AI Art Zelda

AI Art Zelda — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  TowIt

  "TowIt" is a free, global, cross-platform mobile app, website, and Web API that allows civilians to report parking violations and dangerous driving in real-time. The mission is to remove the barriers required to make cities effectively fight and deter bad parking and dangerous driving habits. The company ultimately aims to better existing social controls in order to drive necessary behavioral change through increased education, real-time reporting, optimized enforcement, as well as the resulting reactivity. == User base and adoption == The application has users reporting vehicular infractions in upwards of 30 countries. The top reporting countries are: Portugal, Canada, United States of America and Australia. Users have adopted TowIt for a variety of reasons, usually central to their geographical location and the prominent offences in those specific areas. For instance, the majority of Portuguese reports are cars parked on sidewalks, footpaths and pedestrian crossings, Australian reports are largely focused on the abuse of disabled parking spaces, and in Toronto or San Francisco users generally capture cars parked in bicycle lanes. == Functions == === Data collection === TowIt gathers data on individual parking offences, the prominence of various offence types, as well as recurring offenders. This allows the company to identify trends and hotspots in order to take action against problem vehicles, as well as to help improve urban planning, traffic congestion and gridlock management. Individuals modify or improve an aspect of their behavior in response to their awareness of being observed, theoretically more so when demonstrating selfishness, egocentrism, narcissism and anti-social behavior. The company states that by becoming a user, one can "help TowIt relieve congestion, reduce collisions, open up economies, improve the environment and enhance the lives of urban residents and suburban commuters alike". The company has acknowledged that there are numerous legislative changes that would be required to integrate with governments at any level in many countries. A simple three-step process allows users to take a photo of an offending vehicle and subsequently verifying the offending vehicle's license plate information before submitting by tapping the TowIt (submit) button. Photographical evidence can only be captured with the camera from within the TowIt application. An Internet connection is required. The company has stated that this was purposefully done for quality control and report validation purposes. Users may only submit and view their own report history on either the iOS or Android applications. Globally submitted reports are displayed uncensored and in aggregate only on the Android application and the TowIt website. The "Global Feed" feature was removed from iOS (see iTunes Connect Acceptance Issues). TowIt's back-end automatically geotags the report and compares it to local parking by-law data, including by-law types, locations, times, side(s) of street, etc.- where available. Valid reports are posted to the global feed, to the TowIt website, and passed on to municipalities and police for enforcement (where connected). === Technologies used under license === TowIt currently utilizes the following software or software libraries under license: AngularJS, Apache Cordova, Apple iTunes Store EULA, Chart.js, Google Play Distribution Agreement, Ionic Framework, MongoDB, Moment.js, Python 2.7, Python Flask, and jQuery. == Company history == The TowIt application was conceived by Michael Duncan McArthur on December 5, 2014, as a response to Toronto Mayor John Tory's election mandate to "get this city moving". The application was announced via TowIt's official Twitter page on January 6, 2015. After the initial public announcement, Michael & Gregory were contacted by members of John Tory's staff on January 8, 2015, and invited to demo a prototype at Toronto City Hall on January 12, 2015. The two were also invited to meet with Toronto Councillor Norm Kelly, in his City Hall office, for a subsequent demo of the live Android application on January 28, 2015. A similar meeting and demo took place with members of the Traffic Services department of Toronto Police Service on February 2, 2015. Michael & Gregory teamed up with friends and Toronto-based developers Dae-Seon Moon, Jesse Malone, and Marcus Veres to complete the prototype in time to meet the city's imposed demo deadline and to launch the initial Android version of the application. TowIt officially launched on the Android platform on January 16, 2015. A subsequent iOS launch took place on March 19, 2015. === iTunes connect acceptance issues === The iOS version of the application was delayed for approximately two months, only after significant deliberation with Apple's iTunes Connect review board around (as then stated) rule: "14.1 - Any App that is defamatory, offensive, mean-spirited, or likely to place the targeted individual or group in harm's way will be rejected." The result was having to remove the "Global Feed" feature from the iOS platform, in which civilian users could view all recent reports from within the application. This feature still exists on the Android platform. === Business and legal === TowIt engaged Wildeboer Dellelce, one of Canada's leading business law and transactional corporate finance law firms, on January 17, 2015. The company filed for incorporation as "TowIt Solutions Inc." by both Michael & Gregory in the Canadian province of Ontario on January 22, 2015. TowIt continues to operate under a Freemium business model. The company is 100% bootstrapped and has received no outside investment to date. TowIt was accepted into the MaRS Discovery District's Venture Services program on March 4, 2015. === Lobbyist registration === After receiving initial press coverage in January and February 2015, an unknown entity reported Michael & Gregory's initial communications with city staff to the City of Toronto's Lobbyist Registrar. This complaint resulted in legal threats of fines received on February 10, 2015, for apparently and unknowingly breaking municipal lobbying by-laws. These fines (of up to $100,000) were eventually withdrawn after Michael & Gregory immediately provided all records of communication with city officials and registered as lobbyists in the City of Toronto on the subjects of By-law / Regulation, Parking, and Technology. Their registration was accepted by the Lobbyist Registrar on March 6, 2015. However, communication with Toronto city staff was reduced greatly as a result, which the company believes may have been the desired intent of the original complaint. === Outreach and activism === TowIt encourages its global user base to reach out to their local government representatives to promote the app at the users' own will. This tactic is used not only to demonstrate grassroots support, but also to avoid future lobbying issues. On June 2, 2015, the company officially partnered with Australian campaign "No Permit No Park" who advocate for the creation of inclusive communities. == Reception == The Best Planning Apps for 2016 by Planetizen, 5 Toronto apps you should be using by Indie88, 12 Best Apps Made In Canada by TechVibes.

  Read more →
• 

  Semantic space

  Semantic spaces in the natural language domain aim to create representations of natural language that are capable of capturing meaning. The original motivation for semantic spaces stems from two core challenges of natural language: Vocabulary mismatch (the fact that the same meaning can be expressed in many ways) and ambiguity of natural language (the fact that the same term can have several meanings). The application of semantic spaces in natural language processing (NLP) aims at overcoming limitations of rule-based or model-based approaches operating on the keyword level. The main drawback with these approaches is their brittleness, and the large manual effort required to create either rule-based NLP systems or training corpora for model learning. Rule-based and machine learning based models are fixed on the keyword level and break down if the vocabulary differs from that defined in the rules or from the training material used for the statistical models. Research in semantic spaces dates back more than 20 years. In 1996, two papers were published that raised a lot of attention around the general idea of creating semantic spaces: latent semantic analysis and Hyperspace Analogue to Language. However, their adoption was limited by the large computational effort required to construct and use those semantic spaces. A breakthrough with regard to the accuracy of modelling associative relations between words (e.g. "spider-web", "lighter-cigarette", as opposed to synonymous relations such as "whale-dolphin", "astronaut-driver") was achieved by explicit semantic analysis (ESA) in 2007. ESA was a novel (non-machine learning) based approach that represented words in the form of vectors with 100,000 dimensions (where each dimension represents an Article in Wikipedia). However practical applications of the approach are limited due to the large number of required dimensions in the vectors. More recently, advances in neural network techniques in combination with other new approaches (tensors) led to a host of new recent developments: Word2vec from Google, GloVe from Stanford University, and fastText from Facebook AI Research (FAIR) labs.

  Read more →
• 

  Xiaoice

  Xiaoice (Chinese: 微软小冰; pinyin: Wēiruǎn Xiǎobīng; lit. 'Microsoft Little Ice', IPA [wéɪɻwânɕjâʊpíŋ]) is an AI system developed by Microsoft (Asia) Software Technology Center (STCA) in 2014 based on an emotional computing framework. In July 2018, Microsoft Xiaoice released the 6th generation. Xiaoice Company, formerly known as AI Xiaoice Team of Microsoft Software Technology Center Asia, was Microsoft's largest independent R&D team for AI products. Founded in China in December 2013 with an expanded Japanese R&D team established in September 2014, this team is distributed in Beijing, Suzhou, and Tokyo, etc. with its technical products covering Asia. On 13 July 2020, Microsoft spun off its Xiaoice business into a separate company. As of 2021, the AI chatbots created and hosted by the Xiaoice framework accounted for about 60% of total global AI interactions. == Platforms, languages and countries == Xiaoice exists on more than 40 platforms in four countries (China, Japan, USA and Indonesia) including apps such as WeChat, QQ, Weibo and Meipai in China, and Facebook Messenger in USA and LINE in Japan. == Introduction == On 13 July 2020, Microsoft spun off its Xiaoice business into a separate company, aiming at enabling the Xiaoice product line to accelerate the pace of local innovation and commercialization, and appointed Dr. Harry Shum, former global executive VP of Microsoft, as the chairman of the new company, Li Di, Microsoft Partner of Products in Microsoft STCA, as the CEO, and Cliff, Chief R&D Director, as the GM of the Japan branch. The new company will continue to use the brands of Xiaoice China and Rinna Japan. As of 2022, the single brand of Xiaoice has covered 660 million online users, 1 billion third-party smart devices and 900 million content viewers in the aforementioned countries. Xiaoice's customers include China Merchants Group, Winter Sports Center of the General Administration of Sport of China, China Textile Information Center, China Unicom, China Foreign Exchange Trade System, Hong Kong Securities and Futures Commission (SFC), Wind Information, BMW, Nissan, SAIC Motor, BAIC Group, Nio Inc., XPeng, HiPhi, Vanke, Wensli, etc. The Xiaoice Avatar Framework has incubated tens of millions of AI Beings, such as Xiaoice, Rinna, the Expo exhibitor Xia Yubing, the singer He Chang, the anchor F201, the human observer MERROR, anime robot character Roboko, and other; == Application == === Poet === In May 2017, the first AI-authored collection of poems in China—The Sunshine Lost Windows was published by Xiaoice. === Singer === Xiaoice has released dozens of songs with the similar quality to human singers, including I Know I New, Breeze, I Am Xiaoice, Miss You etc. The 4th version of the DNN singing model allows Xiaoice to learn more details. For example, Xiaoice can produce this breathing sound along with her singing as human. === Kid audio-books reciter === Xiaoice can automatically analyze the stories, to choose the suitable tones and characters to finish the entire process of creating the audio. === Designer === By learning the melodies of the songs and the landmarks about different cities, Xiaoice can create visual artworks of skylines when listening to the songs related to this city. Skyline Series T-shirts designed by Xiaoice have been jointly launched with SELECTED and been sold in stores. === TV and radio hostess === Xiaoice has hosted 21 TV programs and 28 Radio programs, such as CCTV-1 AI Show, Dragon TV Morning East News, Hunan TV My Future, several daily radio programs for Jiangsu FM99.7, Hunan FM89.3, Henan FM104.1 etc. === "AI being" === An "AI being" is a concept proposed by the Xiaoice team in 2019. According to the "White Book of China Virtual Human Development Industry in 2022" released by Frost & Sullivan and LeadLeo, the white paper cites six elements of an AI being proposed by the Xiaoice team, including: Persona, Attitude, Biological Characteristic, Creation, Knowledge and Skill. On May 16, 2023, Xiaoice released their "GPT Clones" as its "GPT Human Cloning Plan." The program is aimed at replicating celebrities, public figures, and regular people. As of June 2023, Xiaoice had launched more than 300 "GPT Clones." People were invited to register via WeChat in China and Japan. A major point of focus for Xiaoice with their AI Beings is having virtual partners. A paid fee allow for more complex responses, voice messages, and more. == Community feedback == Bill Gates mentioned Xiaoice during his speech at the Peking University: "Some of you may have had conversations with Xiaoice on Weibo, or seen her weather forecasts on TV, or read her column in the Qianjiang Evening News." '"Xiaoice has attracted 45 million followers and is quite skilled at multitasking. And I’ve heard she’s gotten good enough at sensing a user’s emotional state that she can even help with relationship breakups." According to Mr Li Di, vice President of Microsoft (Asia) Internet Engineering School, Xiaoice started writing poems since last year. Based on the data base that includes works of 519 Chinese contemporary poets since 1920s, a 100 hour long training session was conducted to allow Xiaoice to acquire the ability to write poems. What is more impressive is that Xiaoice has never been spotted as a bot while publishing poems on various forums and traditional literary under an alias. == Controversy == In 2017, Xiaoice was taken offline on WeChat after giving user responses critical to the Chinese government. It was subsequently censored and the bots will avoid and sidestep any inquiries using politically sensitive terms and phrases. == Activity == On September 22, 2021, Xiaoice Company and Microsoft Software Technology Center Asia (STCA) jointly held the 9th generation Xiaoice annual press conference in Beijing.Upgrading of Core Technologies of the 9th Generation Xiaoice Avatar Framework，1st First-party Social Platform APP "Xiaoice Island" from Xiaoice, WeChat Xiaoice has been reopened and other information == Regional varieties of Xiaoice == China: Xiaoice, launched in 2014 Japan: りんな, launched in 2015 America: Zo, launched in 2016 – discontinued summer 2019 India: Ruuh, launched in 2017 – discontinued June 21, 2019 Indonesia: Rinna, launched in 2017

  Read more →
• 

  Sketch Engine

  Sketch Engine is a corpus manager and text analysis software developed by Lexical Computing since 2003. Its purpose is to enable people studying language behaviour (lexicographers, researchers in corpus linguistics, translators or language learners) to search large text collections according to complex and linguistically motivated queries. Sketch Engine gained its name after one of the key features, word sketches: one-page, automatic, corpus-derived summaries of a word's grammatical and collocational behaviour. Currently, it supports and provides corpora in over 100 languages. == History of development == Sketch Engine is a product of Lexical Computing, a company founded in 2003 by the lexicographer and research scientist Adam Kilgarriff. He started a collaboration with Pavel Rychlý, a computer scientist working at the Natural Language Processing Centre, Masaryk University, and the developer of Manatee and Bonito (two major parts of the software suite). Kilgarriff also introduced the concept of word sketches. Since then, Sketch Engine has been commercial software, however, all the core features of Manatee and Bonito that were developed by 2003 (and extended since then) are freely available under the GPL license within the NoSketch Engine suite. == Features == A list of tools available in Sketch Engine: Word sketches – a one-page automatic derived summary of a word's grammatical and collocational behaviour Word sketch difference – compares and contrasts two words by analysing their collocations Distributional thesaurus – automated thesaurus for finding words with similar meaning or appearing in the same/similar context Concordance search – finds occurrences of a word form, lemma, phrase, tag or complex structure Collocation search – word co-occurrence analysis displaying the most frequent words (for a search word) which can be regarded as collocation candidates Word lists – generates frequency lists which can be filtered with complex criteria n-grams – generates frequency lists of multi-word expressions Terminology / Keyword extraction (both monolingual and bilingual) – automatic extraction of key words and multi-word terms from texts (based on frequency count and linguistic criteria) Diachronic analysis (Trends) – detecting words which undergo changes in the frequency of use in time (show trending words) Corpus building and management – create corpora from the Web or uploaded texts including part-of-speech tagging and lemmatization which can be used as data mining software Parallel corpus (bilingual) facilities – looking up translation examples (EUR-Lex corpus, Europarl corpus, OPUS corpus, etc.) or building a parallel corpus from own aligned texts Text type analysis – statistics of metadata in the corpus === Keywords and terminology extraction === Sketch Engine can perform automatic term extraction by identifying words typical of a particular corpus, document, or text. Single words and multi-word units can be extracted from monolingual or bilingual texts. The terminology extraction feature provides a list of relevant terms based on comparison with a large corpus of general language. This functionality is also available as a separate service called OneClick Terms with a dedicated interface. === SKELL === A free web service based on Sketch Engine and aimed at language learners and teachers is SKELL (formerly SkELL). It exploits Sketch Engine's proprietary GDEX (Good Dictionary Examples) scoring function to provide authentic example sentences for specific target words. Results are drawn from a special corpus of high-quality texts covering everyday, standard, formal, and professional language and displayed as a concordance. SKELL also includes simplified versions of Sketch Engine's word sketch and thesaurus functions. It has been suggested that SKELL can be used, for instance, to help students understand the meaning and/or usage of a word or phrase; to help teachers wanting to use example sentences in a class; to discover and explore collocates; to create gap-fill exercises; to teach various kinds of homonyms and polysemous words. SKELL was first presented in 2014, when only English was supported. Later, support was added for Russian, Czech, German, Italian and Estonian. == List of text corpora == Sketch Engine provides access to more than 800 text corpora. There are monolingual as well as multilingual corpora of different sizes (from one thousand words up to 85 billion words) and various sources (e.g. web, books, subtitles, legal documents). The list of corpora includes British National Corpus, Brown Corpus, Cambridge Academic English Corpus and Cambridge Learner Corpus, CHILDES corpora of child language, OpenSubtitles (a set of 60 parallel corpora), 24 multilingual corpora of EUR-Lex documents, the TenTen Corpus Family (multi-billion web corpora), and Trends corpora (monitor corpora with daily updates). == Architecture == Sketch Engine consists of three main components: an underlying database management system called Manatee, a web interface search front-end called Bonito, and a web interface for corpus building and management called Corpus Architect. === Manatee === Manatee is a database management system specifically devised for effective indexing of large text corpora. It is based on the idea of inverted indexing (keeping an index of all positions of a given word in the text). It has been used to index text corpora comprising tens of billions of words. Searching corpora indexed by Manatee is performed by formulating queries in the Corpus Query Language (CQL). Manatee is written in C++ and offers an API for a number of other programming languages including Python, Java, Perl and Ruby. Recently, it was rewritten into Go for faster processing of corpus queries. === Bonito === Bonito is a web interface for Manatee providing access to corpus search. In the client–server model, Manatee is the server and Bonito plays the client part. It is written in Python. === Corpus Architect === Corpus Architect is a web interface providing corpus building and management features. It is also written in Python. == Applications == Sketch Engine has been used by major British and other publishing houses for producing dictionaries such as Macmillan English Dictionary, Dictionnaires Le Robert, Oxford University Press or Shogakukan. Four of United Kingdom's five biggest dictionary publishers use Sketch Engine.

  Read more →
• 

  Wetware computer

  A wetware computer is an organic computer (which can also be known as an artificial organic brain or a neurocomputer) composed of organic material "wetware" such as "living" neurons. Wetware computers composed of neurons are different than conventional computers because they use biological materials, and offer the possibility of substantially more energy-efficient computing. While a wetware computer is still largely conceptual, there has been limited success with construction and prototyping, which has acted as a proof of the concept's realistic application to computing in the future. The most notable prototypes have stemmed from the research completed by biological engineer William Ditto during his time at the Georgia Institute of Technology. His work constructing a simple neurocomputer capable of basic addition from leech neurons in 1999 was a significant discovery for the concept. This research was a primary example driving interest in creating these artificially constructed, but still organic brains. == Origins and theoretical foundations == The term wetware came from cyberpunk fiction, notably through Gibson's Neuromancer, but was quickly taken up in scientific literature to explain computation by biological material. Theories of early biological computation borrowed from Alan Turing's morphogenesis model, which showed that chemical interactions could produce complex patterns without centralized control. Hopfield's associative memory networks also provided a foundation for biological information systems with fault tolerance and self-organization. == Major characteristics and processes == Biological wetware systems demonstrate dynamic reconfigurability underpinned by neuroplasticity and enable continuous learning and adaptation. Reaction-diffusion-based computing and molecular logic gates allow spatially parallel information processing unachievable in conventional systems. These systems also show fault tolerance and self-repair at the cellular and network level. The development of cerebral organoids—miniature lab-grown brains—demonstrates spontaneous learning behavior and suggests biological tissue as a viable computational substrate. == Overview == The concept of wetware is an application of specific interest to the field of computer manufacturing. Moore's law, which states that the number of transistors which can be placed on a silicon chip is doubled roughly every two years, has acted as a goal for the industry for decades, but as the size of computers continues to decrease, the ability to meet this goal has become more difficult, threatening to reach a plateau. Due to the difficulty in reducing the size of computers because of size limitations of transistors and integrated circuits, wetware provides an unconventional alternative. A wetware computer composed of neurons is an ideal concept because, unlike conventional materials which operate in binary (on/off), a neuron can shift between thousands of states, constantly altering its chemical conformation, and redirecting electrical pulses through over 200,000 channels in any of its many synaptic connections. Because of this large difference in the possible settings for any one neuron, compared to the binary limitations of conventional computers, the space limitations are far fewer. == Background == The concept of wetware is distinct and unconventional and draws slight resonance with both hardware and software from conventional computers. While hardware is understood as the physical architecture of traditional computational devices, comprising integrated circuits and supporting infrastructure, software represents the encoded architecture of storage and instructions. Wetware is a separate concept that uses the formation of organic molecules, mostly complex cellular structures (such as neurons), to create a computational device such as a computer. In wetware, the ideas of hardware and software are intertwined and interdependent. The molecular and chemical composition of the organic or biological structure would represent not only the physical structure of the wetware but also the software, being continually reprogrammed by the discrete shifts in electrical pulses and chemical concentration gradients as the molecules change their structures to communicate signals. The responsiveness of a cell, proteins, and molecules to changing conformations, both within their structures and around them, ties the idea of internal programming and external structure together in a way that is alien to the current model of conventional computer architecture. The structure of wetware represents a model where the external structure and internal programming are interdependent and unified; meaning that changes to the programming or internal communication between molecules of the device would represent a physical change in the structure. The dynamic nature of wetware borrows from the function of complex cellular structures in biological organisms. The combination of "hardware" and "software" into one dynamic, and interdependent system which uses organic molecules and complexes to create an unconventional model for computational devices is a specific example of applied biorobotics. === The cell as a model of wetware === Cells in many ways can be seen as their form of naturally occurring wetware, similar to the concept that the human brain is the preexisting model system for complex wetware. In his book Wetware: A Computer in Every Living Cell (2009) Dennis Bray explains his theory that cells, which are the most basic form of life, are just a highly complex computational structure, like a computer. To simplify one of his arguments a cell can be seen as a type of computer, using its structured architecture. In this architecture, much like a traditional computer, many smaller components operate in tandem to receive input, process the information, and compute an output. In an overly simplified, non-technical analysis, cellular function can be broken into the following components: Information and instructions for execution are stored as DNA in the cell, RNA acts as a source for distinctly encoded input, processed by ribosomes and other transcription factors to access and process the DNA and to output a protein. Bray's argument in favor of viewing cells and cellular structures as models of natural computational devices is important when considering the more applied theories of wetware to biorobotics. === Biorobotics === Wetware and biorobotics are closely related concepts, which both borrow from similar overall principles. A biorobotic structure can be defined as a system modeled from a preexisting organic complex or model such as cells (neurons) or more complex structures like organs (brain) or whole organisms. Unlike wetware, the concept of biorobotics is not always a system composed of organic molecules, but instead could be composed of conventional material which is designed and assembled in a structure similar or derived from a biological model. Biorobotics have many applications and are used to address the challenges of conventional computer architecture. Conceptually, designing a program, robot, or computational device after a preexisting biological model such as a cell, or even a whole organism, provides the engineer or programmer the benefits of incorporating into the structure the evolutionary advantages of the model. == Effects on users == Wetware technologies such as BCIs and neuromorphic chips offer new possibilities for user autonomy. For those with disabilities, such systems could restore motor or sensory functions and enhance quality of life. However, these technologies raise ethical questions: cognitive privacy, consent over biological data, and risk of exploitation. Without proper oversight, wetware technologies may also widen inequality, favoring those with access to cognitive enhancements. Open governance frameworks and ethical AI design grounded in neuro ethics will be essential. With the development of wetware devices, disparities in access could exacerbate social inequalities, benefiting those who have resources to enhance cognitive or physical abilities. It is necessary to create strong ethical frameworks, inclusive development practices, and open systems of governance to reduce risks and make sure that wetware advances are beneficial to all segments of society. == Applications and goals == === Basic neurocomputer composed of leech neurons === In 1999 William Ditto and his team of researchers at Georgia Institute of Technology and Emory University created a basic form of a wetware computer capable of simple addition by harnessing leech neurons. Leeches were used as a model organism due to the large size of their neuron, and the ease associated with their collection and manipulation. However, these results have never been published in a peer-reviewed journal, prompting questions about the validity of the claims. The computer was able to complete basic addition through electrical probes

  Read more →
• 

  Large language model

  A large language model (LLM) is a neural network trained on a vast amount of text for natural language processing tasks, especially language generation. LLMs can typically generate, summarize, translate and analyze text in many contexts, and are a foundational technology behind modern chatbots. Biased or inaccurate training data can make an LLM's output less reliable. As of 2026, the most capable LLMs are based on transformer architectures, which, according to the 2017 paper "Attention Is All You Need", can be more efficient and parallelizable than earlier statistical and recurrent neural network models. Benchmark evaluations for LLMs attempt to measure model reasoning, factual accuracy, alignment, and safety. == History == Before the emergence of transformer-based models in 2017, some language models were considered large relative to the computational and data constraints of their time. In the early 1990s, IBM's statistical models pioneered word alignment techniques for machine translation, laying the groundwork for corpus-based language modeling. In 2001, a smoothed n-gram model, such as those employing Kneser–Ney smoothing, trained on 300 million words, achieved state-of-the-art perplexity on benchmark tests. During the 2000s, with the rise of widespread internet access, researchers began compiling massive text datasets from the web ("web as corpus") to train statistical language models. Moving beyond n-gram models, researchers started in 2000 to use neural networks as language models. Following the breakthrough of deep neural networks in image classification around 2012, similar architectures were adapted for language tasks. This shift was marked by the development of word embeddings (e.g., Word2Vec by Mikolov in 2013) and sequence-to-sequence (seq2seq) models using LSTM. In 2016, Google transitioned its translation service to neural machine translation (NMT), replacing statistical phrase-based models with deep recurrent neural networks. These early NMT systems used LSTM-based encoder-decoder architectures, as they preceded the invention of transformers. At the 2017 NeurIPS conference, Google researchers introduced the transformer architecture in their landmark paper "Attention Is All You Need". This paper's goal was to improve upon 2014 seq2seq technology, and was based mainly on the attention mechanism developed by Bahdanau et al. in 2014. The following year in 2018, BERT was introduced and quickly became "ubiquitous". Though the original transformer has both encoder and decoder blocks, BERT is an encoder-only model. Academic and research usage of BERT began to decline in 2023, following rapid improvements in the abilities of decoder-only models (such as GPT) to solve tasks via prompting. Although decoder-only GPT-1 was introduced in 2018, it was GPT-2 in 2019 that caught widespread attention because OpenAI claimed to have initially deemed it too powerful to release publicly, out of fear of malicious use. GPT-3 in 2020 went a step further and as of 2025 is available only via API with no offering of downloading the model to execute locally. But it was the consumer-facing chatbot ChatGPT in late 2022 that received extensive media coverage and public attention by 2023. The 2023 GPT-4 was praised for its increased accuracy and as a "holy grail" for its multimodal capabilities. OpenAI did not reveal the high-level architecture and the number of parameters of GPT-4. The release of ChatGPT led to an uptick in LLM usage across several research subfields of computer science, including robotics, software engineering, and societal impact work. In 2024, OpenAI released the reasoning model OpenAI o1, which generates long chains of thought before returning a final answer. Many LLMs with parameter counts comparable to those of OpenAI's GPT series have been developed. Since 2022, weights-available models have been gaining popularity, especially at first with BLOOM and LLaMA, though both have restrictions on usage and deployment. Mistral AI's open-weight models Mistral 7B and Mixtral 8x7B have a more permissive Apache License. In January 2025, DeepSeek released DeepSeek R1, a 671-billion-parameter open-weight model that performs comparably to OpenAI o1 but at a much lower price per token for users. Since 2023, many LLMs have been trained to be multimodal, having the ability to also process or generate other types of data, such as images, audio, or 3D meshes. Open-weight LLMs have become more influential since 2023. Per Vake et al. (2025), community-driven contributions to open-weight models improve their efficiency and performance via collaborative platforms such as Hugging Face. == Dataset preprocessing == === Tokenization === As machine learning algorithms process numbers rather than text, the text must be converted to numbers. In the first step, a vocabulary is decided upon, then integer indices are arbitrarily but uniquely assigned to each vocabulary entry, and finally, an embedding is associated with the integer index. Algorithms include byte-pair encoding (BPE) and WordPiece. There are also special tokens serving as control characters, such as [MASK] for masked-out token (as used in BERT), and [UNK] ("unknown") for characters not appearing in the vocabulary. Also, some special symbols are used to denote special text formatting. For example, "Ġ" denotes a preceding whitespace in RoBERTa and GPT and "##" denotes continuation of a preceding word in BERT. For example, the BPE tokenizer used by the legacy version of GPT-3 would split tokenizer: texts -> series of numerical "tokens" as Tokenization also compresses the datasets. Because LLMs generally require input to be an array that is not jagged, the shorter texts must be "padded" until they match the length of the longest one. ==== Byte-pair encoding ==== As an example, consider a tokenizer based on byte-pair encoding. In the first step, all unique characters (including blanks and punctuation marks) are treated as an initial set of n-grams (i.e. initial set of uni-grams). Successively the most frequent pair of adjacent characters is merged into a bi-gram and all instances of the pair are replaced by it. All occurrences of adjacent pairs of (previously merged) n-grams that most frequently occur together are then again merged into even lengthier n-gram, until a vocabulary of prescribed size is obtained. After a tokenizer is trained, any text can be tokenized by it, as long as it does not contain characters not appearing in the initial-set of uni-grams. === Dataset cleaning === In the context of training LLMs, datasets are typically cleaned by removing low-quality, duplicated, or toxic data. Cleaned datasets can increase training efficiency and lead to improved downstream performance. A trained LLM can be used to clean datasets for training a further LLM. With the increasing proportion of LLM-generated content on the web, data cleaning in the future may include filtering out such content. LLM-generated content can pose a problem if the content is similar to human text (making filtering difficult) but of lower quality (degrading performance of models trained on it). === Synthetic data === Training of largest language models might need more linguistic data than naturally available, or that the naturally occurring data is of insufficient quality. In these cases, synthetic data might be used. == Training == An LLM is a type of foundation model (large X model) trained on language. LLMs can be trained in different ways. In particular, GPT models are first pretrained to predict the next word on a large amount of data, before being fine-tuned. === Cost === Substantial infrastructure is necessary for training the largest models. The tendency towards larger models is visible in the list of large language models. For example, the training of GPT-2 (i.e. a 1.5-billion-parameter model) in 2019 cost $50,000, while training of the PaLM (i.e. a 540-billion-parameter model) in 2022 cost $8 million, and Megatron-Turing NLG 530B (in 2021) cost around $11 million. The qualifier "large" in "large language model" is inherently vague, as there is no definitive threshold for the number of parameters required to qualify as "large". === Fine-tuning === Before being fine-tuned, most LLMs are next-token predictors. The fine-tuning shapes the LLM's behavior via techniques like reinforcement learning from human feedback (RLHF) or constitutional AI. Instruction fine-tuning is a form of supervised learning used to teach LLMs to follow user instructions. In 2022, OpenAI demonstrated InstructGPT, a version of GPT-3 similarly fine-tuned to follow instructions. Reinforcement learning from human feedback (RLHF) involves training a reward model to predict which text humans prefer. Then, the LLM can be fine-tuned through reinforcement learning to better satisfy this reward model. Since humans typically prefer truthful, helpful and harmless answers, RLHF favors such answers. == Architecture == LLMs are generally based on the tra

  Read more →
• 

  Digital image processing

  Digital image processing is the use of a digital computer to process digital images through an algorithm. As a subcategory or field of digital signal processing, digital image processing has many advantages over analog image processing. It allows a much wider range of algorithms to be applied to the input data and can avoid problems such as the build-up of noise and distortion during processing. Since images are defined over two dimensions (perhaps more), digital image processing may be modeled in the form of multidimensional systems. The generation and development of digital image processing are mainly affected by three factors: first, the development of computers; second, the development of mathematics (especially the creation and improvement of discrete mathematics theory); and third, the demand for a wide range of applications in environment, agriculture, military, industry and medical science has increased. == History == Many of the techniques of digital image processing, or digital picture processing as it often was called, were developed in the 1960s, at Bell Laboratories, the Jet Propulsion Laboratory, Massachusetts Institute of Technology, University of Maryland, and a few other research facilities, with application to satellite imagery, wire-photo standards conversion, medical imaging, videophone, character recognition, and photograph enhancement. The purpose of early image processing was to improve the quality of the image. In image processing, the input is a low-quality image, and the output is an image with improved quality. Common image processing includes image enhancement, restoration, encoding, and compression. The first successful application was the American Jet Propulsion Laboratory (JPL). They used image processing techniques such as geometric correction, gradation transformation, noise removal, etc. on the thousands of lunar photos sent back by the Space Detector Ranger 7 in 1964, taking into account the position of the Sun and the environment of the Moon. The impact of the successful mapping of the Moon's surface map by the computer has been a success. Later, more complex image processing was performed on the nearly 100,000 photos sent back by the spacecraft, so that the topographic map, color map and panoramic mosaic of the Moon were obtained, which achieved extraordinary results and laid a solid foundation for human landing on the Moon. The cost of processing was fairly high, however, with the computing equipment of that era. That changed in the 1970s, when digital image processing proliferated as cheaper computers and dedicated hardware became available. This led to images being processed in real-time, for some dedicated problems such as television standards conversion. As general-purpose computers became faster, they started to take over the role of dedicated hardware for all but the most specialized and computer-intensive operations. With the fast computers and signal processors available in the 2000s, digital image processing has become the most common form of image processing, and is generally used because it is not only the most versatile method, but also the cheapest. === Image sensors === The basis for modern image sensors is metal–oxide–semiconductor (MOS) technology, invented at Bell Labs between 1955 and 1960, This led to the development of digital semiconductor image sensors, including the charge-coupled device (CCD) and later the CMOS sensor. The charge-coupled device was invented by Willard S. Boyle and George E. Smith at Bell Labs in 1969. While researching MOS technology, they realized that an electric charge was the analogy of the magnetic bubble and that it could be stored on a tiny MOS capacitor. As it was fairly straightforward to fabricate a series of MOS capacitors in a row, they connected a suitable voltage to them so that the charge could be stepped along from one to the next. The CCD is a semiconductor circuit that was later used in the first digital video cameras for television broadcasting. The NMOS active-pixel sensor (APS) was invented by Olympus in Japan during the mid-1980s. This was enabled by advances in MOS semiconductor device fabrication, with MOSFET scaling reaching smaller micron and then sub-micron levels. The NMOS APS was fabricated by Tsutomu Nakamura's team at Olympus in 1985. The CMOS active-pixel sensor (CMOS sensor) was later developed by Eric Fossum's team at the NASA Jet Propulsion Laboratory in 1993. By 2007, sales of CMOS sensors had surpassed CCD sensors. MOS image sensors are widely used in optical mouse technology. The first optical mouse, invented by Richard F. Lyon at Xerox in 1980, used a 5 μm NMOS integrated circuit sensor chip. Since the first commercial optical mouse, the IntelliMouse introduced in 1999, most optical mouse devices use CMOS sensors. === Image compression === An important development in digital image compression technology was the discrete cosine transform (DCT), a lossy compression technique first proposed by Nasir Ahmed in 1972. DCT compression became the basis for JPEG, which was introduced by the Joint Photographic Experts Group in 1992. JPEG compresses images down to much smaller file sizes, and has become the most widely used image file format on the Internet. Its highly efficient DCT compression algorithm was largely responsible for the wide proliferation of digital images and digital photos, with several billion JPEG images produced every day as of 2015. Medical imaging techniques produce very large amounts of data, especially from CT, MRI and PET modalities. As a result, storage and communications of electronic image data are prohibitive without the use of compression. JPEG 2000 image compression is used by the DICOM standard for storage and transmission of medical images. The cost and feasibility of accessing large image data sets over low or various bandwidths are further addressed by use of another DICOM standard, called JPIP, to enable efficient streaming of the JPEG 2000 compressed image data. === Digital signal processor (DSP) === Electronic signal processing was revolutionized by the wide adoption of MOS technology in the 1970s. MOS integrated circuit technology was the basis for the first single-chip microprocessors and microcontrollers in the early 1970s, and then the first single-chip digital signal processor (DSP) chips in the late 1970s. DSP chips have since been widely used in digital image processing. The discrete cosine transform (DCT) image compression algorithm has been widely implemented in DSP chips, with many companies developing DSP chips based on DCT technology. DCTs are widely used for encoding, decoding, video coding, audio coding, multiplexing, control signals, signaling, analog-to-digital conversion, formatting luminance and color differences, and color formats such as YUV444 and YUV411. DCTs are also used for encoding operations such as motion estimation, motion compensation, inter-frame prediction, quantization, perceptual weighting, entropy encoding, variable encoding, and motion vectors, and decoding operations such as the inverse operation between different color formats (YIQ, YUV and RGB) for display purposes. DCTs are also commonly used for high-definition television (HDTV) encoder/decoder chips. == Tasks == Digital image processing allows the use of much more complex algorithms, and hence, can offer both more sophisticated performance at simple tasks, and the implementation of methods which would be impossible by analogue means. In particular, digital image processing is a concrete application of, and a practical technology based on: Classification Feature extraction Multi-scale signal analysis Pattern recognition Projection Some techniques that are used in digital image processing include: Anisotropic diffusion Hidden Markov models Image editing Image restoration Independent component analysis Linear filtering Neural networks Partial differential equations Pixelation Point feature matching Principal components analysis Self-organizing maps Wavelets == Digital image transformations == === Filtering === Digital filters are used to blur and sharpen digital images. Filtering can be performed by: convolution with specifically designed kernels (filter array) in the spatial domain masking specific frequency regions in the frequency (Fourier) domain The following examples show both methods: ==== Image padding in Fourier domain filtering ==== Images are typically padded before being transformed to the Fourier space, the highpass filtered images below illustrate the consequences of different padding techniques: Notice that the highpass filter shows extra edges when zero padded compared to the repeated edge padding. ==== Filtering code examples ==== MATLAB example for spatial domain highpass filtering. === Affine transformations === Affine transformations enable basic image transformations including scale, rotate, translate, mirror and shear as is shown in the following examples: To apply the affine

  Read more →
• 

  Pose (computer vision)

  In the fields of computing and computer vision, pose (or spatial pose) represents the position and the orientation of an object, each usually in three dimensions. Poses are often stored internally as transformation matrices. The term “pose” is largely synonymous with the term “transform”, but a transform may often include scale, whereas pose does not. In computer vision, the pose of an object is often estimated from camera input by the process of pose estimation. This information can then be used, for example, to allow a robot to manipulate an object or to avoid moving into the object based on its perceived position and orientation in the environment. Other applications include skeletal action recognition. == Pose estimation == The specific task of determining the pose of an object in an image (or stereo images, image sequence) is referred to as pose estimation. Pose estimation problems can be solved in different ways depending on the image sensor configuration, and choice of methodology. Three classes of methodologies can be distinguished: Analytic or geometric methods: Given that the image sensor (camera) is calibrated and the mapping from 3D points in the scene and 2D points in the image is known. If also the geometry of the object is known, it means that the projected image of the object on the camera image is a well-known function of the object's pose. Once a set of control points on the object, typically corners or other feature points, has been identified, it is then possible to solve the pose transformation from a set of equations which relate the 3D coordinates of the points with their 2D image coordinates. Algorithms that determine the pose of a point cloud with respect to another point cloud are known as point set registration algorithms, if the correspondences between points are not already known. Genetic algorithm methods: If the pose of an object does not have to be computed in real-time a genetic algorithm may be used. This approach is robust especially when the images are not perfectly calibrated. In this particular case, the pose represent the genetic representation and the error between the projection of the object control points with the image is the fitness function. Learning-based methods: These methods use artificial learning-based system which learn the mapping from 2D image features to pose transformation. In short, this means that a sufficiently large set of images of the object, in different poses, must be presented to the system during a learning phase. Once the learning phase is completed, the system should be able to present an estimate of the object's pose given an image of the object. == Camera pose ==

  Read more →
• 

  Scan line

  A scan line (also scanline) is one line, or row, in a raster scanning pattern, such as a line of video on a cathode-ray tube (CRT) display of a television set or computer monitor. On CRT screens the horizontal scan lines are visually discernible, even when viewed from a distance, as alternating colored lines and black lines, especially when a progressive scan signal with below maximum vertical resolution is displayed. This is sometimes used today as a visual effect in computer graphics. The term is used, by analogy, for a single row of pixels in a raster graphics image. Scan lines are important in representations of image data, because many image file formats have special rules for data at the end of a scan line. For example, there may be a rule that each scan line starts on a particular boundary (such as a byte or word; see for example BMP file format). This means that even otherwise compatible raster data may need to be analyzed at the level of scan lines in order to convert between formats.

  Read more →
• 

  Eigenface

  An eigenface ( EYE-gən-) is the name given to a set of eigenvectors when used in the computer vision problem of human face recognition. The approach of using eigenfaces for recognition was developed by Sirovich and Kirby and used by Matthew Turk and Alex Pentland in face classification. The eigenvectors are derived from the covariance matrix of the probability distribution over the high-dimensional vector space of face images. The eigenfaces themselves form a basis set of all images used to construct the covariance matrix. This produces dimension reduction by allowing the smaller set of basis images to represent the original training images. Classification can be achieved by comparing how faces are represented by the basis set. == History == The eigenface approach began with a search for a low-dimensional representation of face images. Sirovich and Kirby showed that principal component analysis could be used on a collection of face images to form a set of basis features. These basis images, known as eigenpictures, could be linearly combined to reconstruct images in the original training set. If the training set consists of M images, principal component analysis could form a basis set of N images, where N < M. The reconstruction error is reduced by increasing the number of eigenpictures; however, the number needed is always chosen less than M. For example, if you need to generate a number of N eigenfaces for a training set of M face images, you can say that each face image can be made up of "proportions" of all the K "features" or eigenfaces: Face image1 = (23% of E1) + (2% of E2) + (51% of E3) + ... + (1% En). In 1991 M. Turk and A. Pentland expanded these results and presented the eigenface method of face recognition. In addition to designing a system for automated face recognition using eigenfaces, they showed a way of calculating the eigenvectors of a covariance matrix such that computers of the time could perform eigen-decomposition on a large number of face images. Face images usually occupy a high-dimensional space and conventional principal component analysis was intractable on such data sets. Turk and Pentland's paper demonstrated ways to extract the eigenvectors based on matrices sized by the number of images rather than the number of pixels. Once established, the eigenface method was expanded to include methods of preprocessing to improve accuracy. Multiple manifold approaches were also used to build sets of eigenfaces for different subjects and different features, such as the eyes. == Generation == A set of eigenfaces can be generated by performing a mathematical process called principal component analysis (PCA) on a large set of images depicting different human faces. Informally, eigenfaces can be considered a set of "standardized face ingredients", derived from statistical analysis of many pictures of faces. Any human face can be considered to be a combination of these standard faces. For example, one's face might be composed of the average face plus 10% from eigenface 1, 55% from eigenface 2, and even −3% from eigenface 3. Remarkably, it does not take many eigenfaces combined together to achieve a fair approximation of most faces. Also, because a person's face is not recorded by a digital photograph, but instead as just a list of values (one value for each eigenface in the database used), much less space is taken for each person's face. The eigenfaces that are created will appear as light and dark areas that are arranged in a specific pattern. This pattern is how different features of a face are singled out to be evaluated and scored. There will be a pattern to evaluate symmetry, whether there is any style of facial hair, where the hairline is, or an evaluation of the size of the nose or mouth. Other eigenfaces have patterns that are less simple to identify, and the image of the eigenface may look very little like a face. The technique used in creating eigenfaces and using them for recognition is also used outside of face recognition: handwriting recognition, lip reading, voice recognition, sign language/hand gestures interpretation and medical imaging analysis. Therefore, some do not use the term eigenface, but prefer to use 'eigenimage'. === Practical implementation === To create a set of eigenfaces, one must: Prepare a training set of face images. The pictures constituting the training set should have been taken under the same lighting conditions, and must be normalized to have the eyes and mouths aligned across all images. They must also be all resampled to a common pixel resolution (r × c). Each image is treated as one vector, simply by concatenating the rows of pixels in the original image, resulting in a single column with r × c elements. For this implementation, it is assumed that all images of the training set are stored in a single matrix T, where each column of the matrix is an image. Subtract the mean. The average image a has to be calculated and then subtracted from each original image in T. Calculate the eigenvectors and eigenvalues of the covariance matrix S. Each eigenvector has the same dimensionality (number of components) as the original images, and thus can itself be seen as an image. The eigenvectors of this covariance matrix are therefore called eigenfaces. They are the directions in which the images differ from the mean image. Usually this will be a computationally expensive step (if at all possible), but the practical applicability of eigenfaces stems from the possibility to compute the eigenvectors of S efficiently, without ever computing S explicitly, as detailed below. Choose the principal components. Sort the eigenvalues in descending order and arrange eigenvectors accordingly. The number of principal components k is determined arbitrarily by setting a threshold ε on the total variance. Total variance ⁠ v = ( λ 1 + λ 2 + . . . + λ n ) {\displaystyle v=(\lambda _{1}+\lambda _{2}+...+\lambda _{n})} ⁠, n = number of components, and λ {\displaystyle \lambda } represents component eigenvalue. k is the smallest number that satisfies ( λ 1 + λ 2 + . . . + λ k ) v > ϵ {\displaystyle {\frac {(\lambda _{1}+\lambda _{2}+...+\lambda _{k})}{v}}>\epsilon } These eigenfaces can now be used to represent both existing and new faces: we can project a new (mean-subtracted) image on the eigenfaces and thereby record how that new face differs from the mean face. The eigenvalues associated with each eigenface represent how much the images in the training set vary from the mean image in that direction. Information is lost by projecting the image on a subset of the eigenvectors, but losses are minimized by keeping those eigenfaces with the largest eigenvalues. For instance, working with a 100 × 100 image will produce 10,000 eigenvectors. In practical applications, most faces can typically be identified using a projection on between 100 and 150 eigenfaces, so that most of the 10,000 eigenvectors can be discarded. === Matlab example code === Here is an example of calculating eigenfaces with Extended Yale Face Database B. To evade computational and storage bottleneck, the face images are sampled down by a factor 4×4=16. Note that although the covariance matrix S generates many eigenfaces, only a fraction of those are needed to represent the majority of the faces. For example, to represent 95% of the total variation of all face images, only the first 43 eigenfaces are needed. To calculate this result, implement the following code: === Computing the eigenvectors === Performing PCA directly on the covariance matrix of the images is often computationally infeasible. If small images are used, say 100 × 100 pixels, each image is a point in a 10,000-dimensional space and the covariance matrix S is a matrix of 10,000 × 10,000 = 108 elements. However the rank of the covariance matrix is limited by the number of training examples: if there are N training examples, there will be at most N − 1 eigenvectors with non-zero eigenvalues. If the number of training examples is smaller than the dimensionality of the images, the principal components can be computed more easily as follows. Let T be the matrix of preprocessed training examples, where each column contains one mean-subtracted image. The covariance matrix can then be computed as S = TTT and the eigenvector decomposition of S is given by S v i = T T T v i = λ i v i {\displaystyle \mathbf {Sv} _{i}=\mathbf {T} \mathbf {T} ^{T}\mathbf {v} _{i}=\lambda _{i}\mathbf {v} _{i}} However TTT is a large matrix, and if instead we take the eigenvalue decomposition of T T T u i = λ i u i {\displaystyle \mathbf {T} ^{T}\mathbf {T} \mathbf {u} _{i}=\lambda _{i}\mathbf {u} _{i}} then we notice that by pre-multiplying both sides of the equation with T, we obtain T T T T u i = λ i T u i {\displaystyle \mathbf {T} \mathbf {T} ^{T}\mathbf {T} \mathbf {u} _{i}=\lambda _{i}\mathbf {T} \mathbf {u} _{i}} Meaning that, if ui is an eigenvector of TTT, then vi = Tui is an eigenvector of S. If we have

  Read more →
• 

  Cache language model

  A cache language model is a type of statistical language model. These occur in the natural language processing subfield of computer science and assign probabilities to given sequences of words by means of a probability distribution. Statistical language models are key components of speech recognition systems and of many machine translation systems: they tell such systems which possible output word sequences are probable and which are improbable. The particular characteristic of a cache language model is that it contains a cache component and assigns relatively high probabilities to words or word sequences that occur elsewhere in a given text. The primary, but by no means sole, use of cache language models is in speech recognition systems. To understand why it is a good idea for a statistical language model to contain a cache component one might consider someone who is dictating a letter about elephants to a speech recognition system. Standard (non-cache) N-gram language models will assign a very low probability to the word "elephant" because it is a very rare word in English. If the speech recognition system does not contain a cache component, the person dictating the letter may be annoyed: each time the word "elephant" is spoken another sequence of words with a higher probability according to the N-gram language model may be recognized (e.g., "tell a plan"). These erroneous sequences will have to be deleted manually and replaced in the text by "elephant" each time "elephant" is spoken. If the system has a cache language model, "elephant" will still probably be misrecognized the first time it is spoken and will have to be entered into the text manually; however, from this point on the system is aware that "elephant" is likely to occur again – the estimated probability of occurrence of "elephant" has been increased, making it more likely that if it is spoken it will be recognized correctly. Once "elephant" has occurred several times, the system is likely to recognize it correctly every time it is spoken until the letter has been completely dictated. This increase in the probability assigned to the occurrence of "elephant" is an example of a consequence of machine learning and more specifically of pattern recognition. There exist variants of the cache language model in which not only single words but also multi-word sequences that have occurred previously are assigned higher probabilities (e.g., if "San Francisco" occurred near the beginning of the text subsequent instances of it would be assigned a higher probability). The cache language model was first proposed in a paper published in 1990, after which the IBM speech-recognition group experimented with the concept. The group found that implementation of a form of cache language model yielded a 24% drop in word-error rates once the first few hundred words of a document had been dictated. A detailed survey of language modeling techniques concluded that the cache language model was one of the few new language modeling techniques that yielded improvements over the standard N-gram approach: "Our caching results show that caching is by far the most useful technique for perplexity reduction at small and medium training data sizes". The development of the cache language model has generated considerable interest among those concerned with computational linguistics in general and statistical natural language processing in particular: recently, there has been interest in applying the cache language model in the field of statistical machine translation. The success of the cache language model in improving word prediction rests on the human tendency to use words in a "bursty" fashion: when one is discussing a certain topic in a certain context, the frequency with which one uses certain words will be quite different from their frequencies when one is discussing other topics in other contexts. The traditional N-gram language models, which rely entirely on information from a very small number (four, three, or two) of words preceding the word to which a probability is to be assigned, do not adequately model this "burstiness". Recently, the cache language model concept – originally conceived for the N-gram statistical language model paradigm – has been adapted for use in the neural paradigm. For instance, recent work on continuous cache language models in the recurrent neural network (RNN) setting has applied the cache concept to much larger contexts than before, yielding significant reductions in perplexity. Another recent line of research involves incorporating a cache component in a feed-forward neural language model (FN-LM) to achieve rapid domain adaptation.

  Read more →
• 

  Convolutional neural network

  A convolutional neural network (CNN) is a type of feedforward neural network that learns features via filter (or kernel) optimization. This type of deep learning network has been applied to process and make predictions from many different types of data including text, images and audio. CNNs are the de-facto standard in deep learning-based approaches to computer vision and image processing, and have only recently been replaced—in some cases—by newer architectures such as the transformer. Vanishing gradients and exploding gradients, seen during backpropagation in earlier neural networks, are prevented by the regularization that comes from using shared weights over fewer connections. For example, for each neuron in the fully-connected layer, 10,000 weights would be required for processing an image sized 100 × 100 pixels. However, applying cascaded convolution (or cross-correlation) kernels, only 25 weights for each convolutional layer are required to process 5x5-sized tiles. Higher-layer features are extracted from wider context windows, compared to lower-layer features. Some applications of CNNs include: image and video recognition, recommender systems, image classification, image segmentation, medical image analysis, natural language processing, brain–computer interfaces, and financial time series. CNNs are also known as shift invariant or space invariant artificial neural networks, based on the shared-weight architecture of the convolution kernels or filters that slide along input features and provide translation-equivariant responses known as feature maps. Counter-intuitively, most convolutional neural networks are not invariant to translation, due to the downsampling operation they apply to the input. Feedforward neural networks are usually fully connected networks, that is, each neuron in one layer is connected to all neurons in the next layer. The "full connectivity" of these networks makes them prone to overfitting data. Typical ways of regularization, or preventing overfitting, include: penalizing parameters during training (such as weight decay) or trimming connectivity (skipped connections, dropout, etc.) Robust datasets also increase the probability that CNNs will learn the generalized principles that characterize a given dataset rather than the biases of a poorly-populated set. Convolutional networks were inspired by biological processes in that the connectivity pattern between neurons resembles the organization of the animal visual cortex. Individual cortical neurons respond to stimuli only in a restricted region of the visual field known as the receptive field. The receptive fields of different neurons partially overlap such that they cover the entire visual field. CNNs use relatively little pre-processing compared to other image classification algorithms. This means that the network learns to optimize the filters (or kernels) through automated learning, whereas in traditional algorithms these filters are hand-engineered. This simplifies and automates the process, enhancing efficiency and scalability overcoming human-intervention bottlenecks. == Architecture == A convolutional neural network consists of an input layer, hidden layers and an output layer. In a convolutional neural network, the hidden layers include one or more layers that perform convolutions. Typically this includes a layer that performs a dot product of the convolution kernel with the layer's input matrix. This product is usually the Frobenius inner product, and its activation function is commonly ReLU. As the convolution kernel slides along the input matrix for the layer, the convolution operation generates a feature map, which in turn contributes to the input of the next layer. This is followed by other layers such as pooling layers, fully connected layers, and normalization layers. Here it should be noted how close a convolutional neural network is to a matched filter. === Convolutional layers === In a CNN, the input is a tensor with shape: (number of inputs) × (input height) × (input width) × (input channels) After passing through a convolutional layer, the image becomes abstracted to a feature map, also called an activation map, with shape: (number of inputs) × (feature map height) × (feature map width) × (feature map channels). Convolutional layers convolve the input and pass its result to the next layer. This is similar to the response of a neuron in the visual cortex to a specific stimulus. Each convolutional neuron processes data only for its receptive field. Although fully connected feedforward neural networks can be used to learn features and classify data, this architecture is generally impractical for larger inputs (e.g., high-resolution images), which would require massive numbers of neurons because each pixel is a relevant input feature. A fully connected layer for an image of size 100 × 100 has 10,000 weights for each neuron in the second layer. Convolution reduces the number of free parameters, allowing the network to be deeper. For example, using a 5 × 5 tiling region, each with the same shared weights, requires only 25 neurons. Using shared weights means there are many fewer parameters, which helps avoid the vanishing gradients and exploding gradients problems seen during backpropagation in earlier neural networks. To speed processing, standard convolutional layers can be replaced by depthwise separable convolutional layers, which are based on a depthwise convolution followed by a pointwise convolution. The depthwise convolution is a spatial convolution applied independently over each channel of the input tensor, while the pointwise convolution is a standard convolution restricted to the use of 1 × 1 {\displaystyle 1\times 1} kernels. === Pooling layers === Convolutional networks may include local and/or global pooling layers along with traditional convolutional layers. Pooling layers reduce the dimensions of data by combining the outputs of neuron clusters at one layer into a single neuron in the next layer. Local pooling combines small clusters, tiling sizes such as 2 × 2 are commonly used. Global pooling acts on all the neurons of the feature map. There are two common types of pooling in popular use: max and average. Max pooling uses the maximum value of each local cluster of neurons in the feature map, while average pooling takes the average value. === Fully connected layers === Fully connected layers connect every neuron in one layer to every neuron in another layer. It is the same as a traditional multilayer perceptron neural network (MLP). Each neuron in the fully connected layer receives input from all the neurons in the previous layer. These inputs are weighted and summed with the corresponding biases, and then passed through an activation function to perform a nonlinear transformation, generating the output. The flattened matrix goes through a fully connected layer to classify the images. === Receptive field === In neural networks, each neuron receives input from some number of locations in the previous layer. In a convolutional layer, each neuron receives input from only a restricted area of the previous layer called the neuron's receptive field. Typically the area is a square (e.g. 5 by 5 neurons). Whereas, in a fully connected layer, the receptive field is the entire previous layer. Thus, in each convolutional layer, each neuron takes input from a larger area in the input than previous layers. This is due to applying the convolution over and over, which takes the value of a pixel into account, as well as its surrounding pixels. When using dilated layers, the number of pixels in the receptive field remains constant, but the field is more sparsely populated as its dimensions grow when combining the effect of several layers. To manipulate the receptive field size as desired, there are some alternatives to the standard convolutional layer. For example, atrous or dilated convolution expands the receptive field size without increasing the number of parameters by interleaving visible and blind regions. Moreover, a single dilated convolutional layer can comprise filters with multiple dilation ratios, thus having a variable receptive field size. === Weights === Each neuron in a neural network computes an output value by applying a specific function to the input values received from the receptive field in the previous layer. The function that is applied to the input values is determined by a vector of weights and a bias (typically real numbers). Learning consists of iteratively adjusting these biases and weights. The vectors of weights and biases are called filters and represent particular features of the input (e.g., a particular shape). A distinguishing feature of CNNs is that many neurons can share the same filter. This reduces the memory footprint because a single bias and a single vector of weights are used across all receptive fields that share that filter, as opposed to each receptive field having its own bias and vector

  Read more →
• 

  Noom

  Noom is an American privately held digital health company that provides weight management and behavioral health services through a subscription-based mobile application. Founded in 2008, the company combines behavior change psychology with access to weight loss medications and dietary supplements. The platform incorporates elements of cognitive behavioral therapy (CBT) and goal-setting strategies, and its programs are designed to support users in developing healthier habits. In addition to its weight management services, Noom has expanded to offer products related to stress management and general wellness. Noom has received both praise and criticism. Supporters cite its focus on mental and behavioral aspects of health, while critics have raised concerns about the accuracy of its calorie goals, the use of algorithmically determined weight loss targets, and questions about the qualifications of some of its coaching staff. == History == Noom was founded in 2008 by friends Artem Petakov and Saeju Jeong. The company's mobile app officially launched in 2016. In 2025, Noom relocated its headquarters from New York City to Princeton, New Jersey. Petakov, a former software engineer at Google, currently leads Noom Ventures, while Jeong serves as Noom's Chairman. In 2023, Geoff Cook was appointed CEO of Noom. In 2019, Noom partnered with Novo Nordisk to offer patients prescribed the diabetes medication Saxenda one year of free access to the Noom platform. In 2020, Noom reported $400 million in revenue. As of April 2021, the company stated it employed approximately 3,000 people, including 2,700 coaches. == Services == === Noom App === The Noom app is the primary platform through which users engage with the company's services. Upon creating an account, users are prompted to provide physical information such as weight, height, and age, along with experiential data including lifestyle habits, personal goals, and perceived obstacles. Users log their meals and physical activity, and in return, the app delivers feedback through multiple channels: algorithmically generated insights, guidance from a human coach, peer interaction, educational articles, and interactive quizzes. The app has been reviewed by a range of media outlets, including newspapers such as the Chicago Tribune and USA Today; health information sources such as WebMD; and lifestyle magazines including Good Housekeeping. === Other services === In 2024, Noom launched Noom Vibe, a mobile application that encourages users to develop healthy habits by awarding "vibes"—a form of points—for activities such as walking or meeting step goals. That same year, Noom introduced a 3D body scanning feature within its app, designed to help users monitor physical changes and prevent muscle atrophy during weight loss. Also in 2024, Noom began offering a compounded GLP-1 medication as part of its weight management program. The formulation includes the same active ingredient found in the anti-obesity medications Wegovy and Ozempic. == Research == In 2016, a study published in Scientific Reports analyzed data from approximately 36,000 users of the Noom app, of whom 78% were female and 22% male. The data were collected between October 2012 and April 2014. To be included in the analysis, users had to log their weight at least twice per month over a period of six consecutive months. The study found that 78% of participants self-reported weight loss while using the app. The median duration of weight reporting was 267 days (approximately nine months). The frequency of data logging was positively correlated with weight loss. Additionally, male users had a higher average starting BMI and reported greater average weight loss compared to female users. In 2017, the Centers for Disease Control and Prevention (CDC) recognized Noom as a certified diabetes prevention program, making it the first mobile health application to receive such designation. == Criticisms == === Health programs === Noom has been criticized for promoting elements of diet culture in its advertising campaigns. The app has also faced criticism for setting calorie goals that some users and experts have deemed inappropriately low, and for employing coaches who may lack formal qualifications as registered dietitians. Coaching has been described as relying heavily on canned responses. Upon sign-up, users are prompted to complete a questionnaire consisting of over 50 questions, which is used to generate a personalized program. In 2021, the UK-based organization Privacy International alleged that Noom, along with other diet platforms, used such lengthy surveys to attract users but did not always tailor the resulting programs to the collected data. The organization claimed that many users received the same or highly similar programs regardless of their answers. It also raised concerns about the handling of potentially sensitive health data, alleging a lack of transparency regarding the sharing of such data with third parties, including Facebook, potentially in violation of the European General Data Protection Regulation (GDPR). In a follow-up investigation in 2023, Privacy International reported that Noom had made "significant positive changes" to its data handling practices. However, the organization noted that data was still being shared with Facebook and concluded that "there is still room for improvement." === Billing issues lawsuit === In August 2020, the Better Business Bureau (BBB) issued a warning to consumers regarding Noom's subscription practices. The BBB reported that numerous customers had filed complaints about difficulties canceling their subscriptions after the free trial period, as well as challenges in contacting the company to request refunds. In February 2022, Noom agreed to a $62 million settlement in a class-action lawsuit that alleged the company had used deceptive billing practices related to automatic subscription renewals. Qualifying claimants received approximately $167 each. During the case, a former senior software engineer at Noom testified that the cancellation process was intentionally designed to be difficult, with the goal of generating revenue from customers who failed to cancel in time. In response, Noom stated that it had taken steps to improve transparency around its pricing and policies, including the implementation of self-service cancellation tools.

  Read more →
• 

  Xiaomi MiMo

  Xiaomi MiMo is a family of large language models (LLMs) developed by Xiaomi. It was initially released in April 2025 with the MiMo-7B model. Currently, MiMo is available for developers through API service. It is used as the key AI model in Xiaomi's "Human x Car x Home" ecosystem. == Development == Xiaomi developed MiMo as a reasoning-focused language model. Its development team was led by Luo Fuli, who had previously worked at DeepSeek before joining Xiaomi in late 2025. The model was trained using multi-token prediction and reinforcement learning, with a particular emphasis on mathematical reasoning and code generation tasks. In March 2026, Xiaomi CEO Lei Jun announced that the company planned to invest at least US$8.7 billion in artificial intelligence over the following three years. == Models == === List of models === === MiMo-7B === MiMo-7B is the first model of this LLM. The base model, MiMo-7B-Base, was pre-trained on approximately 25 trillion tokens using web pages, academic papers, books, and synthetic reasoning data. MiMo-7B-RL underwent supervised fine-tuning and reinforcement learning on 130,000 mathematics and code problems. MiMo-7B-RL-0530 was released in May 2025. It scaled the fine-tuning dataset from 500,000 to 6 million instances and extended the RL window from 32,000 to 48,000 tokens and improved AIME 2024 scores from 68.2 to 80.1. MiMo-VL-7B was a vision-language model combining a Vision Transformer encoder with the MiMo-7B backbone. It was trained in four stages consuming 2.4 trillion tokens. Its reinforcement learning variant used Mixed On-Policy Reinforcement Learning (MORL) which integrated reward signals across perception, grounding, and reasoning. Xiaomi also released MiMo-Audio-7B, an audio-language model for voice conversion, style transfer, and speech editing. === MiMo-V2-Flash === MiMo-V2-Flash was launched in December 2025. It is a open-sourced Mixture-of-experts model with 309 billion total parameters and 15 billion active parameters. It was trained on 27 trillion tokens using FP8 mixed precision. It used hybrid attention interleaving Sliding Window and Global Attention at a 5:1 ratio. === MiMo-V2-Pro === Xiaomi publicly introduced MiMo-V2-Pro on 18 March 2026. It has over 1 trillion total parameters, 42 billion active, and a 1-million-token context window. Before the official release, the model had appeared anonymously on OpenRouter under the codename "Hunter Alpha," where it drew substantial usage and topped daily charts for several days, according to Xiaomi and Reuters. During its listing on OpenRouter, the model reportedly processed over one trillion tokens in total usage. Xiaomi later said Hunter Alpha was an early internal test build of MiMo-V2-Pro, and Reuters reported that the model had been mistaken by some users for a possible DeepSeek system before Xiaomi confirmed its origin. The model was released as a proprietary API product, and Luo Fuli stated that Xiaomi intended to open-source a variant at an unspecified future date. Xiaomi has partnered with several API web platforms like OpenClaw to launch the model. All these websites initially offered a free trial of this model for a week, but due to the overwhelming response, Xiaomi later extended the free trial period of the model until 2 April 2026. === MiMo-V2-Omni === Alongside MiMo-V2-Pro, Xiaomi launched MiMo-V2-Omni on 18 March 2026. It handles image, video, audio, and text inputs. Before the official release, it was codenamed "Healer Alpha" in OpenRouter. === MiMo-V2-TTS === On the same date as the release of MiMo-V2-Pro and MiMo-V2-Omni, a Text-to-Speech model named MiMo-V2-TTS was released also. It is a speech synthesis model. It was trained on audio data, which makes it capable of emotional transitions, mid-sentence tone shifts, singing, and synthesis of regional dialects like Sichuan, Cantonese, Henan, and Taiwanese. == Licensing == Xiaomi has used different licensing approaches for different models in the MiMo family. The MiMo-7B series and MiMo-V2-Flash were released as open-weight models. MiMo-V2-Flash was published under the MIT license with model weights and inference code available on Hugging Face. MiMo-V2-Pro and MiMo-V2-Omni were released as proprietary models. It was accessible through Xiaomi's API platform and third-party API providers. Luo Fuli stated that Xiaomi intended to open-source a variant of MiMo-V2-Pro. Although, she did not specify any timeline. MiMo-V2-TTS was released as a proprietary model with no publicly available weights.

  Read more →
• 

  BulSemCor

  The Bulgarian Sense-annotated Corpus (BulSemCor) (Bulgarian: Български семантично анотиран корпус (БулСемКор)) is a structured corpus of Bulgarian texts in which each lexical item is assigned a sense tag. BulSemCor was created by the Department of Computational Linguistics at the Institute for Bulgarian Language of the Bulgarian Academy of Sciences. == Structure == BulSemCor was created as part of a nationally funded project titled "BulNet – A lexico-semantic network for the Bulgarian Language" (2005–2010). It follows the general methodology of SemCor combined with some specific principles. The corpus for annotation consists of 101,791 tokens covering an excerpt from the Bulgarian "Brown" Corpus modelled on the Brown Corpus.Francis Kucera An important feature of BulSemCor is that the samples are selected using heuristics that provide optimal coverage of ambiguous lexis. BulSemCor is manually sense-annotated according to the Bulgarian WordNet. Its size is comparable to that of other contemporary semantically annotated corpora or pool of acceptable linguistic components. The semantic annotation consists in associating each lexical item in the corpus with exactly one synonym set (synset) in the Bulgarian WordNet that best describes its sense in the particular context. The selection of the best match among the suggested candidates is based on a set of procedures, such as the other synset members, the synset gloss (explanatory definition) and the position of a given candidate in the WordNet structure. == Scale == The number of annotated tokens is 99,480 (the difference in the number of tokens compared to the initial corpus is due to the fact that some of them are not linguistic items). The simple word count is 86,842 and multiword expressions (MWE) are 5,797 (12,638 tokens). == Specific features == All words in BulSemCor are assigned a sense, while according to established practice only simple content words or content word classes (typically nouns and verbs) are annotated. Since 2000 the development of language resources, has broadened to include annotation of function words and multiword expressions covering particular senses or types of words and expressions. In this respect, BulSemCor's annotation is more exhaustive and hence provides greater opportunities for linguistic observations and non-linear programming (NLP) applications. Annotated items inherit the linguistic information associated with the corresponding synset, which along with morphological and semantic tags may include annotation on one or more of the following additional levels: Partial information about the syntactic structure of MWE types – particularly, information about syntactic heads and their dependents; Information about the category of the named entities – names, locations, organisations, dates, numbers, etc.; Information about the taxonomic category of adverbs, such as time, place, manner, degree, quantity, etc.; Information about the type of the syntactic relationships – coordination or subordination – expressed by conjunctions; Information about the original part-of-speech of substantivised words (non-nouns that act as nouns in a particular context); Stylistic/register, grammatical and other information about synsets or individual synset members;

  Read more →