and points in the same cluster are more likely to share a label (although data that shares a label may spread across multiple clusters). This is a special case of the smoothness assumption and gives rise to feature learning with clustering algorithms. === Manifold assumption === The data lie approximately on a manifold of much lower dimension than the input space. In this case learning the manifold using both the labeled and unlabeled data can avoid the curse of dimensionality. Then learning can proceed using distances and densities defined on the manifold. The manifold assumption is practical when high-dimensional data are generated by some process that may be hard to model directly, but which has only a few degrees of freedom. For instance, human voice is controlled by a few vocal folds, and images of various facial expressions are controlled by a few muscles. In these cases, it is better to consider distances and smoothness in the natural space of the generating problem, rather than in the space of all possible acoustic waves or images, respectively. == History == The heuristic approach of self-training (also known as self-learning or self-labeling) is historically the oldest approach to semi-supervised learning, with examples of applications starting in the 1960s. The transductive learning framework was formally introduced by Vladimir Vapnik in the 1970s. Interest in inductive learning using generative models also began in the 1970s. A probably approximately correct learning bound for semi-supervised learning of a Gaussian mixture was demonstrated by Ratsaby and Venkatesh in 1995. == Methods == === Generative models === Generative approaches to statistical learning first seek to estimate p ( x | y ) {\displaystyle p(x|y)} , the distribution of data points belonging to each class. The probability p ( y | x ) {\displaystyle p(y|x)} that a given point x {\displaystyle x} has label y {\displaystyle y} is then proportional to p ( x | y ) p ( y ) {\displaystyle p(x|y)p(y)} by Bayes' rule. Semi-supervised learning with generative models can be viewed either as an extension of supervised learning (classification plus information about p ( x ) {\displaystyle p(x)} ) or as an extension of unsupervised learning (clustering plus some labels). Generative models assume that the distributions take some particular form p ( x | y , θ ) {\displaystyle p(x|y,\theta )} parameterized by the vector θ {\displaystyle \theta } . If these assumptions are incorrect, the unlabeled data may actually decrease the accuracy of the solution relative to what would have been obtained from labeled data alone. However, if the assumptions are correct, then the unlabeled data necessarily improves performance. The unlabeled data are distributed according to a mixture of individual-class distributions. In order to learn the mixture distribution from the unlabeled data, it must be identifiable, that is, different parameters must yield different summed distributions. Gaussian mixture distributions are identifiable and commonly used for generative models. The parameterized joint distribution can be written as p ( x , y | θ ) = p ( y | θ ) p ( x | y , θ ) {\displaystyle p(x,y|\theta )=p(y|\theta )p(x|y,\theta )} by using the chain rule. Each parameter vector θ {\displaystyle \theta } is associated with a decision function f θ ( x ) = argmax y p ( y | x , θ ) {\displaystyle f_{\theta }(x)={\underset {y}{\operatorname {argmax} }}\ p(y|x,\theta )} . The parameter is then chosen based on fit to both the labeled and unlabeled data, weighted by λ {\displaystyle \lambda } : argmax Θ ( log ⁡ p ( { x i , y i } i = 1 l | θ ) + λ log ⁡ p ( { x i } i = l + 1 l + u | θ ) ) {\displaystyle {\underset {\Theta }{\operatorname {argmax} }}\left(\log p(\{x_{i},y_{i}\}_{i=1}^{l}|\theta )+\lambda \log p(\{x_{i}\}_{i=l+1}^{l+u}|\theta )\right)} === Low-density separation === Another major class of methods attempts to place boundaries in regions with few data points (labeled or unlabeled). One of the most commonly used algorithms is the transductive support vector machine, or TSVM (which, despite its name, may be used for inductive learning as well). Whereas support vector machines for supervised learning seek a decision boundary with maximal margin over the labeled data, the goal of TSVM is a labeling of the unlabeled data such that the decision boundary has maximal margin over all of the data. In addition to the standard hinge loss ( 1 − y f ( x ) ) + {\displaystyle (1-yf(x))_{+}} for labeled data, a loss function ( 1 − | f ( x ) | ) + {\displaystyle (1-|f(x)|)_{+}} is introduced over the unlabeled data by letting y = sign ⁡ f ( x ) {\displaystyle y=\operatorname {sign} {f(x)}} . TSVM then selects f ∗ ( x ) = h ∗ ( x ) + b {\displaystyle f^{}(x)=h^{}(x)+b} from a reproducing kernel Hilbert space H {\displaystyle {\mathcal {H}}} by minimizing the regularized empirical risk: f ∗ = argmin f ( ∑ i = 1 l ( 1 − y i f ( x i ) ) + + λ 1 ‖ h ‖ H 2 + λ 2 ∑ i = l + 1 l + u ( 1 − | f ( x i ) | ) + ) {\displaystyle f^{}={\underset {f}{\operatorname {argmin} }}\left(\displaystyle \sum _{i=1}^{l}(1-y_{i}f(x_{i}))_{+}+\lambda _{1}\|h\|_{\mathcal {H}}^{2}+\lambda _{2}\sum _{i=l+1}^{l+u}(1-|f(x_{i})|)_{+}\right)} An exact solution is intractable due to the non-convex term ( 1 − | f ( x ) | ) + {\displayst

  Read more →
• 

  LRE Map

  The LRE Map (Language Resources and Evaluation) is a freely accessible large database on resources dedicated to Natural language processing. The original feature of LRE Map is that the records are collected during the submission of different major Natural language processing conferences. The records are then cleaned and gathered into a global database called "LRE Map". The LRE Map is intended to be an instrument for collecting information about language resources and to become, at the same time, a community for users, a place to share and discover resources, discuss opinions, provide feedback, discover new trends, etc. It is an instrument for discovering, searching and documenting language resources, here intended in a broad sense, as both data and tools. The large amount of information contained in the Map can be analyzed in many different ways. For instance, the LRE Map can provide information about the most frequent type of resource, the most represented language, the applications for which resources are used or are being developed, the proportion of new resources vs. already existing ones, or the way in which resources are distributed to the community. == Context == Several institutions worldwide maintain catalogues of language resources (ELRA, LDC, NICT Universal Catalogue, ACL Data and Code Repository, OLAC, LT World, etc.) However, it has been estimated that only 10% of existing resources are known, either through distribution catalogues or via direct publicity by providers (web sites and the like). The rest remains hidden, the only occasions where it briefly emerges being when a resource is presented in the context of a research paper or report at some conference. Even in this case, nevertheless, it might be that a resource remains in the background simply because the focus of the research is not on the resource per se. == History == The LRE Map originated under the name "LREC Map" during the preparation of LREC 2010 conference. More specifically, the idea was discussed within the FlaReNet project, and in collaboration with ELRA and the Institute of Computational Linguistics of CNR in Pisa, the Map was put in place at LREC 2010. The LREC organizers asked the authors to provide some basic information about all the resources (in a broad sense, i.e. including tools, standards and evaluation packages), either used or created, described in their papers. All these descriptors were then gathered in a global matrix called the LREC Map. The same methodology and requirements from the authors has been then applied and extended to other conferences, namely COLING-2010, EMNLP-2010, RANLP-2011, LREC 2012, LREC 2014 and LREC 2016. After this generalization to other conferences, the LREC Map has been renamed as the LRE Map. == Size and content == The size of the database increases over time. The data collected amount to 4776 entries. Each resource is described according to the following attributes: Resource type, e.g. lexicon, annotation tool, tagger/parser. Resource production status, e.g. newly created finished, existing-updated. Resource availability, e.g. freely available, from data center. Resource modality, e.g. speech, written, sign language. Resource use, e.g. named entity recognition, language identification, machine translation. Resource language, e.g. English, 23 European Union languages, official languages of India. == Uses == The LRE map is a very important tool to chart the NLP field. Compared to other studied based on subjective scorings, the LRE map is made of real facts. The map has a great potential for many uses, in addition to being an information gathering tool: It is a great instrument for monitoring the evolution of the field (useful for funders), if applied in different contexts and times. It can be seen as a huge joint effort, the beginning of an even larger cooperative action not just among few leaders but among all the researchers. It is also an "educational" means towards the broad acknowledgment of the need of meta-research activities with the active involvement of many. It is also instrumental in introducing the new notion of "citation of resources" that could provide an award and a means of scholarly recognition for researchers engaged in resource creation. It is used to help the organization of the conferences of the field like LREC. == Derived matrices == The data were then cleaned and sorted by Joseph Mariani (CNRS-LIMSI IMMI) and Gil Francopoulo (CNRS-LIMSI IMMI + Tagmatica) in order to compute the various matrices of the final FLaReNet reports. One of them, the matrix for written data at LREC 2010 is as follows: English is the most studied language. Secondly, come French and German languages and then Italian and Spanish. == Future == The LRE Map has been extended to Language Resources and Evaluation Journal and other conferences.

  Read more →
• 

  Machine vision

  Machine vision is the technology and methods used to provide imaging-based automatic inspection and analysis for such applications as automatic inspection, process control, and robot guidance, usually in industry. Machine vision refers to many technologies, software and hardware products, integrated systems, actions, methods and expertise. Machine vision as a systems engineering discipline can be considered distinct from computer vision, a form of computer science. It attempts to integrate existing technologies in new ways and apply them to solve real world problems. The term is the prevalent one for these functions in industrial automation environments but is also used for these functions in other environment vehicle guidance. The overall machine vision process includes planning the details of the requirements and project, and then creating a solution. During run-time, the process starts with imaging, followed by automated analysis of the image and extraction of the required information. == Definition == Definitions of the term "Machine vision" vary, but all include the technology and methods used to extract information from an image on an automated basis, as opposed to image processing, where the output is another image. The information extracted can be a simple good-part/bad-part signal, or more a complex set of data such as the identity, position and orientation of each object in an image. The information can be used for such applications as automatic inspection and robot and process guidance in industry, for security monitoring and vehicle guidance. This field encompasses a large number of technologies, software and hardware products, integrated systems, actions, methods and expertise. Machine vision is practically the only term used for these functions in industrial automation applications; the term is less universal for these functions in other environments such as security and vehicle guidance. Machine vision as a systems engineering discipline can be considered distinct from computer vision, a form of basic computer science; machine vision attempts to integrate existing technologies in new ways and apply them to solve real world problems in a way that meets the requirements of industrial automation and similar application areas. The term is also used in a broader sense by trade shows and trade groups such as the Automated Imaging Association and the European Machine Vision Association. This broader definition also encompasses products and applications most often associated with image processing. The primary uses for machine vision are automatic inspection and industrial robot/process guidance. In more recent times the terms computer vision and machine vision have converged to a greater degree. See glossary of machine vision. == Imaging based automatic inspection and sorting == The primary uses for machine vision are imaging-based automatic inspection and sorting and robot guidance.; in this section the former is abbreviated as "automatic inspection". The overall process includes planning the details of the requirements and project, and then creating a solution. This section describes the technical process that occurs during the operation of the solution. === Methods and sequence of operation === The first step in the automatic inspection sequence of operation is acquisition of an image, typically using cameras, lenses, and lighting that has been designed to provide the differentiation required by subsequent processing. MV software packages and programs developed in them then employ various digital image processing techniques to extract the required information, and often make decisions (such as pass/fail) based on the extracted information. === Equipment === The components of an automatic inspection system usually include lighting, a camera or other imager, a processor, software, and output devices. === Imaging === The imaging device (e.g. camera) can either be separate from the main image processing unit or combined with it in which case the combination is generally called a smart camera or smart sensor. Inclusion of the full processing function into the same enclosure as the camera is often referred to as embedded processing. When separated, the connection may be made to specialized intermediate hardware, a custom processing appliance, or a frame grabber within a computer using either an analog or standardized digital interface (Camera Link, CoaXPress). MV implementations also use digital cameras capable of direct connections (without a framegrabber) to a computer via FireWire, USB or Gigabit Ethernet interfaces. While conventional (2D visible light) imaging is most commonly used in MV, alternatives include multispectral imaging, hyperspectral imaging, imaging various infrared bands, line scan imaging, 3D imaging of surfaces and X-ray imaging. Key differentiations within MV 2D visible light imaging are monochromatic vs. color, frame rate, resolution, and whether or not the imaging process is simultaneous over the entire image, making it suitable for moving processes. Though the vast majority of machine vision applications are solved using two-dimensional imaging, machine vision applications utilizing 3D imaging are a growing niche within the industry. The most commonly used method for 3D imaging is scanning based triangulation which utilizes motion of the product or image during the imaging process. A laser is projected onto the surfaces of an object. In machine vision this is accomplished with a scanning motion, either by moving the workpiece, or by moving the camera & laser imaging system. The line is viewed by a camera from a different angle; the deviation of the line represents shape variations. Lines from multiple scans are assembled into a depth map or point cloud. Stereoscopic vision is used in special cases involving unique features present in both views of a pair of cameras. Other 3D methods used for machine vision are time of flight and grid based. One method is grid array based systems using pseudorandom structured light system as employed by the Microsoft Kinect system circa 2012. === Image processing === After an image is acquired, it is processed. Central processing functions are generally done by a CPU, a GPU, a FPGA or a combination of these. Deep learning training and inference impose higher processing performance requirements. Multiple stages of processing are generally used in a sequence that ends up as a desired result. A typical sequence might start with tools such as filters which modify the image, followed by extraction of objects, then extraction (e.g. measurements, reading of codes) of data from those objects, followed by communicating that data, or comparing it against target values to create and communicate "pass/fail" results. Machine vision image processing methods include; Stitching/Registration: Combining of adjacent 2D or 3D images. Filtering (e.g. morphological filtering) Thresholding: Thresholding starts with setting or determining a gray value that will be useful for the following steps. The value is then used to separate portions of the image, and sometimes to transform each portion of the image to simply black and white based on whether it is below or above that grayscale value. Pixel counting: counts the number of light or dark pixels Segmentation: Partitioning a digital image into multiple segments to simplify and/or change the representation of an image into something that is more meaningful and easier to analyze. Edge detection: finding object edges Color Analysis: Identify parts, products and items using color, assess quality from color, and isolate features using color. Blob detection and extraction: inspecting an image for discrete blobs of connected pixels (e.g. a black hole in a grey object) as image landmarks. Neural network / deep learning / machine learning processing: weighted and self-training multi-variable decision making Circa 2019 there is a large expansion of this, using deep learning and machine learning to significantly expand machine vision capabilities. The most common result of such processing is classification. Examples of classification are object identification,"pass fail" classification of identified objects and OCR. Pattern recognition including template matching. Finding, matching, and/or counting specific patterns. This may include location of an object that may be rotated, partially hidden by another object, or varying in size. Barcode, Data Matrix and "2D barcode" reading Optical character recognition: automated reading of text such as serial numbers Gauging/Metrology: measurement of object dimensions (e.g. in pixels, inches or millimeters) Comparison against target values to determine a "pass or fail" or "go/no go" result. For example, with code or bar code verification, the read value is compared to the stored target value. For gauging, a measurement is compared against the proper value and tolerances. For verification of alpha-numberic codes, the

  Read more →
• 

  Box blur

  A box blur (also known as a box linear filter) is a spatial domain linear filter in which each pixel in the resulting image has a value equal to the average value of its neighboring pixels in the input image. It is a form of low-pass ("blurring") filter. A 3 by 3 box blur ("radius 1") can be written as matrix 1 9 [ 1 1 1 1 1 1 1 1 1 ] . {\displaystyle {\frac {1}{9}}{\begin{bmatrix}1&1&1\\1&1&1\\1&1&1\end{bmatrix}}.} Due to its property of using equal weights, it can be implemented using a much simpler accumulation algorithm, which is significantly faster than using a sliding-window algorithm. Box blurs are frequently used to approximate a Gaussian blur. By the central limit theorem, repeated application of a box blur will approximate a Gaussian blur. In the frequency domain, a box blur has zeros and negative components. That is, a sine wave with a period equal to the size of the box will be blurred away entirely, and wavelengths shorter than the size of the box may be phase-reversed, as seen when two bokeh circles touch to form a bright spot where there would be a dark spot between two bright spots in the original image. == Extensions == Gwosdek, et al. has extended Box blur to take a fractional radius: the edges of the 1-D filter are expanded with a fraction. It makes slightly better gaussian approximation possible due to the elimination of integer-rounding error. Mario Klingemann has a "stack blur" that tries to better emulate gaussian's look in one pass by stacking weights: 1 9 [ 1 2 3 2 1 ] {\displaystyle {\frac {1}{9}}{\begin{bmatrix}1&2&3&2&1\end{bmatrix}}} The triangular impulse response it forms decomposes to two rounds of box blur. Stacked Integral Image by Bhatia et al. takes the weighted average of a few box blurs to fit the gaussian response curve. == Implementation == The following pseudocode implements a 3x3 box blur. The example does not handle the edges of the image, which would not fit inside the kernel, so that these areas remain unblurred. In practice, the issue is better handled by: Introducing an alpha channel to represent the absence of colors; Extending the boundary by filling in values, ranked by quality: Fill in a mirrored image at the border Fill in a constant color extending from the last pixel Pad in a fixed color A number of optimizations can be applied when implementing the box blur of a radius r and N pixels: The box blur is a separable filter, so that only two 1D passes of averaging 2 r + 1 pixels will be needed, one horizontal and one vertical, for each pixel. This lowers the complexity from O(Nr2) to O(Nr). In digital signal processing terminology, each pass is a moving-average filter. Accumulation. Instead of discarding the sum for each pixel, the algorithm re-uses the previous sum, and updates it by subtracting away the old pixel and adding the new pixel in the blurring range. A summed-area table can be used similarly. This lowers the complexity from O(Nr) to O(N). When being used in multiple passes to approximate a Gaussian blur, the cascaded integrator–comb filter construction allows for doing the equivalent operation in a single pass.

  Read more →
• 

  BERT (language model)

  Bidirectional encoder representations from transformers (BERT) is a language model introduced in October 2018 by researchers at Google. It learns to represent text as a sequence of vectors using self-supervised learning. It uses the encoder-only transformer architecture. BERT dramatically improved the state of the art for large language models. As of 2020, BERT is a ubiquitous baseline in natural language processing (NLP) experiments. BERT is trained by masked token prediction and next sentence prediction. With this training, BERT learns contextual, latent representations of tokens in their context, similar to ELMo and GPT-2. It found applications for many natural language processing tasks, such as coreference resolution and polysemy resolution. It improved on ELMo and spawned the study of "BERTology", which attempts to interpret what is learned by BERT. BERT was originally implemented in the English language at two model sizes, BERTBASE (110 million parameters) and BERTLARGE (340 million parameters). Both were trained on the Toronto BookCorpus (800M words) and English Wikipedia (2,500M words). The weights were released on GitHub. On March 11, 2020, 24 smaller models were released, the smallest being BERTTINY with just 4 million parameters. == Architecture == BERT is an "encoder-only" transformer architecture. At a high level, BERT consists of 4 modules: Tokenizer: This module converts a piece of English text into a sequence of integers ("tokens"). Embedding: This module converts the sequence of tokens into an array of real-valued vectors representing the tokens. It represents the conversion of discrete token types into a lower-dimensional Euclidean space. Encoder: a stack of Transformer blocks with self-attention, but without causal masking. Task head: This module converts the final representation vectors into one-shot encoded tokens again by producing a predicted probability distribution over the token types. It can be viewed as a simple decoder, decoding the latent representation into token types, or as an "un-embedding layer". The task head is necessary for pre-training, but it is often unnecessary for so-called "downstream tasks," such as question answering or sentiment classification. Instead, one removes the task head and replaces it with a newly initialized module suited for the task, and finetune the new module. The latent vector representation of the model is directly fed into this new module, allowing for sample-efficient transfer learning. === Embedding === This section describes the embedding used by BERTBASE. The other one, BERTLARGE, is similar, just larger. The tokenizer of BERT is WordPiece, which is a sub-word strategy like byte-pair encoding. Its vocabulary size is 30,000, and any token not appearing in its vocabulary is replaced by [UNK] ("unknown"). The first layer is the embedding layer, which contains three components: token type embeddings, position embeddings, and segment type embeddings. Token type: The token type is a standard embedding layer, translating a one-hot vector into a dense vector based on its token type. Position: The position embeddings are based on a token's position in the sequence. BERT uses absolute position embeddings, where each position in a sequence is mapped to a real-valued vector. Each dimension of the vector consists of a sinusoidal function that takes the position in the sequence as input. Segment type: Using a vocabulary of just 0 or 1, this embedding layer produces a dense vector based on whether the token belongs to the first or second text segment in that input. In other words, type-1 tokens are all tokens that appear after the [SEP] special token. All prior tokens are type-0. The three embedding vectors are added together representing the initial token representation as a function of these three pieces of information. After embedding, the vector representation is normalized using a LayerNorm operation, outputting a 768-dimensional vector for each input token. After this, the representation vectors are passed forward through 12 Transformer encoder blocks, and are decoded back to 30,000-dimensional vocabulary space using a basic affine transformation layer. === Architectural family === The encoder stack of BERT has 2 free parameters: L {\displaystyle L} , the number of layers, and H {\displaystyle H} , the hidden size. There are always H / 64 {\displaystyle H/64} self-attention heads, and the feed-forward/filter size is always 4 H {\displaystyle 4H} . By varying these two numbers, one obtains an entire family of BERT models. For BERT: the feed-forward size and filter size are synonymous. Both of them denote the number of dimensions in the middle layer of the feed-forward network. the hidden size and embedding size are synonymous. Both of them denote the number of real numbers used to represent a token. The notation for encoder stack is written as L/H. For example, BERTBASE is written as 12L/768H, BERTLARGE as 24L/1024H, and BERTTINY as 2L/128H. == Training == === Pre-training === BERT was pre-trained simultaneously on two tasks: Masked language modeling (MLM): In this task, BERT ingests a sequence of words, where one word may be randomly changed ("masked"), and BERT tries to predict the original words that had been changed. For example, in the sentence "The cat sat on the [MASK]," BERT would need to predict "mat." This helps BERT learn bidirectional context, meaning it understands the relationships between words not just from left to right or right to left but from both directions at the same time. Next sentence prediction (NSP): In this task, BERT is trained to predict whether one sentence logically follows another. For example, given two sentences, "The cat sat on the mat" and "It was a sunny day", BERT has to decide if the second sentence is a valid continuation of the first one. This helps BERT understand relationships between sentences, which is important for tasks like question answering or document classification. ==== Masked language modeling ==== In masked language modeling, 15% of tokens would be randomly selected for masked-prediction task, and the training objective was to predict the masked token given its context. In more detail, the selected token is: replaced with a [MASK] token with probability 80%, replaced with a random word token with probability 10%, not replaced with probability 10%. The reason not all selected tokens are masked is to avoid the dataset shift problem. The dataset shift problem arises when the distribution of inputs seen during training differs significantly from the distribution encountered during inference. A trained BERT model might be applied to word representation (like Word2Vec), where it would be run over sentences not containing any [MASK] tokens. It is later found that more diverse training objectives are generally better. As an illustrative example, consider the sentence "my dog is cute". It would first be divided into tokens like "my1 dog2 is3 cute4". Then a random token in the sentence would be picked. Let it be the 4th one "cute4". Next, there would be three possibilities: with probability 80%, the chosen token is masked, resulting in "my1 dog2 is3 [MASK]4"; with probability 10%, the chosen token is replaced by a uniformly sampled random token, such as "happy", resulting in "my1 dog2 is3 happy4"; with probability 10%, nothing is done, resulting in "my1 dog2 is3 cute4". After processing the input text, the model's 4th output vector is passed to its decoder layer, which outputs a probability distribution over its 30,000-dimensional vocabulary space. ==== Next sentence prediction ==== Given two sentences, the model predicts if they appear sequentially in the training corpus, outputting either [IsNext] or [NotNext]. During training, the algorithm sometimes samples two sentences from a single continuous span in the training corpus, while at other times, it samples two sentences from two discontinuous spans. The first sentence starts with a special token, [CLS] (for "classify"). The two sentences are separated by another special token, [SEP] (for "separate"). After processing the two sentences, the final vector for the [CLS] token is passed to a linear layer for binary classification into [IsNext] and [NotNext]. For example: Given "[CLS] my dog is cute [SEP] he likes playing [SEP]", the model should predict [IsNext]. Given "[CLS] my dog is cute [SEP] how do magnets work [SEP]", the model should predict [NotNext]. === Fine-tuning === BERT is meant as a general pretrained model for various applications in natural language processing. That is, after pre-training, BERT can be fine-tuned with fewer resources on smaller datasets to optimize its performance on specific tasks such as natural language inference and text classification, and sequence-to-sequence-based language generation tasks such as question answering and conversational response generation. The original BERT paper published results demonstrating that a small amount of fine

  Read more →
• 

  Augmented Analytics

  Augmented Analytics is an approach of data analytics that employs the use of machine learning and natural language processing to automate analysis processes normally done by a specialist or data scientist. The term was introduced in 2017 by Rita Sallam, Cindi Howson, and Carlie Idoine in a Gartner research paper. Augmented analytics is based on business intelligence and analytics. In the graph extraction step, data from different sources are investigated. == Defining Augmented Analytics == Machine Learning – a systematic computing method that uses algorithms to sift through data to identify relationships, trends, and patterns. It is a process that allows algorithms to dynamically learn from data instead of having a set base of programmed rules. Natural language generation (NLG) – a software capability that takes unstructured data and translates it into plain-English, readable, language. Automating Insights – using machine learning algorithms to automate data analysis processes. Natural Language Query – enabling users to query data using business terms that are either typed onto a search box or spoken. == Data Democratization == Data Democratization is the democratizing data access in order to relieve data congestion and get rid of any sense of data "gatekeepers". This process must be implemented alongside a method for users to make sense of the data. This process is used in hopes of speeding up company decision making and uncovering opportunities hidden in data. There are three aspects to democratising data: Data Parameterisation and Characterisation. Data Decentralisation using an OS of blockchain and DLT technologies, as well as an independently governed secure data exchange to enable trust. Consent Market-driven Data Monetisation. When it comes to connecting assets, there are two features that will accelerate the adoption and usage of data democratisation: decentralized identity management and business data object monetization of data ownership. It enables multiple individuals and organizations to identify, authenticate, and authorize participants and organizations, enabling them to access services, data or systems across multiple networks, organizations, environments, and use cases. It empowers users and enables a personalized, self-service digital onboarding system so that users can self-authenticate without relying on a central administration function to process their information. Simultaneously, decentralized identity management ensures the user is authorized to perform actions subject to the system’s policies based on their attributes (role, department, organization, etc.) and/ or physical location. == Use cases == Agriculture – Farmers collect data on water use, soil temperature, moisture content and crop growth, augmented analytics can be used to make sense of this data and possibly identify insights that the user can then use to make business decisions. Smart Cities – Many cities across the United States, known as Smart Cities collect large amounts of data on a daily basis. Augmented analytics can be used to simplify this data in order to increase effectiveness in city management (transportation, natural disasters, etc.). Analytic Dashboards – Augmented analytics has the ability to take large data sets and create highly interactive and informative analytical dashboards that assist in many organizational decisions. Augmented Data Discovery – Using an augmented analytics process can assist organizations in automatically finding, visualizing and narrating potentially important data correlations and trends. Data Preparation – Augmented analytics platforms have the ability to take large amounts of data and organize and "clean" the data in order for it to be usable for future analyses. Business – Businesses collect large amounts of data, daily. Some examples of types of data collected in business operations include; sales data, consumer behavior data, distribution data. An augmented analytics platform provides access to analysis of this data, which could be used in making business decisions.

  Read more →
• 

  Gradient vector flow

  Gradient vector flow (GVF), a computer vision framework introduced by Chenyang Xu and Jerry L. Prince, is the vector field that is produced by a process that smooths and diffuses an input vector field. It is usually used to create a vector field from images that points to object edges from a distance. It is widely used in image analysis and computer vision applications for object tracking, shape recognition, segmentation, and edge detection. In particular, it is commonly used in conjunction with active contour model. == Background == Finding objects or homogeneous regions in images is a process known as image segmentation. In many applications, the locations of object edges can be estimated using local operators that yield a new image called an edge map. The edge map can then be used to guide a deformable model, sometimes called an active contour or a snake, so that it passes through the edge map in a smooth way, therefore defining the object itself. A common way to encourage a deformable model to move toward the edge map is to take the spatial gradient of the edge map, yielding a vector field. Since the edge map has its highest intensities directly on the edge and drops to zero away from the edge, these gradient vectors provide directions for the active contour to move. When the gradient vectors are zero, the active contour will not move, and this is the correct behavior when the contour rests on the peak of the edge map itself. However, because the edge itself is defined by local operators, these gradient vectors will also be zero far away from the edge and therefore the active contour will not move toward the edge when initialized far away from the edge. Gradient vector flow (GVF) is the process that spatially extends the edge map gradient vectors, yielding a new vector field that contains information about the location of object edges throughout the entire image domain. GVF is defined as a diffusion process operating on the components of the input vector field. It is designed to balance the fidelity of the original vector field, so it is not changed too much, with a regularization that is intended to produce a smooth field on its output. Although GVF was designed originally for the purpose of segmenting objects using active contours attracted to edges, it has been since adapted and used for many alternative purposes. Some newer purposes including defining a continuous medial axis representation, regularizing image anisotropic diffusion algorithms, finding the centers of ribbon-like objects, constructing graphs for optimal surface segmentations, creating a shape prior, and much more. == Theory == The theory of GVF was originally described by Xu and Prince. Let f ( x , y ) {\displaystyle \textstyle f(x,y)} be an edge map defined on the image domain. For uniformity of results, it is important to restrict the edge map intensities to lie between 0 and 1, and by convention f ( x , y ) {\displaystyle \textstyle f(x,y)} takes on larger values (close to 1) on the object edges. The gradient vector flow (GVF) field is given by the vector field v ( x , y ) = [ u ( x , y ) , v ( x , y ) ] {\displaystyle \textstyle \mathbf {v} (x,y)=[u(x,y),v(x,y)]} that minimizes the energy functional In this equation, subscripts denote partial derivatives and the gradient of the edge map is given by the vector field ∇ f = ( f x , f y ) {\displaystyle \textstyle \nabla f=(f_{x},f_{y})} . Figure 1 shows an edge map, the gradient of the (slightly blurred) edge map, and the GVF field generated by minimizing E {\displaystyle \textstyle {\mathcal {E}}} . Equation 1 is a variational formulation that has both a data term and a regularization term. The first term in the integrand is the data term. It encourages the solution v {\displaystyle \textstyle \mathbf {v} } to closely agree with the gradients of the edge map since that will make v − ∇ f {\displaystyle \textstyle \mathbf {v} -\nabla f} small. However, this only needs to happen when the edge map gradients are large since v − ∇ f {\displaystyle \textstyle \mathbf {v} -\nabla f} is multiplied by the square of the length of these gradients. The second term in the integrand is a regularization term. It encourages the spatial variations in the components of the solution to be small by penalizing the sum of all the partial derivatives of v {\displaystyle \textstyle \mathbf {v} } . As is customary in these types of variational formulations, there is a regularization parameter μ > 0 {\displaystyle \textstyle \mu >0} that must be specified by the user in order to trade off the influence of each of the two terms. If μ {\displaystyle \textstyle \mu } is large, for example, then the resulting field will be very smooth and may not agree as well with the underlying edge gradients. Theoretical Solution. Finding v ( x , y ) {\displaystyle \textstyle \mathbf {v} (x,y)} to minimize Equation 1 requires the use of calculus of variations since v ( x , y ) {\displaystyle \textstyle \mathbf {v} (x,y)} is a function, not a variable. Accordingly, the Euler equations, which provide the necessary conditions for v {\displaystyle \textstyle \mathbf {v} } to be a solution can be found by calculus of variations, yielding where ∇ 2 {\displaystyle \textstyle \nabla ^{2}} is the Laplacian operator. It is instructive to examine the form of the equations in (2). Each is a partial differential equation that the components u {\displaystyle u} and v {\displaystyle v} of v {\displaystyle \mathbf {v} } must satisfy. If the magnitude of the edge gradient is small, then the solution of each equation is guided entirely by Laplace's equation, for example ∇ 2 u = 0 {\displaystyle \textstyle \nabla ^{2}u=0} , which will produce a smooth scalar field entirely dependent on its boundary conditions. The boundary conditions are effectively provided by the locations in the image where the magnitude of the edge gradient is large, where the solution is driven to agree more with the edge gradients. Computational Solutions. There are two fundamental ways to compute GVF. First, the energy function E {\displaystyle {\mathcal {E}}} itself (1) can be directly discretized and minimized, for example, by gradient descent. Second, the partial differential equations in (2) can be discretized and solved iteratively. The original GVF paper used an iterative approach, while later papers introduced considerably faster implementations such as an octree-based method, a multi-grid method, and an augmented Lagrangian method. In addition, very fast GPU implementations have been developed in Extensions and Advances. GVF is easily extended to higher dimensions. The energy function is readily written in a vector form as which can be solved by gradient descent or by finding and solving its Euler equation. Figure 2 shows an illustration of a three-dimensional GVF field on the edge map of a simple object (see ). The data and regularization terms in the integrand of the GVF functional can also be modified. A modification described in , called generalized gradient vector flow (GGVF) defines two scalar functions and reformulates the energy as While the choices g ( ∇ f | ) = μ {\displaystyle \textstyle g(\nabla f|)=\mu } and h ( | ∇ f | ) = | ∇ f | 2 {\displaystyle \textstyle h(|\nabla f|)=|\nabla f|^{2}} reduce GGVF to GVF, the alternative choices g ( | ∇ f | ) = exp ⁡ { − | ∇ f | / K } {\displaystyle \textstyle g(|\nabla f|)=\exp\{-|\nabla f|/K\}} and h ( ∇ f | ) = 1 − g ( | ∇ f | ) {\displaystyle \textstyle h(\nabla f|)=1-g(|\nabla f|)} , for K {\displaystyle K} a user-selected constant, can improve the tradeoff between the data term and its regularization in some applications. The GVF formulation has been further extended to vector-valued images in where a weighted structure tensor of a vector-valued image is used. A learning based probabilistic weighted GVF extension was proposed in to further improve the segmentation for images with severely cluttered textures or high levels of noise. The variational formulation of GVF has also been modified in motion GVF (MGVF) to incorporate object motion in an image sequence. Whereas the diffusion of GVF vectors from a conventional edge map acts in an isotropic manner, the formulation of MGVF incorporates the expected object motion between image frames. An alternative to GVF called vector field convolution (VFC) provides many of the advantages of GVF, has superior noise robustness, and can be computed very fast. The VFC field v V F C {\displaystyle \textstyle \mathbf {v} _{\mathrm {VFC} }} is defined as the convolution of the edge map f {\displaystyle f} with a vector field kernel k {\displaystyle \mathbf {k} } where The vector field kernel k {\displaystyle \textstyle \mathbf {k} } has vectors that always point toward the origin but their magnitudes, determined in detail by the function m {\displaystyle m} , decrease to zero with increasing distance from the origin. The beauty of VFC is that it can be computed very rapidly using a fast Fourier tra

  Read more →
• 

  Mean shift

  Mean shift is a non-parametric feature-space mathematical analysis technique for locating the maxima of a density function, a so-called mode-seeking algorithm. Application domains include cluster analysis in computer vision and image processing. == History == The mean shift procedure is usually credited to work by Fukunaga and Hostetler in 1975. It is, however, reminiscent of earlier work by Schnell in 1964. == Overview == Mean shift is a procedure for locating the maxima—the modes—of a density function given discrete data sampled from that function. This is an iterative method, and we start with an initial estimate x {\displaystyle x} . Let a kernel function K ( x i − x ) {\displaystyle K(x_{i}-x)} be given. This function determines the weight of nearby points for re-estimation of the mean. Typically a Gaussian kernel on the distance to the current estimate is used, K ( x i − x ) = e − c | | x i − x | | 2 {\displaystyle K(x_{i}-x)=e^{-c||x_{i}-x||^{2}}} . The weighted mean of the density in the window determined by K {\displaystyle K} is m ( x ) = ∑ x i ∈ N ( x ) K ( x i − x ) x i ∑ x i ∈ N ( x ) K ( x i − x ) {\displaystyle m(x)={\frac {\sum _{x_{i}\in N(x)}K(x_{i}-x)x_{i}}{\sum _{x_{i}\in N(x)}K(x_{i}-x)}}} where N ( x ) {\displaystyle N(x)} is the neighborhood of x {\displaystyle x} , a set of points for which K ( x i − x ) ≠ 0 {\displaystyle K(x_{i}-x)\neq 0} . The difference m ( x ) − x {\displaystyle m(x)-x} is called mean shift in Fukunaga and Hostetler. The mean-shift algorithm now sets x ← m ( x ) {\displaystyle x\leftarrow m(x)} , and repeats the estimation until m ( x ) {\displaystyle m(x)} converges. Although the mean shift algorithm has been widely used in many applications, a rigid proof for the convergence of the algorithm using a general kernel in a high dimensional space is still not known. Aliyari Ghassabeh showed the convergence of the mean shift algorithm in one dimension with a differentiable, convex, and strictly decreasing profile function. However, the one-dimensional case has limited real world applications. Also, the convergence of the algorithm in higher dimensions with a finite number of the stationary (or isolated) points has been proved. However, sufficient conditions for a general kernel function to have finite stationary (or isolated) points have not been provided. Gaussian Mean-Shift is an Expectation–maximization algorithm. == Details == Let data be a finite set S {\displaystyle S} embedded in the n {\displaystyle n} -dimensional Euclidean space, X {\displaystyle X} . Let K {\displaystyle K} be a flat kernel that is the characteristic function of the λ {\displaystyle \lambda } -ball in X {\displaystyle X} , In each iteration of the algorithm, s ← m ( s ) {\displaystyle s\leftarrow m(s)} is performed for all s ∈ S {\displaystyle s\in S} simultaneously. The first question, then, is how to estimate the density function given a sparse set of samples. One of the simplest approaches is to just smooth the data, e.g., by convolving it with a fixed kernel of width h {\displaystyle h} , where x i {\displaystyle x_{i}} are the input samples and k ( r ) {\displaystyle k(r)} is the kernel function (or Parzen window). h {\displaystyle h} is the only parameter in the algorithm and is called the bandwidth. This approach is known as kernel density estimation or the Parzen window technique. Once we have computed f ( x ) {\displaystyle f(x)} from the equation above, we can find its local maxima using gradient ascent or some other optimization technique. The problem with this "brute force" approach is that, for higher dimensions, it becomes computationally prohibitive to evaluate f ( x ) {\displaystyle f(x)} over the complete search space. Instead, mean shift uses a variant of what is known in the optimization literature as multiple restart gradient descent. Starting at some guess for a local maximum, y k {\displaystyle y_{k}} , which can be a random input data point x 1 {\displaystyle x_{1}} , mean shift computes the gradient of the density estimate f ( x ) {\displaystyle f(x)} at y k {\displaystyle y_{k}} and takes an uphill step in that direction. == Types of kernels == Kernel definition: Let X {\displaystyle X} be the n {\displaystyle n} -dimensional Euclidean space, R n {\displaystyle \mathbb {R} ^{n}} . The norm of x {\displaystyle x} is a non-negative number, ‖ x ‖ 2 = x ⊤ x ≥ 0 {\displaystyle \|x\|^{2}=x^{\top }x\geq 0} . A function K : X → R {\displaystyle K:X\rightarrow \mathbb {R} } is said to be a kernel if there exists a profile, k : [ 0 , ∞ ] → R {\displaystyle k:[0,\infty ]\rightarrow \mathbb {R} } , such that K ( x ) = k ( ‖ x ‖ 2 ) {\displaystyle K(x)=k(\|x\|^{2})} and k is non-negative. k is non-increasing: k ( a ) ≥ k ( b ) {\displaystyle k(a)\geq k(b)} if a < b {\displaystyle a Read more →

• 

  Rule-based machine translation

  Rule-based machine translation (RBMT) is a classical approach of machine translation systems based on linguistic information about source and target languages. Such information is retrieved from (unilingual, bilingual or multilingual) dictionaries and grammars covering the main semantic, morphological, and syntactic regularities of each language. Having input sentences, an RBMT system generates output sentences on the basis of analysis of both the source and the target languages involved. RBMT has been progressively superseded by more efficient methods, particularly neural machine translation. == History == The first RBMT systems were developed in the early 1970s. The most important steps of this evolution were the emergence of the following RBMT systems: Systran Japanese MT systems Today, other common RBMT systems include: Apertium GramTrans == Types of RBMT == There are three different types of rule-based machine translation systems: Direct Systems (Dictionary Based Machine Translation) map input to output with basic rules. Transfer RBMT Systems (Transfer Based Machine Translation) employ morphological and syntactical analysis. Interlingual RBMT Systems (Interlingua) use an abstract meaning. RBMT systems can also be characterized as the systems opposite to Example-based Systems of Machine Translation (Example Based Machine Translation), whereas Hybrid Machine Translations Systems make use of many principles derived from RBMT. == Basic principles == The main approach of RBMT systems is based on linking the structure of the given input sentence with the structure of the demanded output sentence, necessarily preserving their unique meaning. The following example can illustrate the general frame of RBMT: A girl eats an apple. Source Language = English; Demanded Target Language = German Minimally, to get a German translation of this English sentence one needs: A dictionary that will map each English word to an appropriate German word. Rules representing regular English sentence structure. Rules representing regular German sentence structure. And finally, we need rules according to which one can relate these two structures together. Accordingly, we can state the following stages of translation: 1st: getting basic part-of-speech information of each source word: a = indef.article; girl = noun; eats = verb; an = indef.article; apple = noun 2nd: getting syntactic information about the verb "to eat": NP-eat-NP; here: eat – Present Simple, 3rd Person Singular, Active Voice 3rd: parsing the source sentence: (NP an apple) = the object of eat Often only partial parsing is sufficient to get to the syntactic structure of the source sentence and to map it onto the structure of the target sentence. 4th: translate English words into German a (category = indef.article) => ein (category = indef.article) girl (category = noun) => Mädchen (category = noun) eat (category = verb) => essen (category = verb) an (category = indef. article) => ein (category = indef.article) apple (category = noun) => Apfel (category = noun) 5th: Mapping dictionary entries into appropriate inflected forms (final generation): A girl eats an apple. => Ein Mädchen isst einen Apfel. == Ontologies == An ontology is a formal representation of knowledge that includes the concepts (such as objects, processes etc.) in a domain and some relations between them. If the stored information is of linguistic nature, one can speak of a lexicon. In NLP, ontologies can be used as a source of knowledge for machine translation systems. With access to a large knowledge base, rule-based systems can be enabled to resolve many (especially lexical) ambiguities on their own. In the following classic examples, as humans, we are able to interpret the prepositional phrase according to the context because we use our world knowledge, stored in our lexicons:I saw a man/star/molecule with a microscope/telescope/binoculars.Since the syntax does not change, a traditional rule-based machine translation system may not be able to differentiate between the meanings. With a large enough ontology as a source of knowledge however, the possible interpretations of ambiguous words in a specific context can be reduced. === Building ontologies === The ontology generated for the PANGLOSS knowledge-based machine translation system in 1993 may serve as an example of how an ontology for NLP purposes can be compiled: A large-scale ontology is necessary to help parsing in the active modules of the machine translation system. In the PANGLOSS example, about 50,000 nodes were intended to be subsumed under the smaller, manually-built upper (abstract) region of the ontology. Because of its size, it had to be created automatically. The goal was to merge the two resources LDOCE online and WordNet to combine the benefits of both: concise definitions from Longman, and semantic relations allowing for semi-automatic taxonomization to the ontology from WordNet. A definition match algorithm was created to automatically merge the correct meanings of ambiguous words between the two online resources, based on the words that the definitions of those meanings have in common in LDOCE and WordNet. Using a similarity matrix, the algorithm delivered matches between meanings including a confidence factor. This algorithm alone, however, did not match all meanings correctly on its own. A second hierarchy match algorithm was therefore created which uses the taxonomic hierarchies found in WordNet (deep hierarchies) and partially in LDOCE (flat hierarchies). This works by first matching unambiguous meanings, then limiting the search space to only the respective ancestors and descendants of those matched meanings. Thus, the algorithm matched locally unambiguous meanings (for instance, while the word seal as such is ambiguous, there is only one meaning of seal in the animal subhierarchy). Both algorithms complemented each other and helped constructing a large-scale ontology for the machine translation system. The WordNet hierarchies, coupled with the matching definitions of LDOCE, were subordinated to the ontology's upper region. As a result, the PANGLOSS MT system was able to make use of this knowledge base, mainly in its generation element. == Components == The RBMT system contains: a SL morphological analyser - analyses a source language word and provides the morphological information; a SL parser - is a syntax analyser which analyses source language sentences; a translator - used to translate a source language word into the target language; a TL morphological generator - works as a generator of appropriate target language words for the given grammatica information; a TL parser - works as a composer of suitable target language sentences; Several dictionaries - more specifically a minimum of three dictionaries: a SL dictionary - needed by the source language morphological analyser for morphological analysis, a bilingual dictionary - used by the translator to translate source language words into target language words, a TL dictionary - needed by the target language morphological generator to generate target language words. The RBMT system makes use of the following: a Source Grammar for the input language which builds syntactic constructions from input sentences; a Source Lexicon which captures all of the allowable vocabulary in the domain; Source Mapping Rules which indicate how syntactic heads and grammatical functions in the source language are mapped onto domain concepts and semantic roles in the interlingua; a Domain Model/Ontology which defines the classes of domain concepts and restricts the fillers of semantic roles for each class; Target Mapping Rules which indicate how domain concepts and semantic roles in the interlingua are mapped onto syntactic heads and grammatical functions in the target language; a Target Lexicon which contains appropriate target lexemes for each domain concept; a Target Grammar for the target language which realizes target syntactic constructions as linearized output sentences. == Advantages == No bilingual texts are required. This makes it possible to create translation systems for languages that have no texts in common, or even no digitized data whatsoever. Domain independent. Rules are usually written in a domain independent manner, so the vast majority of rules will "just work" in every domain, and only a few specific cases per domain may need rules written for them. No quality ceiling. Every error can be corrected with a targeted rule, even if the trigger case is extremely rare. This is in contrast to statistical systems where infrequent forms will be washed away by default. Total control. Because all rules are hand-written, you can easily debug a rule-based system to see exactly where a given error enters the system, and why. Reusability. Because RBMT systems are generally built from a strong source language analysis that is fed to a transfer step and target language generator, the source language analysis and targe

  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 →
• 

  Adversarial stylometry

  Adversarial stylometry is the practice of altering writing style to reduce the potential for stylometry to discover the author's identity or their characteristics. This task is also known as authorship obfuscation or authorship anonymisation. Stylometry poses a significant privacy challenge in its ability to unmask anonymous authors or to link pseudonyms to an author's other identities, which, for example, creates difficulties for whistleblowers, activists, and hoaxers and fraudsters. The privacy risk is expected to grow as machine learning techniques and text corpora develop. All adversarial stylometry shares the core idea of faithfully paraphrasing the source text so that the meaning is unchanged but the stylistic signals are obscured. Such a faithful paraphrase is an adversarial example for a stylometric classifier. Several broad approaches to this exist, with some overlap: imitation, substituting the author's own style for another's; translation, applying machine translation with the hope that this eliminates characteristic style in the source text; and obfuscation, deliberately modifying a text's style to make it not resemble the author's own. Manually obscuring style is possible, but laborious; in some circumstances, it is preferable or necessary. Automated tooling, either semi- or fully-automatic, could assist an author. How best to perform the task and the design of such tools is an open research question. While some approaches have been shown to be able to defeat particular stylometric analyses, particularly those that do not account for the potential of adversariality, establishing safety in the face of unknown analyses is an issue. Ensuring the faithfulness of the paraphrase is a critical challenge for automated tools. It is uncertain if the practice of adversarial stylometry is detectable in itself. Some studies have found that particular methods produced signals in the output text, but a stylometrist who is uncertain of what methods may have been used may not be able to reliably detect them. == History == Rao & Rohatgi (2000), an early work in adversarial stylometry, identified machine translation as a possibility, but noted that the quality of translators available at the time presented severe challenges. Kacmarcik & Gamon (2006) is another early work. Brennan, Afroz & Greenstadt (2012) performed the first evaluation of adversarial stylometric methods on actual texts. Brennan & Greenstadt (2009) introduced the first corpus of adversarially authored texts specifically for evaluating stylometric methods; other corpora include the International Imitation Hemingway Competition, the Faux Faulkner contest, and the hoax blog A Gay Girl in Damascus. == Motivations == Rao & Rohatgi (2000) suggest that short, unattributed documents (i.e., anonymous posts) are not at risk of stylometric identification, but pseudonymous authors who have not practiced adversarial stylometry in producing corpuses of thousands of words may be vulnerable. Narayanan et al. (2012) attempted large-scale deanonymisation of 100,000 blog authors with mixed results: the identifications were significantly better than chance, but only accurately matched the blog and author a fifth of the time; identification improved with the number of posts written by the author in the corpus. Even if an author is not identified, some of their characteristics may still be deduced stylometrically, or stylometry may narrow the anonymity set of potential authors sufficiently for other information to complete the identification. Detecting author characteristics (e.g., gender or age) is often simpler than identifying an author from a large, possibly open, set of candidates. Modern machine learning techniques offer powerful tools for identification; further development of corpora and computational stylometric techniques are likely to raise further privacy issues. Gröndahl & Asokan (2020a) say that the general validity of the hypothesis underlying stylometry—that authors have invariant, content-independent 'style fingerprints'—is uncertain, but "the deanonymisation attack is a real privacy concern". Those interested in practicing adversarial stylometry and stylistic deception include whistleblowers avoiding retribution; journalists and activists; perpetrators of frauds and hoaxes; authors of fake reviews; literary forgers; criminals disguising their identity from investigators; and, generally, anyone with a desire for anonymity or pseudonymity. Authors, or agents acting on behalf of authors, may also attempt to remove stylistic clues to author characteristics (e.g., race or gender) so that knowledge of those characteristics cannot be used for discrimination (e.g., through algorithmic bias). Another possible use for adversarial stylometry is in disguising automatically generated text as human-authored. == Methods == With imitation, the author attempts to mislead stylometry by matching their style to another author's. An incomplete imitation, where some of the true author's unique characteristics appear alongside the imitated author's, can be a detectable signal for the use of adversarial stylometry. Imitation can be performed automatically with style transfer systems, though this typically requires a large corpus in the target style for the system to learn from. Another approach is translation, which employs machine translation of a source text to eliminate characteristic style, often through multiple translators in sequence to produce a round-trip translation. Such chained translation can lead to texts being significantly altered, even to the point of incomprehensibility; improved translation tools reduce this risk. More simply-structured texts can be easier to machine translate without losing the original meaning. Machine translation blurs into direct stylistic imitation or obfuscation achieved through automated style transfer, which can be viewed as a "translation" with the same language as input and output. With low-quality translation tools, an author can be required to manually correct major translation errors while avoiding the hazard of re-introducing stylistic characteristics. Wang, Juola & Riddell (2022) found that gross errors introduced by Google Translate were rare, but more common with several intermediate translations—however, occasional simple or short sentences and misspellings in the source text appeared verbatim in the output, potentially providing an identifying signal. Chain translation can leave characteristic traces of its application in a document, which may allow reconstruction of the intermediate languages used and the number of translation steps performed. Obfuscation involves deliberately changing the style of a text to reduce its similarity to other texts by some metric; this may be performed at the time of writing by conscious modification, or as part of a revision process with feedback from the metric being targeted as an input to decide when the text has been sufficiently obfuscated. In contrast to translation, complex texts can offer more opportunities for effective obfuscation without altering meaning, and likewise genres with more permissible variation allow more obfuscation. However, longer texts are harder to thoroughly obfuscate. Obfuscation can blend into imitation if the author develops a novel target style, distinct from their original style. With respect to masking author characteristics, obfuscation may aim to achieve a union (adding signals for imitated characteristics) or an intersection (removing signals and normalising) of other authors' styles. Avoiding the author's own idiosyncrasies and producing a "normalised" text is a critical obfuscatory step: an author may have a unique tendency to misspell certain words, use particular variants, or to format a document in a characteristic way. Stylometric signals vary in how simply they can be adversarially masked; an author may easily change their vocabulary by conscious choice, but altering the pattern of grammar or the letter frequency in their text may be harder to achieve, though Juola & Vescovi (2011) report that imitation typically succeeds at masking more characteristics than obfuscation. Automated obfuscation may require large amounts of training data written by the author. Concerning automated implementations of adversarial stylometry, two possible implementations are rule-based systems for paraphrasing; and encoder–decoder architectures, where the text passes through an intermediate format that is (intended to be) style-neutral. Another division in automated methods is whether there is feedback from an identification system or not. With such feedback, finding paraphrases for author masking has been characterised as a heuristic search problem, exploring textual variants until the result is stylistically sufficiently far (in the case of obfuscation) or near (in the case of imitation), which then constitutes an adversarial example for that identification system. == Evaluation == How

  Read more →
• 

  Marq (company)

  Marq (formerly Lucidpress) is a cloud-based software platform for brand management and templated content creation. The platform integrates with digital asset management (DAM) systems—including Aprimo and Bynder and customer relationship management (CRM) tools such as Salesforce and HubSpot. Marq also includes AI-assisted features for brand compliance and content automation. Trade publications have described the product as a brand templating and creative automation platform. == History == In October 2013, Lucid Software, Inc. announced Lucidpress as a public beta version. Following its release, Lucidpress was featured in TechCrunch, VentureBeat and PC World, with TechCrunch noting: "I had a chance to test the app before its launch and it is indeed very easy to use. If you've ever used a desktop publishing app in the past, you'll feel right at home with Marq, as it features the same kind of standard top-bar menu and layout options as most other publishing apps. In terms of features, it can also hold its own against similar desktop-based apps." In May 2021, Lucidpress announced that it had been acquired by Charles Thayne Capital ("CTC"), a growth-oriented and technology-focused private investment firm. In May 2021, following its acquisition by Charles Thayne Capital, Lucidpress became fully independent. Owen Fuller, who had served as General Manager since 2017, was appointed Chief Executive Officer. In 2022, Lucidpress was rebranded as Marq to reflect the company’s shift toward brand templating and creative automation tools, while continuing to support its publishing features. == Features == Marq integrates with customer relationship management (CRM) platforms such as Salesforce and HubSpot, enabling the creation of personalized, on-brand sales and marketing materials. The platform also connects with multiple digital asset management (DAM) systems, including Bynder, Aprimo, MediaValet, PhotoShelter, Acquia, and Canto. == Investment == Lucid Software raised $1 million in Seed in 2011, led by Google Ventures. In May 2014, the company received a $5 million investment. The round was led by Salt Lake-based Kickstart Seed Fund. In September 2016, the company received a $36 million investment from Spectrum Equity.

  Read more →
• 

  ACL Data Collection Initiative

  The ACL Data Collection Initiative (ACL/DCI) was a project established in 1989 by the Association for Computational Linguistics (ACL) to create and distribute large text and speech corpora for computational linguistics research. The initiative aimed to address the growing need for substantial text databases that could support research in areas such as natural language processing, speech recognition, and computational linguistics. By 1993, the initiative’s activities had effectively ceased, with its functions and datasets absorbed by the Linguistic Data Consortium (LDC), which was founded in 1992. == Objectives == The ACL/DCI had several key objectives: To acquire a large and diverse text corpus from various sources To transform the collected texts into a common format based on the Standard Generalized Markup Language (SGML) To make the corpus available for scientific research at low cost with minimal restrictions To provide a common database that would allow researchers to replicate or extend published results To reduce duplication of effort among researchers in obtaining and preparing text data These objectives were designed to address the growing demand for very large amounts of text arising from applications in recognition and analysis of text and speech. Its core objective was to "oversee the acquisition and preparation of a large text corpus to be made available for scientific research at cost and without royalties". == History == By the late 1980s, researchers in computational linguistics and speech recognition faced a significant problem: the lack of large-scale, accessible text corpora for developing statistical models and testing algorithms. Existing generally available text databases were too small to meet the needs of developing applications in text and speech recognition. The initiative was formed to meet this need by collecting, standardizing, and distributing large quantities of text data with minimal restrictions for scientific research. As stated by Liberman (1990), "research workers have been severely hampered by the lack of appropriate materials, and specially by the lack of a large enough body of text on which published results can be replicated or extended by others." The ACL/DCI committee was established in February 1989. The committee included members from academic and industrial research laboratories in the United States and Europe. The initiative was chaired by Mark Liberman from the University of Pennsylvania (formerly of AT&T Bell Laboratories). Other committee members included representatives from organizations such as Bellcore, IBM T.J. Watson Research Center, Cambridge University, Virginia Polytechnic Institute & State University, Northeastern University, University of Pennsylvania, SRI International, MCC, Xerox PARC, ISSCO, and University of Pisa. The project operated initially without dedicated funding, relying on volunteer efforts from committee members and their affiliated institutions. Key supporters included AT&T Bell Labs, Bellcore, IBM, Xerox, and the University of Pennsylvania, which allowed the use of their computing facilities for ACL/DCI-related work. Previously running on volunteer effort pro bono, in 1991, it obtained funding from General Electric and the National Science Foundation (IRI-9113530). == Data == As of 1990, the ACL/DCI had collected hundreds of millions of words of diverse text. The collection included: Wall Street Journal articles (25 to 50 million words); Canadian Hansard (parliamentary records) in parallel English and French versions: cleaned-up English Hansard donated by the IBM alignment models group (100 million words), and original Bilingual Hansard (from a different time period) obtained directly (200 million words). Collins English Dictionary (1979 edition), both as fulltext (3 million words) and as various "database" versions, constructed using "typographers' tape" donated by Collins, which were computer tapes containing the structured digital data used to typeset and print the 1979 edition of the dictionary; Emails from ARPANET newsletters for the ACM Special Interest Group on Information Retrieval Forum (IRLIST) and AIList Digest issues distributed over the ARPANET (AILIST) (5 million words), both collected by Edward A. Fox at VIPSU; Articles on networking (2 million words); U.S. Department of Agriculture Extension Service Fact Sheets (>1 million words); 200,000 scientific abstracts of about 1,500 words each from the Department of Energy (25 million words); Archives of the Challenger Investigation Commission, including transcripts of depositions and hearings (2.5 million words); Books from the Library of America, including works by Mark Twain, Eugene O'Neill, Ralph Waldo Emerson, Herman Melville, W.E.B. DuBois, Willa Cather, and Benjamin Franklin (130 books, 20 million words); Public domain books like the King James Bible, Tristram Shandy, The Federalist Papers; Several million words of transcribed radiologists' reports, donated by Francis Ganong at Kurzweil Applied Intelligence Inc (about 5 million words); The Child Language Data Exchange corpus of child language acquisition transcripts; U.S. Department of Justice Justice Retrieval and Inquiry System (JURIS) materials; The Swiss Civil Code in parallel German, French and Italian; Economic reports from the Union Bank of Switzerland, in parallel English, German, French and Italian; About 12K words of administrative policy manuals and 14K words of administrative memos, contributed by Geoff Pullum of U.C.S.C.; Material from various ACM journals and the ACL journal Computational Linguistics; The CSLI publications series: 50-100 reports (8K words each) and 5-10 books (80K words each). The initiative started with North American English text but expanded to include Canadian French and planned to include Japanese, Chinese, and other Asian languages. At least 5 million words from the collection were tagged under the Penn Treebank project, and those tags were distributed by DCI as well. After DCI was absorbed by the LDC, the datasets were curated under LDC. == Format == The ACL/DCI corpus was coded in a standard form based on SGML (Standard Generalized Markup Language, ISO 8879), consistent with the recommendations of the Text Encoding Initiative (TEI), of which the DCI was an affiliated project. The TEI was a joint project of the ACL, the Association for Computers and the Humanities, and the Association for Literary and Linguistic Computing, aiming to provide a common interchange format for literary and linguistic data. The initiative planned to add annotations reflecting consensually approved linguistic features like part of speech and various aspects of syntactic and semantic structure over time. == Examples == As an example of the use of ACL/DCI, consider the Wall Street Journal (WSJ) corpus for speech recognition research. The WSJ corpus was used as the basis for the DARPA Spoken Language System (SLS) community's Continuous Speech Recognition (CSR) Corpus. The WSJ corpus became a standard benchmark for evaluating speech recognition systems and has been used in numerous research papers. The WSJ CSR Corpus provided DARPA with its first general-purpose English, large vocabulary, natural language, high perplexity corpus containing speech (400 hours) and text (47 million words) during 1987–89. The text corpus was 313 MB in size. The text was preprocessed to remove ambiguity in the word sequence that a reader might choose, ensuring that the unread text used to train language models was representative of the spoken test material. The preprocessing included converting numbers into orthographics, expanding abbreviations, resolving apostrophes and quotation marks, and marking punctuation. As another example, the Yarowsky algorithm used bitext data from DCI to train a simple word-sense disambiguation model that was competitive with advanced models trained on smaller datasets. == Distribution == Materials from the ACL/DCI collection were distributed to research groups on a non-commercial basis. By 1990, about 25 research groups and individual researchers had received tapes containing various portions of the collected material. To obtain the data, researchers had to sign an agreement not to redistribute the data or make direct commercial use of it. However, commercial application of "analytical materials" derived from the text, such as statistical tables or grammar rules, was explicitly permitted. The initiative first distributed data via 12-inch reels of 9-track tape, then via CD-ROMs. Each such tape could contain 30 million words compressed via the Lempel-Ziv algorithms. The first CD-ROM distribution was in 1991, funded by Dragon Systems Inc. It contained Collins English Dictionary, WSJ, scientific abstracts provided by the U.S. Department of Energy, and the Penn Treebank.

  Read more →
• 

  Production (computer science)

  In computer science, a production or production rule is a rewrite rule that replaces some symbols with other symbols. A finite set of productions P {\displaystyle P} is the main component in the specification of a formal grammar (specifically a generative grammar). In such grammars, a set of productions is a special case of relation on the set of strings V ∗ {\displaystyle V^{}} (where ∗ {\displaystyle {}^{}} is the Kleene star operator) over a finite set of symbols V {\displaystyle V} called a vocabulary that defines which non-empty strings can be substituted with others. The set of productions is thus a special kind subset P ⊂ V ∗ × V ∗ {\displaystyle P\subset V^{}\times V^{}} and productions are then written in the form u → v {\displaystyle u\to v} to mean that ( u , v ) ∈ P {\displaystyle (u,v)\in P} (not to be confused with → {\displaystyle \to } being used as function notation, since there may be multiple rules for the same u {\displaystyle u} ). Given two subsets A , B ⊂ V ∗ {\displaystyle A,B\subset V^{}} , productions can be restricted to satisfy P ⊂ A × B {\displaystyle P\subset A\times B} , in which case productions are said "to be of the form A → B {\displaystyle A\to B} . Different choices and constructions of A , B {\displaystyle A,B} lead to different types of grammars. In general, any production of the form u → ϵ , {\displaystyle u\to \epsilon ,} where ϵ {\displaystyle \epsilon } is the empty string (sometimes also denoted λ {\displaystyle \lambda } ), is called an erasing rule, while productions that would produce strings out of nowhere, namely of the form ϵ → v , {\displaystyle \epsilon \to v,} are never allowed. In order to allow the production rules to create meaningful sentences, the vocabulary is partitioned into (disjoint) sets Σ {\displaystyle \Sigma } and N {\displaystyle N} providing two different roles: Σ {\displaystyle \Sigma } denotes the terminal symbols known as an alphabet containing the symbols allowed in a sentence; N {\displaystyle N} denotes nonterminal symbols, containing a distinguished start symbol S ∈ N {\displaystyle S\in N} , that are needed together with the production rules to define how to build the sentences. In the most general case of an unrestricted grammar, a production u → v {\displaystyle u\to v} , is allowed to map arbitrary strings u {\displaystyle u} and v {\displaystyle v} in V {\displaystyle V} (terminals and nonterminals), as long as u {\displaystyle u} is not empty. So unrestricted grammars have productions of the form V ∗ ∖ { ϵ } → V ∗ {\displaystyle V^{}\setminus \{\epsilon \}\to V^{}} or if we want to disallow changing finished sentences V ∗ N V ∗ = ( V ∗ ∖ Σ ∗ ) → V ∗ {\displaystyle V^{}NV^{}=(V^{}\setminus \Sigma ^{})\to V^{}} , where V ∗ N V ∗ {\displaystyle V^{}NV^{}} indicates concatenation and forces a non-terminal symbol to always be present on the left-hand side of the productions, and ∖ {\displaystyle \setminus } denotes set minus or set difference. If we do not allow the start symbol to occur in v {\displaystyle v} (the word on the right side), we have to replace V ∗ {\displaystyle V^{}} with ( V ∖ { S } ) ∗ {\displaystyle (V\setminus \{S\})^{}} on the right-hand side. The other types of formal grammar in the Chomsky hierarchy impose additional restrictions on what constitutes a production. Notably in a context-free grammar, the left-hand side of a production must be a single nonterminal symbol. So productions are of the form: N → V ∗ {\displaystyle N\to V^{}} == Grammar generation == To generate a string in the language, one begins with a string consisting of only a single start symbol, and then successively applies the rules (any number of times, in any order) to rewrite this string. This stops when a string containing only terminals is obtained. The language consists of all the strings that can be generated in this manner. Any particular sequence of legal choices taken during this rewriting process yields one particular string in the language. If there are multiple different ways of generating this single string, then the grammar is said to be ambiguous. For example, assume the alphabet consists of a {\displaystyle a} and b {\displaystyle b} , with the start symbol S {\displaystyle S} , and we have the following rules: 1. S → a S b {\displaystyle S\rightarrow aSb} 2. S → b a {\displaystyle S\rightarrow ba} then we start with S {\displaystyle S} , and can choose a rule to apply to it. If we choose rule 1, we replace S {\displaystyle S} with a S b {\displaystyle aSb} and obtain the string a S b {\displaystyle aSb} . If we choose rule 1 again, we replace S {\displaystyle S} with a S b {\displaystyle aSb} and obtain the string a a S b b {\displaystyle aaSbb} . This process is repeated until we only have symbols from the alphabet (i.e., a {\displaystyle a} and b {\displaystyle b} ). If we now choose rule 2, we replace S {\displaystyle S} with b a {\displaystyle ba} and obtain the string a a b a b b {\displaystyle aababb} , and are done. We can write this series of choices more briefly, using symbols: S ⇒ a S b ⇒ a a S b b ⇒ a a b a b b {\displaystyle S\Rightarrow aSb\Rightarrow aaSbb\Rightarrow aababb} . The language of the grammar is the set of all the strings that can be generated using this process: { b a , a b a b , a a b a b b , a a a b a b b b , … } {\displaystyle \{ba,abab,aababb,aaababbb,\dotsc \}} .

  Read more →