AI Art Video

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

• 

  Averbis

  Averbis has a focus on healthcare, pharma, automotive and intellectual property analytics. Averbis is involved in various research projects of the German Federal Ministry of Economics and Energy and the European Union such as DebugIT, EUCases, Mantra and SEMCARE. In addition to these projects, Averbis was also involved in the following projects: Greenpilot is a virtual library, which provides technical information in the fields of nutrition, environment and agriculture. Medpilot is a virtual library, which provides information about medicine and related sciences. In 2013, Averbis has been nominated for the German Founder Prize 2013. Averbis GmbH provides text analytics and text mining software to transform unstructured text into actionable information. It was founded in 2007 by IT experts after years of relevant scientific experience in the field of text mining and multilingual information retrieval. Averbis works in the field of terminology management, natural language processing, machine learning and semantic search. Its text mining software is embedded into the text mining framework UIMA.

  Read more →
• 

  Best AI Art Generators in 2026

  Curious about the best AI art generator? An AI art generator is software that uses machine learning to help you get more done — it combines speed, accuracy, and an interface that just works. Hands-on testing shows real-world results vary, so a short free trial is the smartest way to decide. Whether you are a beginner or a pro, the right AI art generator slots into your workflow and pays for itself fast. This guide breaks down the top picks, their pros and cons, and who each one is best for.

  Read more →
• 

  Hebbian theory

  Hebbian theory is a neuropsychological theory claiming that an increase in synaptic efficacy arises from a presynaptic cell's repeated and persistent stimulation of a postsynaptic cell. It is an attempt to explain synaptic plasticity, the adaptation of neurons during the learning process. Hebbian theory was introduced by Donald Hebb in his 1949 book The Organization of Behavior. The theory is also called Hebb's rule, Hebb's law, Hebb's postulate, and cell assembly theory. Hebb states it as follows: Let us assume that the persistence or repetition of a reverberatory activity (or "trace") tends to induce lasting cellular changes that add to its stability. ... When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A's efficiency, as one of the cells firing B, is increased. The theory is often summarized as "Neurons that fire together, wire together." However, Hebb emphasized that cell A needs to "take part in firing" cell B, and such causality can occur only if cell A fires just before, not at the same time as, cell B. This aspect of causation in Hebb's work foreshadowed what is now known about spike-timing-dependent plasticity, which requires temporal precedence. Hebbian theory attempts to explain associative or Hebbian learning, in which simultaneous activation of cells leads to pronounced increases in synaptic strength between those cells. It also provides a biological basis for errorless learning methods for education and memory rehabilitation. In the study of neural networks in cognitive function, it is often regarded as the neuronal basis of unsupervised learning. == Engrams, cell assembly theory, and learning == Hebbian theory provides an explanation for how neurons might connect to become engrams, which may be stored in overlapping cell assemblies, or groups of neurons that encode specific information. Initially created as a way to explain recurrent activity in specific groups of cortical neurons, Hebb's theories on the form and function of cell assemblies can be understood from the following: The general idea is an old one, that any two cells or systems of cells that are repeatedly active at the same time will tend to become 'associated' so that activity in one facilitates activity in the other. Hebb also wrote: When one cell repeatedly assists in firing another, the axon of the first cell develops synaptic knobs (or enlarges them if they already exist) in contact with the soma of the second cell. D. Alan Allport posits additional ideas regarding cell assembly theory and its role in forming engrams using the concept of auto-association, or the brain's ability to retrieve information based on a partial cue, described as follows: If the inputs to a system cause the same pattern of activity to occur repeatedly, the set of active elements constituting that pattern will become increasingly strongly inter-associated. That is, each element will tend to turn on every other element and (with negative weights) to turn off the elements that do not form part of the pattern. To put it another way, the pattern as a whole will become 'auto-associated'. We may call a learned (auto-associated) pattern an engram. Research conducted in the laboratory of Nobel laureate Eric Kandel has provided evidence supporting the role of Hebbian learning mechanisms at synapses in the marine gastropod Aplysia californica. Because synapses in the peripheral nervous system of marine invertebrates are much easier to control in experiments, Kandel's research found that Hebbian long-term potentiation along with activity-dependent presynaptic facilitation are both necessary for synaptic plasticity and classical conditioning in Aplysia californica. While research on invertebrates has established fundamental mechanisms of learning and memory, much of the work on long-lasting synaptic changes between vertebrate neurons involves the use of non-physiological experimental stimulation of brain cells. However, some of the physiologically relevant synapse modification mechanisms that have been studied in vertebrate brains do seem to be examples of Hebbian processes. One such review indicates that long-lasting changes in synaptic strengths can be induced by physiologically relevant synaptic activity using both Hebbian and non-Hebbian mechanisms. == Principles == In artificial neurons and artificial neural networks, Hebb's principle can be described as a method of determining how to alter the weights between model neurons. The weight between two neurons increases if the two neurons activate simultaneously, and reduces if they activate separately. Nodes that tend to be either both positive or both negative at the same time have strong positive weights, while those that tend to be opposite have strong negative weights. The following is a formulaic description of Hebbian learning (many other descriptions are possible): w i j = x i x j , {\displaystyle \,w_{ij}=x_{i}x_{j},} where w i j {\displaystyle w_{ij}} is the weight of the connection from neuron j {\displaystyle j} to neuron i {\displaystyle i} , and x i {\displaystyle x_{i}} is the input for neuron i {\displaystyle i} . This is an example of pattern learning, where weights are updated after every training example. In a Hopfield network, connections w i j {\displaystyle w_{ij}} are set to zero if i = j {\displaystyle i=j} (no reflexive connections allowed). With binary neurons (activations either 0 or 1), connections would be set to 1 if the connected neurons have the same activation for a pattern. When several training patterns are used, the expression becomes an average of the individuals: w i j = 1 p ∑ k = 1 p x i k x j k , {\displaystyle w_{ij}={\frac {1}{p}}\sum _{k=1}^{p}x_{i}^{k}x_{j}^{k},} where w i j {\displaystyle w_{ij}} is the weight of the connection from neuron j {\displaystyle j} to neuron i {\displaystyle i} , p {\displaystyle p} is the number of training patterns and x i k {\displaystyle x_{i}^{k}} the k {\displaystyle k} -th input for neuron i {\displaystyle i} . This is learning by epoch, with weights updated after all the training examples are presented and is last term applicable to both discrete and continuous training sets. Again, in a Hopfield network, connections w i j {\displaystyle w_{ij}} are set to zero if i = j {\displaystyle i=j} (no reflexive connections). A variation of Hebbian learning that takes into account phenomena such as blocking and other neural learning phenomena is the mathematical model of Harry Klopf. Klopf's model assumes that parts of a system with simple adaptive mechanisms can underlie more complex systems with more advanced adaptive behavior, such as neural networks. == Relationship to unsupervised learning, stability, and generalization == Because of the simple nature of Hebbian learning, based only on the coincidence of pre- and post-synaptic activity, it may not be intuitively clear why this form of plasticity leads to meaningful learning. However, it can be shown that Hebbian plasticity does pick up the statistical properties of the input in a way that can be categorized as unsupervised learning. This can be mathematically shown in a simplified example. Let us work under the simplifying assumption of a single rate-based neuron of rate y ( t ) {\displaystyle y(t)} , whose inputs have rates x 1 ( t ) . . . x N ( t ) {\displaystyle x_{1}(t)...x_{N}(t)} . The response of the neuron y ( t ) {\displaystyle y(t)} is usually described as a linear combination of its input, ∑ i w i x i {\displaystyle \sum _{i}w_{i}x_{i}} , followed by a response function f {\displaystyle f} : y = f ( ∑ i = 1 N w i x i ) . {\displaystyle y=f\left(\sum _{i=1}^{N}w_{i}x_{i}\right).} As defined in the previous sections, Hebbian plasticity describes the evolution in time of the synaptic weight w {\displaystyle w} : d w i d t = η x i y . {\displaystyle {\frac {dw_{i}}{dt}}=\eta x_{i}y.} Assuming, for simplicity, an identity response function f ( a ) = a {\displaystyle f(a)=a} , we can write d w i d t = η x i ∑ j = 1 N w j x j {\displaystyle {\frac {dw_{i}}{dt}}=\eta x_{i}\sum _{j=1}^{N}w_{j}x_{j}} or in matrix form: d w d t = η x x T w . {\displaystyle {\frac {d\mathbf {w} }{dt}}=\eta \mathbf {x} \mathbf {x} ^{T}\mathbf {w} .} As in the previous chapter, if training by epoch is done an average ⟨ … ⟩ {\displaystyle \langle \dots \rangle } over discrete or continuous (time) training set of x {\displaystyle \mathbf {x} } can be done: d w d t = ⟨ η x x T w ⟩ = η ⟨ x x T ⟩ w = η C w . {\displaystyle {\frac {d\mathbf {w} }{dt}}=\langle \eta \mathbf {x} \mathbf {x} ^{T}\mathbf {w} \rangle =\eta \langle \mathbf {x} \mathbf {x} ^{T}\rangle \mathbf {w} =\eta C\mathbf {w} .} where C = ⟨ x x T ⟩ {\displaystyle C=\langle \,\mathbf {x} \mathbf {x} ^{T}\rangle } is the correlation matrix of the input under the additional assumption that ⟨ x ⟩ = 0 {\displaystyle \langle \mathbf

  Read more →
• 

  Best AI Blog Writers in 2026

  Trying to pick the best AI blog writer? An AI blog writer is software that uses machine learning to help you get more done — it scales effortlessly from a single task to thousands. The best picks balance beginner-friendly simplicity with the depth power users need, and they ship updates often. Whether you are a beginner or a pro, the right AI blog writer slots into your workflow and pays for itself fast. Read on for hands-on impressions, pricing tiers, and the standout features that matter.

  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 →
• 

  Frederick Jelinek

  Frederick Jelinek (18 November 1932 – 14 September 2010) was a Czech-American researcher in information theory, automatic speech recognition, and natural language processing. He is well known for his oft-quoted statement, "Every time I fire a linguist, the performance of the speech recognizer goes up". Jelinek was born in Czechoslovakia before World War II and emigrated with his family to the United States in the early years of the communist regime. He studied engineering at the Massachusetts Institute of Technology and taught for 10 years at Cornell University before accepting a job at IBM Research. In 1961, he married Czech screenwriter Milena Jelinek. At IBM, his team advanced approaches to computer speech recognition and machine translation. After IBM, he went to head the Center for Language and Speech Processing at Johns Hopkins University for 17 years, where he was still working on the day he died. == Personal life == Jelinek was born on November 18, 1932, as Bedřich Jelínek in Kladno to Vilém and Trude Jelínek. His father was Jewish; his mother was born in Switzerland to Czech Catholic parents and had converted to Judaism. Jelínek senior, a dentist, had planned early to escape Nazi occupation and flee to England; he arranged for a passport, visa, and the shipping of his dentistry materials. The couple planned to send their son to an English private school. However, Vilém decided to stay at the last minute and was eventually sent to the Theresienstadt concentration camp, where he died in 1945. The family was forced to move to Prague in 1941, but Frederick, his sister and mother—thanks to the latter's background—escaped the concentration camps. After the war, Jelinek entered in the gymnasium, despite having missed several years of schooling because education of Jewish children had been forbidden since 1942. His mother, anxious that her son should get a good education, made great efforts for their emigration, especially when it became clear he would not be allowed to even attempt the graduation examination. His mother hoped her son would become a physician, but Jelinek dreamed of being a lawyer. He studied engineering in evening classes at the City College of New York and received stipends from the National Committee for a Free Europe that allowed him to study at the Massachusetts Institute of Technology. About his choice of specialty, he said: "Fortunately, to electrical engineering there belonged a discipline whose aim was not the construction of physical systems: the theory of information". He obtained his Ph.D. in 1962, with Robert Fano as his adviser. In 1957, Jelinek paid an unexpected visit to Prague. He had been in Vienna and applied for a visa, hoping to see his former acquaintances again. He met with his old friend Miloš Forman, who introduced him to film student Milena Tobolová—whose screenplay had been the basis for the movie Easy Life (Snadný život). His flight back to the U.S. had a stopover in Munich, during which he called her to propose. Tobolová was considered a dissident and the authorities were not happy with her film. Jelinek asked for help from Jerome Wiesner and Cyrus Eaton, the latter who lobbied Nikita Khrushchev. Following the inauguration of John F. Kennedy, a group of Czech dissidents were allowed to emigrate in January 1961. Thanks to the lobbying, the future Milena Jelinek was one of them. After completing his graduate studies, Jelinek, who had developed an interest in linguistics, had plans to work with Charles F. Hockett at Cornell University. However these fell through and during the next ten years he continued to study information theory. Having previously worked at IBM during a sabbatical, he began full-time work there in 1972—at first on leave for Cornell, but permanently from 1974. He remained there for over twenty years. Although at first he had been offered a regular research job, upon his arrival he learned that Josef Raviv had recently been promoted to head of the newly opened IBM Haifa Research Laboratory, and became head of the Continuous Speech Recognition group at the Thomas J. Watson Research Center. Despite his team's successes in this area, Jelinek's work remained little known in his home country because Czech scientists were not allowed to participate in key conferences. After the 1989 fall of communism, Jelinek helped establish scientific relationships, regularly visiting to lecture and helping to persuade IBM to establish a computing centre at Charles University. In 1993, he retired from IBM and went to Johns Hopkins University's Center for Language and Speech Processing, where he was director and Julian Sinclair Smith Professor of Electrical and Computer Engineering. He was still working there at the time of his death; Jelinek died of a heart attack at the close of an otherwise normal workday in mid-September 2010. He was survived by his wife, daughter and son, sister, stepsister, and three grandchildren, including Sophie Gold Jelinek. == Research and legacy == Information theory was a fashionable scientific approach in the mid '50s. However, pioneer Claude Shannon wrote in 1956 that this trendiness was dangerous. He said, "Our fellow scientists in many different fields, attracted by the fanfare and by the new avenues opened to scientific analysis, are using these ideas in their own problems ... It will be all too easy for our somewhat artificial prosperity to collapse overnight when it is realized that the use of a few exciting words like information, entropy, redundancy, do not solve all our problems." During the next decade, a combination of factors shut down the application of information theory to natural language processing (NLP) problems—in particular machine translation. One factor was the 1957 publication of Noam Chomsky's Syntactic Structures, which stated, "probabilistic models give no insight into the basic problems of syntactic structure". This accorded well with the philosophy of the artificial intelligence research of the time, which promoted rule-based approaches. The other factor was the 1966 ALPAC report, which recommended that the government should stop funding research into machine translation. ALPAC chairman John Pierce later said that the field was filled with "mad inventors or untrustworthy engineers". He said that the underlying linguistic problems must be solved before attempts at NLP could be reasonably made. These elements essentially halted research in the field. Jelinek had begun to develop an interest in linguistics after the immigration of his wife, who initially enrolled in the MIT linguistics program with the help of Roman Jakobson. Jelinek often accompanied her to Chomsky's lectures, and even discussed the possibility of changing orientation with his adviser. Fano was "really upset", and after the failure of his project with Hockett at Cornell, he did not return to this field of research until starting work at IBM. The scope of research at IBM was considerably different from that of most other teams. According to Mark Liberman, "While [Jelinek] was leading IBM's effort to solve the general dictation problem during the decade or so following 1972, most other U.S. companies and academic researchers were working on very limited problems ... or were staying out of the field entirely". Jelinek regarded speech recognition as an information theory problem—a noisy channel, in this case the acoustic signal—which some observers considered a daring approach. The concept of perplexity was introduced in their first model, New Raleigh Grammar, which was published in 1976 as the paper "Continuous Speech Recognition by Statistical Methods" in the journal Proceedings of the IEEE. According to Young, the basic noisy channel approach "reduced the speech recognition problem to one of producing two statistical models". Whereas New Raleigh Grammar was a hidden Markov model, their next model, called Tangora, was broader and involved n-grams, specifically trigrams. Even though "it was obvious to everyone that this model was hopelessly impoverished", it was not improved upon until Jelinek presented another paper in 1999. The same trigram approach was applied to phones in single words. Although the identification of parts of speech turned out not to be very useful for speech recognition, tagging methods developed during these projects are now used in various NLP applications. The incremental research techniques developed at IBM eventually became dominant in the field after DARPA, in the mid-80s, returned to NLP research and imposed that methodology to participating teams, shared common goals, data, and precise evaluation metrics. The Continuous Speech Recognition Group's research, which required large amounts of data to train the algorithms, eventually led to the creation of the Linguistic Data Consortium. In the 1980s, although the broader problem of speech recognition remained unsolved, they sought to apply the methods developed to other problems; machine translat

  Read more →
• 

  Probabilistic automaton

  In mathematics and computer science, the probabilistic automaton (PA) is a generalization of the nondeterministic finite automaton; it includes the probability of a given transition into the transition function, turning it into a transition matrix. Thus, the probabilistic automaton also generalizes the concepts of a Markov chain and of a subshift of finite type. The languages recognized by probabilistic automata are called stochastic languages; these include the regular languages as a subset. The number of stochastic languages is uncountable. The concept was introduced by Michael O. Rabin in 1963; a certain special case is sometimes known as the Rabin automaton (not to be confused with the subclass of ω-automata also referred to as Rabin automata). In recent years, a variant has been formulated in terms of quantum probabilities, the quantum finite automaton. == Informal Description == For a given initial state and input character, a deterministic finite automaton (DFA) has exactly one next state, and a nondeterministic finite automaton (NFA) has a set of next states. A probabilistic automaton (PA) instead has a weighted set (or vector) of next states, where the weights must sum to 1 and therefore can be interpreted as probabilities (making it a stochastic vector). The notions states and acceptance must also be modified to reflect the introduction of these weights. The state of the machine as a given step must now also be represented by a stochastic vector of states, and a state accepted if its total probability of being in an acceptance state exceeds some cut-off. A PA is in some sense a half-way step from deterministic to non-deterministic, as it allows a set of next states but with restrictions on their weights. However, this is somewhat misleading, as the PA utilizes the notion of the real numbers to define the weights, which is absent in the definition of both DFAs and NFAs. This additional freedom enables them to decide languages that are not regular, such as the p-adic languages with irrational parameters. As such, PAs are more powerful than both DFAs and NFAs (which are famously equally powerful). == Formal Definition == The probabilistic automaton may be defined as an extension of a nondeterministic finite automaton ( Q , Σ , δ , q 0 , F ) {\displaystyle (Q,\Sigma ,\delta ,q_{0},F)} , together with two probabilities: the probability P {\displaystyle P} of a particular state transition taking place, and with the initial state q 0 {\displaystyle q_{0}} replaced by a stochastic vector giving the probability of the automaton being in a given initial state. For the ordinary non-deterministic finite automaton, one has a finite set of states Q {\displaystyle Q} a finite set of input symbols Σ {\displaystyle \Sigma } a transition function δ : Q × Σ → ℘ ( Q ) {\displaystyle \delta :Q\times \Sigma \to \wp (Q)} a set of states F {\displaystyle F} distinguished as accepting (or final) states F ⊆ Q {\displaystyle F\subseteq Q} . Here, ℘ ( Q ) {\displaystyle \wp (Q)} denotes the power set of Q {\displaystyle Q} . By use of currying, the transition function δ : Q × Σ → ℘ ( Q ) {\displaystyle \delta :Q\times \Sigma \to \wp (Q)} of a non-deterministic finite automaton can be written as a membership function δ : Q × Σ × Q → { 0 , 1 } {\displaystyle \delta :Q\times \Sigma \times Q\to \{0,1\}} so that δ ( q , a , q ′ ) = 1 {\displaystyle \delta (q,a,q^{\prime })=1} if q ′ ∈ δ ( q , a ) {\displaystyle q^{\prime }\in \delta (q,a)} and 0 {\displaystyle 0} otherwise. The curried transition function can be understood to be a matrix with matrix entries [ θ a ] q q ′ = δ ( q , a , q ′ ) {\displaystyle \left[\theta _{a}\right]_{qq^{\prime }}=\delta (q,a,q^{\prime })} The matrix θ a {\displaystyle \theta _{a}} is then a square matrix, whose entries are zero or one, indicating whether a transition q → a q ′ {\displaystyle q{\stackrel {a}{\rightarrow }}q^{\prime }} is allowed by the NFA. Such a transition matrix is always defined for a non-deterministic finite automaton. The probabilistic automaton replaces these matrices by a family of right stochastic matrices P a {\displaystyle P_{a}} , for each symbol a in the alphabet Σ {\displaystyle \Sigma } so that the probability of a transition is given by [ P a ] q q ′ {\displaystyle \left[P_{a}\right]_{qq^{\prime }}} A state change from some state to any state must occur with probability one, of course, and so one must have ∑ q ′ [ P a ] q q ′ = 1 {\displaystyle \sum _{q^{\prime }}\left[P_{a}\right]_{qq^{\prime }}=1} for all input letters a {\displaystyle a} and internal states q {\displaystyle q} . The initial state of a probabilistic automaton is given by a row vector v {\displaystyle v} , whose components are the probabilities of the individual initial states q {\displaystyle q} , that add to 1: ∑ q [ v ] q = 1 {\displaystyle \sum _{q}\left[v\right]_{q}=1} The transition matrix acts on the right, so that the state of the probabilistic automaton, after consuming the input string a b c {\displaystyle abc} , would be v P a P b P c {\displaystyle vP_{a}P_{b}P_{c}} In particular, the state of a probabilistic automaton is always a stochastic vector, since the product of any two stochastic matrices is a stochastic matrix, and the product of a stochastic vector and a stochastic matrix is again a stochastic vector. This vector is sometimes called the distribution of states, emphasizing that it is a discrete probability distribution. Formally, the definition of a probabilistic automaton does not require the mechanics of the non-deterministic automaton, which may be dispensed with. Formally, a probabilistic automaton PA is defined as the tuple ( Q , Σ , P , v , F ) {\displaystyle (Q,\Sigma ,P,v,F)} . A Rabin automaton is one for which the initial distribution v {\displaystyle v} is a coordinate vector; that is, has zero for all but one entries, and the remaining entry being one. == Stochastic languages == The set of languages recognized by probabilistic automata are called stochastic languages. They include the regular languages as a subset. Let F = Q accept ⊆ Q {\displaystyle F=Q_{\text{accept}}\subseteq Q} be the set of "accepting" or "final" states of the automaton. By abuse of notation, Q accept {\displaystyle Q_{\text{accept}}} can also be understood to be the column vector that is the membership function for Q accept {\displaystyle Q_{\text{accept}}} ; that is, it has a 1 at the places corresponding to elements in Q accept {\displaystyle Q_{\text{accept}}} , and a zero otherwise. This vector may be contracted with the internal state probability, to form a scalar. The language recognized by a specific automaton is then defined as L η = { s ∈ Σ ∗ | v P s Q accept > η } {\displaystyle L_{\eta }=\{s\in \Sigma ^{}\vert vP_{s}Q_{\text{accept}}>\eta \}} where Σ ∗ {\displaystyle \Sigma ^{}} is the set of all strings in the alphabet Σ {\displaystyle \Sigma } (so that is the Kleene star). The language depends on the value of the cut-point η {\displaystyle \eta } , normally taken to be in the range 0 ≤ η < 1 {\displaystyle 0\leq \eta <1} . A language is called η-stochastic if and only if there exists some PA that recognizes the language, for fixed η {\displaystyle \eta } . A language is called stochastic if and only if there is some 0 ≤ η < 1 {\displaystyle 0\leq \eta <1} for which L η {\displaystyle L_{\eta }} is η-stochastic. A cut-point is said to be an isolated cut-point if and only if there exists a δ > 0 {\displaystyle \delta >0} such that | v P ( s ) Q accept − η | ≥ δ {\displaystyle \vert vP(s)Q_{\text{accept}}-\eta \vert \geq \delta } for all s ∈ Σ ∗ {\displaystyle s\in \Sigma ^{}} == Properties == Every regular language is stochastic, and more strongly, every regular language is η-stochastic. A weak converse is that every 0-stochastic language is regular; however, the general converse does not hold: there are stochastic languages that are not regular. Every η-stochastic language is stochastic, for some 0 < η < 1 {\displaystyle 0<\eta <1} . Every stochastic language is representable by a Rabin automaton. If η {\displaystyle \eta } is an isolated cut-point, then L η {\displaystyle L_{\eta }} is a regular language. == p-adic languages == The p-adic languages provide an example of a stochastic language that is not regular, and also show that the number of stochastic languages is uncountable. A p-adic language is defined as the set of strings L η ( p ) = { n 1 n 2 n 3 … | 0 ≤ n k < p and 0. n 1 n 2 n 3 … > η } {\displaystyle L_{\eta }(p)=\{n_{1}n_{2}n_{3}\ldots \vert 0\leq n_{k}\eta \}} in the letters 0 , 1 , 2 , … , ( p − 1 ) {\displaystyle 0,1,2,\ldots ,(p-1)} . That is, a p-adic language is merely the set of real numbers in [0, 1], written in base-p, such that they are greater than η {\displaystyle \eta } . It is straightforward to show that all p-adic languages are stochastic. In particular, this implies that the number of stochastic languages is uncountable. A p-adic

  Read more →
• 

  Tamara Berg

  Tamara Lee Berg is a tenured associate professor at the University of North Carolina at Chapel Hill and a research scientist manager at Facebook AML/FAIR. == Education == Berg obtained her PhD in computer science from the University of California, Berkeley in 2007 as a member of the Berkeley Computer Vision Group. She was an assistant professor at Stony Brook University from 2008 to 2013 before joining University of North Carolina Chapel Hill in 2013. == Research == Berg's research interests are at the boundary of computer vision and natural language processing. In particular, she focuses on understanding the connections between vision and language, for example, to automatically identify people in news photographs, for generating natural language descriptions for images, or for recognising clothing and style. == Selected awards and honours == 2019 Mark Everingham Prize 2013 Marr Prize at the International Conference on Computer Vision 2011 National Science Foundation Career Award == Personal life == Berg is married to fellow computer vision researcher Alexander Berg.

  Read more →
• 

  PropBank

  PropBank is a corpus that is annotated with verbal propositions and their arguments—a "proposition bank". Although "PropBank" refers to a specific corpus produced by Martha Palmer et al., the term propbank is also coming to be used as a common noun referring to any corpus that has been annotated with propositions and their arguments. The PropBank project has played a role in research in natural language processing, and has been used in semantic role labelling. == Comparison == PropBank differs from FrameNet, the resource to which it is most frequently compared, in several ways. PropBank is a verb-oriented resource, while FrameNet is centered on the more abstract notion of frames, which generalizes descriptions across similar verbs (e.g. "describe" and "characterize") as well as nouns and other words (e.g. "description"). PropBank does not annotate events or states of affairs described using nouns. PropBank commits to annotating all verbs in a corpus, whereas the FrameNet project chooses sets of example sentences from a large corpus and only in a few cases has annotated longer continuous stretches of text. PropBank-style annotations often remain close to the syntactic level, while FrameNet-style annotations are sometimes more semantically motivated. From the start, PropBank was developed with the idea of serving as training data for machine learning-based semantic role labeling systems in mind. It requires that all arguments to a verb be syntactic constituents and different senses of a word are only distinguished if the differences bear on the arguments. Due to such differences, semantic role labeling with respect to PropBank is often a somewhat easier task than producing FrameNet-style annotations.

  Read more →
• 

  Dilek Hakkani-Tür

  Dilek Z. Hakkani-Tür is a Turkish-American computer scientist focusing on speech processing, speech recognition, and dialogue systems. She is a professor of computer science at the University of Illinois Urbana-Champaign. == Education and career == Hakkani-Tür is a 1994 graduate of Middle East Technical University in Ankara, Turkey. She continued her studies at Bilkent University, also in Ankara, where she earned a master's degree in 1996 and completed her Ph.D. in 2000. She worked as a researcher at AT&T Labs from 2001 to 2005, at the International Computer Science Institute from 2006 to 2010, at Microsoft Research from 2010 to 2016, at Google Research from 2016 to 2018, and at Amazon Alexa from 2018 to 2023. At Microsoft, she was in the team of scientists that built the first prototype of the Cortana virtual assistant. While working for Amazon Alexa, she also taught at the University of California, Santa Cruz as a distinguished visiting instructor. She joined the University of Illinois Urbana-Champaign faculty in 2023. She was editor-in-chief of IEEE/ACM Transactions on Audio, Speech and Language Processing from 2019 to 2021, and is president of the Special Interest Group on Discourse and Dialogue of the Association for Computational Linguistics for the 2023–2025 term. She has served as co-editor-in-chief of Transactions of the Association for Computational Linguistics since 2024. == Recognition == In 2014, Hakkani-Tür was elected as an IEEE Fellow "for contributions to spoken language processing", and as a Fellow of the International Speech Communication Association "for contributions to advancing the state-of-the-art in spoken language processing, especially for human/human and human/machine conversational understanding". In 2024, she was elected as a Fellow of the Association for Computational Linguistics for her contributions to spoken dialogue systems.

  Read more →
• 

  Top 10 AI Writing Assistants Compared (2026)

  Trying to pick the best AI writing assistant? An AI writing assistant is software that uses machine learning to help you get more done — it scales effortlessly from a single task to thousands. The best picks balance beginner-friendly simplicity with the depth power users need, and they ship updates often. Whether you are a beginner or a pro, the right AI writing assistant slots into your workflow and pays for itself fast. Read on for hands-on impressions, pricing tiers, and the standout features that matter.

  Read more →
• 

  How to Choose an AI Text-to-video Tool

  Comparing the best AI text-to-video tool? An AI text-to-video tool is software that uses machine learning to help you get more done — it lowers the barrier so anyone can produce professional output. Privacy matters too: check whether your data trains the model and whether a no-log or enterprise tier is available. Whether you are a beginner or a pro, the right AI text-to-video tool slots into your workflow and pays for itself fast. Below we compare features, pricing, and real output so you can choose with confidence.

  Read more →
• 

  Sahara Net

  Sahara Net is an information and communications technology provider (ICT) serving the Saudi market, the company has rapidly grown since 1989 to offer various complementary services such as connectivity, internet, hosting, cloud, optimization, cyber security, and managed services. == History == Sahara Net is a Saudi Joint Stock Company (JSC) and its history goes back to 1989 when Sahara Net established the 1st Saudi Bulletin Board Service (BBS) in the Kingdom. During this period, it operated as a hub for email exchange in the FidoNet network. And in 1994 Sahara Net started offering Internet connectivity and other related services like internet email, web design, web hosting, and Domain name registry services. These services made the first ISP in Saudi Arabia before the official licensing in 1998, when the Saudi Internet market was regulated and Sahara Net received Internet Service Provider (ISP) license and was appointed as the official Local Internet Registry (LIR) in the Kingdom of Saudi Arabia. == Today == The company grew over these years to become one of the main ICTs in the Saudi Arabian market, extending network coverage to all major cities in Saudi Arabia, and offering various connectivity options to business as well as home users. In 2009, the company was partially acquired by Telindus (the ICT investment arm of Belgacom), the famous telecom operator in Belgium and Europe. Then, in 2014, the company was fully acquired by its original founders. Recently, Sahara Net was converted from an LLC to a JSC with over 1200 shareholders by a capital raise (original founders still control 70% of the shares).

  Read more →
• 

  Round-trip translation

  Round-trip translation (RTT), also known as back-and-forth translation, recursive translation and bi-directional translation, is the process of translating a word, phrase or text into another language (forward translation), then translating the result back into the original language (back translation), using machine translation (MT) software. It is often used by laypeople to evaluate a machine translation system, or to test whether a text is suitable for MT when they are unfamiliar with the target language. Because the resulting text can often differ substantially from the original, RTT can also be a source of entertainment. == Software quality == To compare the quality of different machine translation systems, users perform RTT and compare the resulting text to the original. The theory is that the closer the result of the RTT is to the original text, the higher the quality of the machine translation system. One of the problems with this technique is that if there is a problem with the resulting text it is impossible to know whether the error occurred in the forward translation, in the back translation, or in both. In addition, it is possible to get a good back translation from a bad forward translation. A study using the automatic evaluation methods BLEU and F-score compared five different free online translation programs, evaluating the quality of both the forward translation and the back translation, and found no correlation between the quality of the forward translation and the quality of the back translation (i.e., a high quality forward translation did not always correspond to a high quality back translation). The author concluded that RTT was a poor method of predicting the quality of machine translation software. This conclusion was reinforced by a more in-depth study also using automatic evaluation methods. A subsequent study which included human evaluation of the back translation in addition to automatic evaluation methods found that RTT might have some ability to predict the quality of a machine translation system not on a sentence-by-sentence basis but for larger texts. == Suitability of text for machine translation == It is also suggested that RTT can be used to determine whether a text is suitable for machine translation. The idea being that if RTT results in a text that is close to the original, the text is suitable for MT. If after using RTT, the resulting text is inaccurate, the source text can then be edited until a satisfactory result is achieved. One of the studies looking at RTT as a means of measuring MT system quality also looked at its ability to predict whether a text was suitable for machine translation. It found that using different types of text also did not result in any correlation between the quality of the forward translation and the quality of the back translation. In contrast another study using human evaluation found that there was a correlation between the quality of the forward translation and the back translation and that this correlation could be used to estimate the quality of the forward translation. This correlation could be used to estimate the quality of the forward translation and by simplifying the source text, improve the quality of the forward translation. == Entertainment == Although the use of RTT for assessing MT system quality or the suitability of a text for MT is in doubt, it is a way to have fun with machine translation. The text produced from an RTT can be comically bad. At one time websites existed for the sole purpose of performing RTT for fun. Other variations send the text through several languages before translating it back into the original or continue translating the text back and forth until it reaches equilibrium (i.e., the result of the back translation is identical to the text used for the forward translation). RTT as entertainment appeared in Philip K. Dick's novel Galactic Pot-Healer. The main character runs book titles and sayings through RTT then has his friends try to guess the original. The Australian television show Spicks and Specks had a contest called "Turning Japanese" which used RTT on song lyrics. Contestants needed to correctly guess the title of the song from which the lyrics were taken.

  Read more →
• 

  Cortana (virtual assistant)

  Cortana is a discontinued virtual assistant developed by Microsoft that used the Bing search engine to perform tasks such as setting reminders and answering questions for users. Cortana was available in English, Portuguese, French, German, Italian, Spanish, Chinese, and Japanese language editions, depending on the software platform and region in which it was used. In 2019, Microsoft began reducing the prevalence of Cortana and converting it from an assistant into different software integrations. It was split from the Windows 10 search bar in April 2019. In January 2020, the Cortana mobile app was removed from certain markets, and on March 31, 2021, the Cortana mobile app was shut down globally. On June 2, 2023, Microsoft announced that support for the Cortana standalone app on Microsoft Windows would end in late 2023 and would be replaced by Microsoft Copilot, an AI chatbot. Support for Cortana in the Microsoft Outlook and Microsoft 365 mobile apps was discontinued in fall of 2023. == History == === Beginnings (2009–2014) === The development of Cortana started in 2009 in the Microsoft Speech products team with general manager Zig Serafin and Chief Scientist Larry Heck. Heck and Serafin established the vision, mission, and long-range plan for Microsoft's digital personal assistant and they built a team with the expertise to create the initial prototypes for Cortana. Some of the key researchers in these early efforts included Microsoft Research researchers Dilek Hakkani-Tür, Gokhan Tur, Andreas Stolcke, and Malcolm Slaney, research software developer Madhu Chinthakunta, and user experience designer Lisa Stifelman. To develop the Cortana digital assistant, the team interviewed human personal assistants. The interviews inspired a number of unique features in Cortana, including the assistant's "notebook" feature. Originally, Cortana was meant to be only a codename, but a petition on Windows Phone's UserVoice site proved to be popular and made the codename official. Cortana was demonstrated for the first time at the Microsoft Build developer conference in San Francisco in April 2014. It was launched as a key ingredient of Microsoft's planned "makeover" of future operating systems for Windows Phone and Windows. It was named after Cortana, a synthetic intelligence character in Microsoft's Halo video game franchise originating in Bungie folklore, with Jen Taylor, the character's voice actress, returning to voice the personal assistant's US-specific version. === Expansion (2015–2018) === In January 2015, Microsoft announced the availability of Cortana for Windows 10 desktops and mobile devices as part of merging Windows Phone into the operating system at large. On May 26, 2015, Microsoft announced that Cortana would also be available on other mobile platforms. An Android release was set for July 2015, but the Android APK file containing Cortana was leaked ahead of its release. It was officially released, along with an iOS version, in December 2015. During E3 2015, Microsoft announced that Cortana would come to the Xbox One as part of a universally designed Windows 10 update for the console. Microsoft integrated Cortana into numerous products such as Microsoft Edge. Microsoft's Cortana assistant was deeply integrated into the browser. Cortana was able to find opening hours when on restaurant sites, show retail coupons for websites, or show weather information in the address bar. At the Worldwide Partners Conference 2015 Microsoft demonstrated Cortana integration with products such as GigJam. Conversely, Microsoft announced in late April 2016 that it would block anything other than Bing and Edge from being used to complete Cortana searches, again raising questions of anti-competitive practices by the company. Microsoft's "Windows in the car" concept included Cortana. The concept makes it possible for drivers to make restaurant reservations and see places before they go there. At Microsoft Build 2016, Microsoft announced plans to integrate Cortana into Skype (Microsoft's video-conferencing and instant messaging service) as a bot to allow users to order food, book trips, transcribe video messages and make calendar appointments through Cortana in addition to other bots. As of 2016, Cortana was able to underline certain words and phrases in Skype conversations that relate to contacts and corporations. A writer from Engadget has criticised the Cortana integration in Skype for responding only to very specific keywords, feeling as if she was "chatting with a search engine" due to the impersonal way the bots replied to certain words such as "Hello" causing the Bing Music bot to bring up Adele's song of that name. Microsoft also announced at Microsoft Build 2016 that Cortana would be able to cloud-synchronise notifications between Windows 10 Mobile's and Windows 10's Action Center, as well as notifications from Android devices. In December 2016, Microsoft announced the preview of Calendar.help, a service that enabled people to delegate the scheduling of meetings to Cortana. Users interact with Cortana by including her in email conversations. Cortana would then check people's availability in Outlook Calendar or Google Calendar, and work with others Cc'd on the email to schedule the meeting. The service relied on automation and human-based computation. In May 2017, Microsoft announced INVOKE, a voice-activated speaker featuring Cortana, in collaboration with Harman Kardon. The premium speaker has a cylindrical design and offers 360-degree sound, the ability to make and receive calls with Skype, and all of the other features currently available with Cortana. In 2017, Microsoft partnered with Amazon to integrate Echo and Cortana with each other, allowing users of each smart assistant to summon the other via a command. This feature preview was released in August 2018. Windows 10 users were able to just say "Hey Cortana, open Alexa" and Echo users were able to say "Alexa, open Cortana" to summon the other assistant. === Decreasing focus and discontinuation (2019–2024) === In January 2019, Microsoft CEO Satya Nadella stated that he no longer saw Cortana as a direct competitor against Alexa and Siri. Shortly thereafter, Microsoft began reducing the prevalence of Cortana and converting it from an assistant into different software integrations. It was split from the Windows 10 search bar in April 2019. In January 2020, the Cortana mobile app was removed from certain markets, and then, on July 24, 2020, Cortana was removed from the Xbox dashboard as part of a redesign. On January 31, 2021, Microsoft removed the Cortana mobile application in many markets, including the UK, Australia, Germany, Mexico, China, Spain, Canada, and India. On March 31, 2021, Microsoft shut down the Cortana apps globally for iOS and Android and removed the apps entirely from their corresponding app stores. To access previously recorded content, users had to use Cortana on Windows 10 or other specialized Microsoft applications. Microsoft also reduced emphasis on Cortana in Windows with the 2021 release of Windows 11. Cortana was not used during the device setup process or pinned to the taskbar by default. On June 2, 2023, Microsoft announced the Cortana standalone app on Windows 10 and Windows 11 which would shut down later in the year. In its support article, Microsoft listed several alternatives, most of which have since been rebranded as Microsoft Copilot. They also added that the change would not impact Cortana in Office 365 and Teams environments. On August 11, 2023, Microsoft updated the Cortana standalone app in Windows, informing that it was deprecated and can no longer be used. Microsoft's support article announcing the deprecation of Cortana was updated to reflect this change. Along with the deprecation of the standalone app, it was announced that Cortana support in Teams mobile, Microsoft Teams displays, and Teams rooms would end in late 2023. The support article states that Cortana in the “Play my emails” feature of the Microsoft Outlook mobile app would continue to be available. Later in June 2024, the support article was updated, stating that Cortana in the voice search and the "Play my emails" feature is now removed from the Microsoft Outlook mobile app, officially marking the discontinuation of Cortana across all Microsoft products. On May 22, 2024, Microsoft announced the Windows 11 24H2 update, which removed Cortana, Tips, and WordPad from systems. == Functionality == Cortana was able to set reminders, recognize natural voice without the requirement for keyboard input, and answer questions using information from the Bing search engine. Searches using Windows 10 are made only with the Microsoft Bing search engine, and all links will open with Microsoft Edge, except when a screen reader such as Narrator was being used, where the links will open in Internet Explorer. Windows Phone 8.1's universal Bing SmartSearch features were incorporated into Cortana, which replaced the

  Read more →