AI Content Writing

AI Content Writing — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

Cybersecurity in space

Cybersecurity in space involves the defense of all space assets (e.g. navigation systems, satellites, ground antennas, networks, etc.). The security of space can be affected by attacks such as disruption, corruption as well as the destruction of depended-upon assets/collected data. Government (e.g. militaries) and non-government sectors (e.g. financial industries) have started to become more reliant on numerous space-based services. Due to the criticality of these services, space security experts have identified these assets as high-value targets (HVT) that can cause detrimental consequences to all of Earth. == Scope and definitions == Space assets are broken down by three sub-sectors: the space component, the ground component, and the individual user component. The architecture of space assets is extremely complex and allows for a frequent attack vector utilized, the disruption by radio frequency (RF) cyber-attacks. In 2020, a memorandum was published by President Donald Trump, Space Policy Directive‑5 (SPD‑5). It established principles to ensure the safeguarding of all space assets. In 2023, the National Institute of Standards and Technology’s (NIST) published IR 8270, Introduction to Cybersecurity for Commercial Satellite Operations. This report established a baseline risk-management framework (RMF) to be implemented into space operations. == History == During the Cold War in the 1950s-1960s, the United States and Russia entered what was called the “Space Race”. By 1957, the Soviet Union successfully launched the first satellite into space named Sputnik. By 1961, the first key milestone was accomplished when the Soviet Union’s Yuri Gagarin became the first human to orbit Earth. This was later followed by the first American, Alan Shepard, to be launched into space; this was followed by John Glenn becoming the first American to orbit Earth in 1962. In 1969, a pinnacle milestone was reached when Apollo 11 launched into space and Neil Armstrong became the first man to walk on the moon. As space operations furthered, Commercial off-the-shelf products became increasingly popular but resulted in a rapid increase to the cyber-attack surface. Public awareness of space security did not increase until 2022, when the Viasat KA-SAT incident occurred, resulting in the disruption of a large number of modems across Europe. The attack was later accredited to Russia by the U.S. and the U.K. Policy and standards started to rapidly increase by 2020. The establishment of SPD-5 was released in 2020 followed by asset hardening instructions in 2022, and NIST’s IR 8270 in 2023. It was not until 2025 that Europe published their own findings in the Space Threat Landscape 2025 Report. This document led to the EU’s security proposals and standards. == Threats == === Radio-frequency Interference and Global Navigation Satellite Systems (GNSS) Spoofing === Space services are highly dependent on RF links for systems such as GNSS, however, a consequence of this dependency on RF is denial of service and deception. In 2017, the Black Sea maritime event occurred when numerous ships were subject to spoofing. Space services depend on RF links susceptible to jamming (denial) and spoofing (deception), including for GNSS/Positioning, Navigation, and Timing (PNT). Annotated incidents include the 2017 Black Sea maritime spoofing event affecting numerous ships, and extensive aviation GNSS spoofing patterns surveyed in various regions during 2024–2025. === Network intrusion and malware === Cyber threats can intrude and infect assets with malware. They do this by finding misconfiguration vulnerabilities, remote-management interfaces, and/or supply-chain vulnerabilities mainly in ground networks and user terminals. When KA-SAT occurred, it resulted from bulk modem disturbances. Forensic analysts later suggested malicious management controls and wiper malware as the root cause. === Supply-chain and lifecycle risks === The outsource of COTS components, external vendors, and software defined payloads allowed for vulnerabilities to emerge in the System/Product Lifecycle. In response, EU recommended the implementation of lifecycle-wide controls as mitigating factors. === Espionage, disruption, and influence === As Advanced Persistent Threats (APTs), Global Positioning System (GPS) intervention, and information warfare increased, assets like transponders became more frequent targets of attack. == Noteworthy incidents == The Viasat KA‑SAT incident of 2022, where a large number of modems in Europe were disrupted, resulted in the loss of telemetry access to a significant amount of wind turbines in Germany. The mass GNSS deception of the Black Sea in 2017 affected numerous ships when they started to convey fake central locations in Russia. Between 2024 and 2025, there was a mass, repetitive aviation GNSS spoofing that affected the aircraft of various regions. == Standards, guidelines, and best practices == SPD‑5 (U.S.) – This established risk-based engineering, verifying and ensuring positive control, and the implementation of risk mitigation controls. NIST IR 8270 – This created a RMF for COTS satellites. CISA/FBI SATCOM Advisory (AA22‑076) – Provided guidance on hardening techniques such as least-privileged, access control, encryption, etc.). ENISA Space Threat Landscape 2025 – It established the categorization of assets to organize threats, ensuring the observation of system/product lifecycle, and an RMF for COTS satellites. ECSS‑E‑ST‑80C (2024) – This established a standard for securing lifecycles in space, covering all segments (e.g. ground, launch, etc.). == Regulation and governance == As of 2025, there is no international regulations established for space assets, but the U.S., EU, and ESA institutional initiatives have published standards to address security concerns. The U.S. implemented SPD-5 and the Federal Communications Commission (FCC); the FCC addressed orbital debris. While the EU created standards to address technological mandates and support the implementation of NIS2. Lastly, the ESA created a special operations center to safeguard their satellites. International governance is still evolving, but forums have been held by the United Nations Committee on the Peaceful Uses of Outer Space. International conversations under forums such as the UN Committee on the Peaceful Uses of Outer Space (COPUOS) progressively note the cyber–space safety relationship, though formal global norms specific to space cybersecurity continue evolving. == Risk management approaches == Through RMF, mitigation controls have been implemented to reduce the risk of exploitation while increasing the security of space. Controls addressing mitigation include proper configuration, system hardening, zero-trust architectures, encryption, etc. Both the government and industries have placed an emphasis on incident response procedures to identify, contain, and remediate breaches.
Read more →
Generalized multidimensional scaling

Generalized multidimensional scaling (GMDS) is an extension of metric multidimensional scaling, in which the target space is non-Euclidean. When the dissimilarities are distances on a surface and the target space is another surface, GMDS allows finding the minimum-distortion embedding of one surface into another. GMDS is an emerging research direction. Currently, main applications are recognition of deformable objects (e.g. for three-dimensional face recognition) and texture mapping.
Read more →
Prescription monitoring program

In the United States, prescription monitoring programs (PMPs) or prescription drug monitoring programs (PDMPs) are state-run programs which collect and distribute data about the prescription and dispensation of federally controlled substances and, depending on state requirements, other potentially abusable prescription drugs. PMPs are meant to help prevent adverse drug-related events such as opioid overdoses, drug diversion, and substance abuse by decreasing the amount and/or frequency of opioid prescribing, and by identifying those patients who are obtaining prescriptions from multiple providers (i.e., "doctor shopping") or those physicians overprescribing opioids. Most US health care workers support the idea of PMPs, which intend to assist physicians, physician assistants, nurse practitioners, dentists and other prescribers, the pharmacists, chemists and support staff of dispensing establishments. The database, whose use is required by State law, typically requires prescribers and pharmacies dispensing controlled substances to register with their respective state PMPs and (for pharmacies and providers who dispense from their offices) to report the dispensation of such prescriptions to an electronic online database. The majority of PMPs are authorized to notify law enforcement agencies or licensing boards or physicians when a prescriber, or patients receiving prescriptions, exceed thresholds established by the state or prescription recipient exceeds thresholds established by the State. All states have implemented PDMPs, although evidence for the effectiveness of these programs is mixed. While prescription of opioids has decreased with PMP use, overdose deaths in many states have actually increased, with those states sharing data with neighboring jurisdictions or requiring reporting of more drugs experiencing highest increases in deaths. This may be because those declined opioid prescriptions turn to street drugs, whose potency and contaminants carry greater overdose risk. == History == Prescription drug monitoring programs, or PDMPs, are an example of one initiative proposed to alleviate effects of the opioid crisis. The programs are designed to restrict prescription drug abuse by limiting a patient's ability to obtain similar prescriptions from multiple providers (i.e. “doctor shopping”) and reducing diversion of controlled substances. This is meant to reduce risk of fatal overdose caused by high doses of opioids or interactions between opioids and benzodiazepenes, and to enable better decision making on the part of healthcare providers who may be unaware of a patient's prescription drug use, history or other prescriptions. PDMPs have been implemented in state legislations since 1939 in California, a time before electronic medical records, though implementation rose alongside increased awareness of overprescribing of opioids and overdose. A later New York state program was struck down by the U.S. Supreme Court in Whalen v. Roe. But, by 2019, 49 states, the District of Columbia, and Guam had enacted PDMP legislation. In 2021 Missouri, the last State to not use a PMP, adopted legislation to create one. PMPs are constantly being updated to increase speed of data collection, sharing of data across States, and ease of interpretation. This is being done by integrating PDMP reports with other health information technologies such as health information exchanges (HIE), electronic health record (EHR) systems, and/ or pharmacy dispensing software systems. One program that has been implemented in nine states is called the PDMP Electronic Health Records Integration and Interoperability Expansion, also known as PEHRIIE. Another software, marketed by Bamboo Health and integrated with PMPs in 43 states, uses an algorithm to track factors thought to increase risk of diversion, abuse or overdose, and assigns patients a three digit score based on presumed indicators of risk. While some studies have suggested that PDMP-HIT integration and sharing of interstate data brings benefits such as reduced opioid-related inpatient morbidity, others have found no or negative impact on mortality compared to states without PMP data sharing. Patient and media reports suggest need for testing and evaluation of algorithmic software used to score risk, with some patients reporting denial of prescriptions without c explanation or clarity of data. == Goals == Most health care workers support PMPs which intend to assist physicians, physician assistants, nurse practitioners, dentists and other prescribers, the pharmacists, chemists and support staff of dispensing establishments, as well as law-enforcement agencies. The collaboration supports the legitimate medical use of controlled substances while limiting their abuse and diversion. Pharmacies dispensing controlled substances and prescribers typically must register with their respective state PMPs and (for pharmacies and providers who dispense controlled substances from their offices) report the dispensation to an electronic online database. Some pharmacy software can submit these reports automatically to multiple states. == Usage == === List of programs by state === === Software systems === NarxCare is a prescription drug monitoring program (PDMP) run by Bamboo Health. Bamboo Health was formerly known as Appriss. It is widely used across the United States by pharmacies including Rite Aid as well as those at Walmart and Sam’s Club. The NarxCare software allows doctors to view data about a patient, combining data from the prescription registries of various U.S. states to make the registries interoperable nationally. It also uses machine learning to generate an "Overdose Risk Score" that potentially includes EMS and criminal justice data; these scores have been criticized by researchers and patient advocates for the lack of transparency in the process as well as the potential for disparate treatment of women and minority groups. Advertised as an "analytics tool and care management platform", the NarxCare software allows doctors to view data about a patient including how many pharmacies they have visited and the combinations of medication they are prescribed. It combines data from the prescription registries of various U.S. states, making the registries interoperable nationally. It additionally uses machine learning to generate various three-digit "risk scores" and an overall "Overdose Risk Score", collectively referred to as Narx Scores, in a process that potentially includes EMS and criminal justice data as well as court records. == Controversy == Many doctors and researchers support the idea of PDMPs as a tool in combatting the opioid epidemic. Opioid prescribing, opioid diversion and supply, opioid misuse, and opioid-related morbidity and mortality are common elements in data entered into PDMPs. Prescription Monitoring Programs are purported to offer economic benefits for the states who implement them by decreasing overall health care costs, lost productivity, and investigation times. However, there are many studies that conclude the impact of PDMPs is unclear. While use of PMPs has been accompanied by decrease in opioid prescribing, few analyses consider corresponding use of street opioids, extramedical use, or diversion, which might provide a more holistic method for evaluation of PMP intent and efficacy. Evidence for PDMP impact on fatal overdoses is decidedly mixed, with multiple studies finding increased overdose rates in some states, decreases in others, or no clear impact. Interestingly, an increase in heroin overdoses after PDMP implementation has been commonly reported, presumably as denial of prescription opioids sends patients in search of street drugs. Narx Scores have been criticized by researchers and patient advocates for the lack of transparency in the generation process as well as the potential for disparate treatment of women and minority groups. Writing in Duke Law Journal, Jennifer Oliva stated that "black-box algorithms" are used to generate the scores.
Read more →
SqueezeNet

SqueezeNet is a deep neural network for image classification released in 2016. SqueezeNet was developed by researchers at DeepScale, University of California, Berkeley, and Stanford University. In designing SqueezeNet, the authors' goal was to create a smaller neural network with fewer parameters while achieving competitive accuracy. Their best-performing model achieved the same accuracy as AlexNet on ImageNet classification, but has a size 510x less than it. == Version history == SqueezeNet was originally released on February 22, 2016. This original version of SqueezeNet was implemented on top of the Caffe deep learning software framework. Shortly thereafter, the open-source research community ported SqueezeNet to a number of other deep learning frameworks. On February 26, 2016, Eddie Bell released a port of SqueezeNet for the Chainer deep learning framework. On March 2, 2016, Guo Haria released a port of SqueezeNet for the Apache MXNet framework. On June 3, 2016, Tammy Yang released a port of SqueezeNet for the Keras framework. In 2017, companies including Baidu, Xilinx, Imagination Technologies, and Synopsys demonstrated SqueezeNet running on low-power processing platforms such as smartphones, FPGAs, and custom processors. As of 2018, SqueezeNet ships "natively" as part of the source code of a number of deep learning frameworks such as PyTorch, Apache MXNet, and Apple CoreML. In addition, third party developers have created implementations of SqueezeNet that are compatible with frameworks such as TensorFlow. Below is a summary of frameworks that support SqueezeNet. == Relationship to other networks == === AlexNet === SqueezeNet was originally described in SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size. AlexNet is a deep neural network that has 240 MB of parameters, and SqueezeNet has just 5 MB of parameters. This small model size can more easily fit into computer memory and can more easily be transmitted over a computer network. However, it's important to note that SqueezeNet is not a "squeezed version of AlexNet." Rather, SqueezeNet is an entirely different DNN architecture than AlexNet. What SqueezeNet and AlexNet have in common is that both of them achieve approximately the same level of accuracy when evaluated on the ImageNet image classification validation dataset. === Model compression === Model compression (e.g. quantization and pruning of model parameters) can be applied to a deep neural network after it has been trained. In the SqueezeNet paper, the authors demonstrated that a model compression technique called Deep Compression can be applied to SqueezeNet to further reduce the size of the parameter file from 5 MB to 500 KB. Deep Compression has also been applied to other DNNs, such as AlexNet and VGG. == Variants == Some of the members of the original SqueezeNet team have continued to develop resource-efficient deep neural networks for a variety of applications. A few of these works are noted in the following table. As with the original SqueezeNet model, the open-source research community has ported and adapted these newer "squeeze"-family models for compatibility with multiple deep learning frameworks. In addition, the open-source research community has extended SqueezeNet to other applications, including semantic segmentation of images and style transfer.
Read more →
Verbal overshadowing

Verbal overshadowing is a phenomenon where giving a verbal description of sensory input impairs formation of memories of that input. This was first reported by Schooler and Engstler-Schooler (1990) where it was shown that the effects can be observed across multiple domains of cognition which are known to rely on non-verbal knowledge and perceptual expertise. One example of this is memory, which has been known to be influenced by language. Seminal work by Carmichael and collaborators (1932) demonstrated that when verbal labels are connected to non-verbal forms during an individual's encoding process, it could potentially bias the way those forms are reproduced. Because of this, memory performance relying on reportable aspects of memory that encode visual forms should be vulnerable to the effects of verbalization. == Initial findings == Schooler and Engstler-Schooler (1990) were the first to report findings of verbal overshadowing. In their study, participants watched a video of a simulated robbery and were instructed to either verbally describe the robber or engage in a control task. Those who engaged in giving a verbal description were less likely to correctly identify the robber from a test lineup, compared to those who engaged in the control task. A larger effect was detected when the verbal description was provided 20, rather than 5, minutes after the video, and immediately before the test lineup. A meta-analysis by Meissner and Brigham (2001) supported the effects of verbal overshadowing, showing a small but reliably negative effect. == General effects of verbal overshadowing == The effects of verbal overshadowing have been generalized across multiple domains of cognition that are known to rely on non-verbal knowledge and perceptual expertise, such as memory. Memory has been known to be influenced by language. Seminal work by Carmichael and collaborators (1932) demonstrated that labels attached to, or associated with, non-verbal forms during memory encoding can affect the way the forms were subsequently reproduced. Because of this, memory performance that relies on reportable aspects of memory that encode visual forms should be vulnerable to the effects of verbalization. Pelizzon, Brandimonte, and Luccio (2002) found that visual memory representations appear to incorporate visual, spatial, and temporal characteristics. It is explained as follows: With the temporal code (where the only information available is the sequence of the stimuli), performance levels remain high, unless participants are required to retrieve the stimuli in a different order from that used at encoding (visual cue). In this case, performance is significantly impaired, even in the presence of a visual cue. The study showed that order information acts as a link between the two separate representations of figure and background, hence preventing verbal overshadowing at encoding (temporal component) or attenuating its influence at retrieval (spatial component).(p. 960) Hatano, Ueno, Kitagami, and Kawaguchi found that verbal overshadowing is likely to occur when participants verbally described targets in detail. Detailed verbal descriptions resulted in more frequently inaccurate descriptions that in turn created inaccurate representations in the memories of participants. Inaccuracies are also likely to occur when face recognition comes immediately after verbalization. Other forms of non-verbal knowledge affected by verbal overshadowing include the following: [Verbal overshadowing] has also been observed when participants attempt to generate descriptions of other 'difficult-to-describe' stimuli such as colors (Schooler and Engstler-Schooler, 1990) or abstract figures (Brandimonte et al., 1997), or other non-visual tasks such as wine tasting (Melcher and Schooler, 1996), decision making (Wilson and Schooler, 1991), and insight problem-solving. (p. 871) (Schooler et al., 1993) Verbalization of stimuli leads to the disruption of non-reportable processes that are necessary for achieving insight solutions, which are distinct from language processes. Schooler, Ohlsson, and Brooks (1993) found that face recognition requires information that cannot be adequately verbalized, giving rise to difficulty in describing factors in recognition judgments. Subjects were less effective in solving insight problems when compelled to put their thoughts in words, which suggests that language may interfere with thought. The verbal overshadowing effect was not seen when participants engaged in articulatory suppression. Performance was reduced in both the verbal and non-verbal description conditions. This is evidence that verbal encoding plays a role in face recognition. By testing with distracting faces presented between study and test, Lloyd-Jones and Brown (2008) suggested a dual-process approach to recognition memory took place, that verbalization influenced familiarity-based processes at first, but its effects were later seen on recollection, when discrimination between items became more difficult. == Verbal overshadowing in facial recognition == The verbal overshadowing effect can be found for facial recognition because faces are predominately processed in a holistic or configurable manner. (Tanaka & Farah, 1993; Tanaka & Sengco, 1997) Verbalizing one's memory for a face is done using a featural or analytic strategy, leading to a drift from the configurable information about the face and to impaired recognition performance. However, Fallshore & Schooler (1995) found that the verbal overshadowing effect was not found when participants described faces of races different from their own. A study by Brown and Lloyd-Jones (2003) found that there was no verbal overshadowing effect found in car descriptions; it was only seen in facial descriptions. The authors noted that descriptions were no different on any measure including accuracy. It is suggested that less expertise in verbalizing faces rather than cars invokes a stronger shift in verbal and featural processing. This supports the concept of a transfer inappropriate retrieval framework and addresses some limitations of the effect. Wickham and Swift (2006) suggested that the verbal overshadowing effect is not seen in describing all faces, and one aspect that determines this is distinctiveness. Results showed that typical faces produce verbal overshadowing, while distinctive faces did not. In studies of eyewitness reports, variation in response criteria given by participants influenced the quality of the descriptions generated and accuracy on identification task, known as the retrieval-based effect. Face recognition was also impaired when subjects described a familiar face, such as a parent, or when describing a previously seen but novel face. Dodson, Johnson, and Schooler (1997) found that recognition was also impaired when participants were provided with a description of a previously seen face, and they were able to ignore provided versus self-generated descriptions more easily. This finding of verbal overshadowing suggested that eyewitness recognition is not only affected by their own descriptions, but of descriptions heard from others, such other eyewitness testimonies. == Voice recognition == The verbal overshadowing effect has also been found to affect voice identification. Research shows that describing a non-verbal stimuli leads to a decrease in recognition accuracy. In an unpublished study by Schooler, Fiore, Melcher, and Ambadar (1996), participants listened to a tape-recorded voice, after which they were asked either to verbally describe it or to not do so, and then asked to distinguish the voice from 3 similar distractor voices. The results showed that verbal overshadowing impaired accuracy of recognition based on gut feeling, suggesting an overall verbal overshadowing for voice recognition. Due to the forensic relevance of voices heard over the telephone and harassing phone calls that are often a problem for police, Perfect, Hunt, and Harris (2002) examined the influence of three factors on accuracy and confidence in voice recognition from a line-up. They expected to find an effect, because voice represents a class of stimuli that is difficult to describe verbally. This meets Schooler et al.'s (1997) modality mismatch criterion, meaning that describing the speakers age, gender, or accent is difficult, making voice recognition susceptible to the verbal overshadowing phenomenon. It was found that the method of memory encoding had no impact on performance, and that hearing a telephone voice reduced confidence but did not affect accuracy. They also found that providing a verbal description impaired accuracy but had no effect on confidence. The data showed an effect of verbal overshadowing in voice recognition and provided yet another disassociation between confidence and performance. Although there was a difference in confidence level, witnesses were able to identify voices over the telephone as accurately as voices heard direc
Read more →
Persian Speech Corpus

The Persian Speech Corpus is a Modern Persian speech corpus for speech synthesis. The corpus contains phonetic and orthographic transcriptions of about 2.5 hours of Persian speech aligned with recorded speech on the phoneme level, including annotations of word boundaries. Previous spoken corpora of Persian include FARSDAT, which consists of read aloud speech from newspaper texts from 100 Persian speakers and the Telephone FARsi Spoken language DATabase (TFARSDAT) which comprises seven hours of read and spontaneous speech produced by 60 native speakers of Persian from ten regions of Iran. The Persian Speech Corpus was built using the same methodologies laid out in the doctoral project on Modern Standard Arabic of Nawar Halabi at the University of Southampton. The work was funded by MicroLinkPC, who own an exclusive license to commercialise the corpus, though the corpus is available for non-commercial use through the corpus' website. It is distributed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. The corpus was built for speech synthesis purposes, but has been used for building HMM based voices in Persian. It can also be used to automatically align other speech corpora with their phonetic transcript and could be used as part of a larger corpus for training speech recognition systems. == Contents == The corpus is downloadable from its website, and contains the following: 396 .wav files containing spoken utterances 396 .lab files containing text utterances 396 .TextGrid files containing the phoneme labels with time stamps of the boundaries where these occur in the .wav files. phonetic-transcript.txt which has the form "[wav_filename]" "[Phoneme Sequence]" in every line orthographic-transcript.txt which has the form "[wav_filename]" "[Orthographic Transcript]" in every line
Read more →
PVLV

The primary value learned value (PVLV) model is a possible explanation for the reward-predictive firing properties of dopamine (DA) neurons. It simulates behavioral and neural data on Pavlovian conditioning and the midbrain dopaminergic neurons that fire in proportion to unexpected rewards. It is an alternative to the temporal-differences (TD) algorithm. It is used as part of Leabra.
Read more →
Transkribus

Transkribus is a platform for the text recognition, image analysis and structure recognition of historical documents. The platform was created in the context of the two EU projects "tranScriptorium" (2013–2015) and "READ" (Recognition and Enrichment of Archival Documents – 2016–2019). It was developed by the University of Innsbruck. Since July 1, 2019 the platform has been directed and further developed by the READ-COOP, a non-profit cooperative. The platform integrates tools developed by research groups throughout Europe, including the Pattern Recognition and Human Language Technology (PRHLT) group of the Technical University of Valencia and the Computational Intelligence Technology Lab (CITlab) group of University of Rostock. Comparable programs that offer similar functions are eScriptorium and OCR4All.
Read more →
BERT (language model)

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

In statistics, machine learning and algorithms, a tensor sketch is a type of dimensionality reduction that is particularly efficient when applied to vectors that have tensor structure. Such a sketch can be used to speed up explicit kernel methods, bilinear pooling in neural networks and is a cornerstone in many numerical linear algebra algorithms. == Mathematical definition == Mathematically, a dimensionality reduction or sketching matrix is a matrix M ∈ R k × d {\displaystyle M\in \mathbb {R} ^{k\times d}} , where k < d {\displaystyle k Read more →
Information gain (decision tree)

In the context of decision trees in information theory and machine learning, information gain refers to the conditional expected value of the Kullback–Leibler divergence of the univariate probability distribution of one variable from the conditional distribution of this variable given the other one. (In broader contexts, information gain can also be used as a synonym for either Kullback–Leibler divergence or mutual information, but the focus of this article is on the more narrow meaning below.) Explicitly, the information gain of a random variable X {\displaystyle X} obtained from an observation of a random variable A {\displaystyle A} taking value a {\displaystyle a} is defined as: I G ( X , a ) = D KL ( P X ∣ a ∥ P X ) {\displaystyle {\mathit {IG}}(X,a)=D_{\text{KL}}{\bigl (}P_{X\mid a}\parallel P_{X}{\bigr )}} In other words, it is the Kullback–Leibler divergence of P X ( x ) {\displaystyle P_{X}(x)} (the prior distribution for X {\displaystyle X} ) from P X ∣ a ( x ) {\displaystyle P_{X\mid a}(x)} (the posterior distribution for X {\displaystyle X} given A = a {\displaystyle A=a} ). The expected value of the information gain is the mutual information I ( X ; A ) {\displaystyle I(X;A)} : E A ⁡ [ I G ( X , A ) ] = I ( X ; A ) {\displaystyle \operatorname {E} _{A}[{\mathit {IG}}(X,A)]=I(X;A)} i.e. the reduction in the entropy of X {\displaystyle X} achieved by learning the state of the random variable A {\displaystyle A} . In machine learning, this concept can be used to define a preferred sequence of attributes to investigate to most rapidly narrow down the state of X. Such a sequence (which depends on the outcome of the investigation of previous attributes at each stage) is called a decision tree, and when applied in the area of machine learning is known as decision tree learning. Usually an attribute with high mutual information should be preferred to other attributes. == General definition == In general terms, the expected information gain is the reduction in information entropy Η from a prior state to a state that takes some information as given: I G ( T , a ) = H ( T ) − H ( T | a ) , {\displaystyle IG(T,a)=\mathrm {H} {(T)}-\mathrm {H} {(T|a)},} where H ( T | a ) {\displaystyle \mathrm {H} {(T|a)}} is the conditional entropy of T {\displaystyle T} given the value of attribute a {\displaystyle a} . This is intuitively plausible when interpreting entropy Η as a measure of uncertainty of a random variable T {\displaystyle T} : by learning (or assuming) a {\displaystyle a} about T {\displaystyle T} , our uncertainty about T {\displaystyle T} is reduced (i.e. I G ( T , a ) {\displaystyle IG(T,a)} is positive), unless of course T {\displaystyle T} is independent of a {\displaystyle a} , in which case H ( T | a ) = H ( T ) {\displaystyle \mathrm {H} (T|a)=\mathrm {H} (T)} , meaning I G ( T , a ) = 0 {\displaystyle IG(T,a)=0} . == Formal definition == Let T denote a set of training examples, each of the form ( x , y ) = ( x 1 , x 2 , x 3 , . . . , x k , y ) {\displaystyle ({\textbf {x}},y)=(x_{1},x_{2},x_{3},...,x_{k},y)} where x a ∈ v a l s ( a ) {\displaystyle x_{a}\in \mathrm {vals} (a)} is the value of the a th {\displaystyle a^{\text{th}}} attribute or feature of example x {\displaystyle {\textbf {x}}} and y is the corresponding class label. The information gain for an attribute a is defined in terms of Shannon entropy H ( − ) {\displaystyle \mathrm {H} (-)} as follows. For a value v taken by attribute a, let S a ( v ) = { x ∈ T | x a = v } {\displaystyle S_{a}{(v)}=\{{\textbf {x}}\in T|x_{a}=v\}} be defined as the set of training inputs of T for which attribute a is equal to v. Then the information gain of T for attribute a is the difference between the a priori Shannon entropy H ( T ) {\displaystyle \mathrm {H} (T)} of the training set and the conditional entropy H ( T | a ) {\displaystyle \mathrm {H} {(T|a)}} . H ( T | a ) = ∑ v ∈ v a l s ( a ) | S a ( v ) | | T | ⋅ H ( S a ( v ) ) . {\displaystyle \mathrm {H} (T|a)=\sum _{v\in \mathrm {vals} (a)}{{\frac {|S_{a}{(v)}|}{|T|}}\cdot \mathrm {H} \left(S_{a}{\left(v\right)}\right)}.} I G ( T , a ) = H ( T ) − H ( T | a ) {\displaystyle IG(T,a)=\mathrm {H} (T)-\mathrm {H} (T|a)} The mutual information is equal to the total entropy for an attribute if for each of the attribute values a unique classification can be made for the result attribute. In this case, the relative entropies subtracted from the total entropy are 0. In particular, the values v ∈ v a l s ( a ) {\displaystyle v\in vals(a)} defines a partition of the training set data T into mutually exclusive and all-inclusive subsets, inducing a categorical probability distribution P a ( v ) {\textstyle P_{a}{(v)}} on the values v ∈ v a l s ( a ) {\textstyle v\in vals(a)} of attribute a. The distribution is given P a ( v ) := | S a ( v ) | | T | {\textstyle P_{a}{(v)}:={\frac {|S_{a}{(v)}|}{|T|}}} . In this representation, the information gain of T given a can be defined as the difference between the unconditional Shannon entropy of T and the expected entropy of T conditioned on a, where the expectation value is taken with respect to the induced distribution on the values of a. I G ( T , a ) = H ( T ) − ∑ v ∈ v a l s ( a ) P a ( v ) H ( S a ( v ) ) = H ( T ) − E P a [ H ( S a ( v ) ) ] = H ( T ) − H ( T | a ) . {\displaystyle {\begin{alignedat}{2}IG(T,a)&=\mathrm {H} (T)-\sum _{v\in \mathrm {vals} (a)}{P_{a}{(v)}\mathrm {H} \left(S_{a}{(v)}\right)}\\&=\mathrm {H} (T)-\mathbb {E} _{P_{a}}{\left[\mathrm {H} {(S_{a}{(v)})}\right]}\\&=\mathrm {H} (T)-\mathrm {H} {(T|a)}.\end{alignedat}}} == Example == In engineering applications, information is analogous to signal, and entropy is analogous to noise. It determines how a decision tree chooses to split data. The leftmost figure below is very impure and has high entropy corresponding to higher disorder and lower information value. As we go to the right, the entropy decreases, and the information value increases. Now, it is clear that information gain is the measure of how much information a feature provides about a class. Let's visualize information gain in a decision tree as shown in the right: The node t is the parent node, and the sub-nodes tL and tR are child nodes. In this case, the parent node t has a collection of cancer and non-cancer samples denoted as C and NC respectively. We can use information gain to determine how good the splitting of nodes is in a decision tree. In terms of entropy, information gain is defined as: To understand this idea, let's start by an example in which we create a simple dataset and want to see if gene mutations could be related to patients with cancer. Given four different gene mutations, as well as seven samples, the training set for a decision can be created as follows: In this dataset, a 1 means the sample has the mutation (True), while a 0 means the sample does not (False). A sample with C denotes that it has been confirmed to be cancerous, while NC means it is non-cancerous. Using this data, a decision tree can be created with information gain used to determine the candidate splits for each node. For the next step, the entropy at parent node t of the above simple decision tree is computed as:H(t) = −[pC,t log2(pC,t) + pNC,t log2(pNC,t)] where, probability of selecting a class ‘C’ sample at node t, pC,t = n(t, C) / n(t), probability of selecting a class ‘NC’ sample at node t, pNC,t = n(t, NC) / n(t), n(t), n(t, C), and n(t, NC) are the number of total samples, ‘C’ samples and ‘NC’ samples at node t respectively.Using this with the example training set, the process for finding information gain beginning with H ( t ) {\displaystyle \mathrm {H} {(t)}} for Mutation 1 is as follows: pC, t = 4/7 pNC, t = 3/7 H ( t ) {\displaystyle \mathrm {H} {(t)}} = −(4/7 × log2(4/7) + 3/7 × log2(3/7)) = 0.985 Note: H ( t ) {\displaystyle \mathrm {H} {(t)}} will be the same for all mutations at the root. The relatively high value of entropy H ( t ) = 0.985 {\displaystyle \mathrm {H} {(t)}=0.985} (1 is the optimal value) suggests that the root node is highly impure and the constituents of the input at the root node would look like the leftmost figure in the above Entropy Diagram. However, such a set of data is good for learning the attributes of the mutations used to split the node. At a certain node, when the homogeneity of the constituents of the input occurs (as shown in the rightmost figure in the above Entropy Diagram), the dataset would no longer be good for learning. Moving on, the entropy at left and right child nodes of the above decision tree is computed using the formulae:H(tL) = −[pC,L log2(pC,L) + pNC,L log2(pNC,L)]H(tR) = −[pC,R log2(pC,R) + pNC,R log2(pNC,R)]where, probability of selecting a class ‘C’ sample at the left child node, pC,L = n(tL, C) / n(tL), probability of selecting a class ‘NC’ sample at the left child node, pNC,L = n(tL, NC) / n(tL), probability of selecting a class ‘C’ sample at the right child node, pC,R = n(tR, C) / n(tR), prob
Read more →
Sample exclusion dimension

In computational learning theory, sample exclusion dimensions arise in the study of exact concept learning with queries. In algorithmic learning theory, a concept over a domain X is a Boolean function over X. Here we only consider finite domains. A partial approximation S of a concept c is a Boolean function over Y ⊆ X {\displaystyle Y\subseteq X} such that c is an extension to S. Let C be a class of concepts and c be a concept (not necessarily in C). Then a specifying set for c w.r.t. C, denoted by S is a partial approximation S of c such that C contains at most one extension to S. If we have observed a specifying set for some concept w.r.t. C, then we have enough information to verify a concept in C with at most one more mind change. The exclusion dimension, denoted by XD(C), of a concept class is the maximum of the size of the minimum specifying set of c' with respect to C, where c' is a concept not in C.
Read more →
Meesho

Meesho Limited (short for Meri shop, transl. My shop) is an Indian e-commerce company, headquartered in Bengaluru. Founded by Vidit Aatrey and Sanjeev Barnwal in December 2015, Meesho is an online marketplace in categories such as fashion, home and kitchen, beauty and personal care, electronics accessories, and daily use products. == History == Meesho Private Limited, formerly Fashnear Technologies Private Limited, was established by IIT Delhi graduates Vidit Aatrey and Sanjeev Barnwal in December, 2015 In 2016, the founders came up with the idea of re-establishing the platform as Meesho, one that would enable country-wide shipping for resellers with the use of social media sites as tools for marketing. In February 2019, the platform reported having around 209,000 users and about 1.2 million monthly orders, and in March 2020, it reported approximately 563,000 users and 3.1 million monthly orders. In 2021, the Meesho mobile application was ranked among the most downloaded shopping apps globally. In 2022, Meesho had about 120 million monthly users and about 910 million orders were made through the platform, with a gross merchandise value (GMV) of about $5 billion. According to report as of August 2023 Meesho delisted 42 lakh counterfeit listings and 10 lakh restricted products under its initiative Project Suraksha. During the same period, the platform blocked access for over 12,000 user accounts flagged for policy violations. The Court granted injunctive relief by directing domain registrars to suspend the infringing websites. Additionally, the Court ordered law enforcement authorities to initiate criminal investigations, freeze associated financial accounts against the identified offenders. In 2023, Meesho became the fastest shopping app to cross over 500 million downloads. In 2024, Meesho introduced Valmo, a logistics marketplace, to provide shipment services to sellers by aggregating multiple logistics providers. Meesho employs over 3,000 small businesses and 10-12 large firms for warehousing and sorting operations within its logistics framework. According to media reports, Valmo operating in approximately 15,000 pincodes in India with around 6,000 partners. It is reported to handle over 50% of Meesho's daily orders. In November 2024, Meesho introduced a generative AI-powered voice bot for customer support, managing approximately 60,000 calls daily in English and Hindi. According to media reports, the system resolves the majority of queries without human assistance, with only a small fraction of calls requiring manual intervention. According to media reports, in 2024, Meesho prevented over 22 million suspicious or potentially fraudulent transactions on its platform. The company initiated legal proceedings, resulting in the filing of twelve cases, including nine specifically targeting over forty individuals in the cities of Kolkata and Ranchi. The company filed a suit in the Delhi High Court for a permanent injunction against parties operating deceptive websites misappropriating its brand identity. Meesha went public through an initial public offering in December 2025, raising $603 million. It is listed on both the BSE and NSE. == Recognition == In 2023, Meesho was named one of the most influential companies of the year by Time (magazine).
Read more →
Common Voice

Common Voice is a crowdsourcing project started by Mozilla to create a free and open speech corpus. The project is supported by volunteers who record sample sentences with a microphone and review recordings of other users. The transcribed sentences are collected in a voice database available under the public domain license CC0. This license ensures that developers can use the database for voice-to-text and text-to-voice applications without restrictions or costs. == Aims == Common Voice aims to provide diverse voice samples. According to Mozilla's Katharina Borchert, many existing projects took datasets from public radio or otherwise had datasets that underrepresented both women and people with pronounced accents. == Voice database == The first dataset was released in November 2017. More than 20,000 users worldwide had recorded 500 hours of English sentences. In February 2019, the first batch of languages was released for use. This included 18 languages such as English, French, German and Mandarin Chinese, but also less prevalent languages like Welsh and Kabyle. In total, this included almost 1,400 hours of recorded voice data from more than 42,000 contributors. By July 2020 the database had amassed 7,226 hours of voice recordings in 54 languages, 5,591 hours of which had been verified by volunteers. In May 2021, following the work to add Kinyarwanda, the project received a grant to add Kiswahili. At the beginning of 2022, Bengali.AI partnered with Common Voice to launch the "Bangla Speech Recognition" project that aims to make machines understand the Bangla language. 2000 hours of voice was collected. In September 2022, it was announced that the Twi language of Ghana was the 100th language to be added to the database. As of December 2025, Mozilla Common Voice collects voice data for over 250 languages, with the most hours having been collected in English, Catalan, Kinyarwanda, Belarusian and Esperanto.
Read more →
Iris flower data set

The Iris flower data set or Fisher's Iris data set is a multivariate data set used and made famous by the British statistician and biologist Ronald Fisher in his 1936 paper The use of multiple measurements in taxonomic problems as an example of linear discriminant analysis. It is sometimes called Anderson's Iris data set because Edgar Anderson collected the data to quantify the morphologic variation of Iris flowers of three related species. Two of the three species were collected in the Gaspé Peninsula "all from the same pasture, and picked on the same day and measured at the same time by the same person with the same apparatus". The data set consists of 50 samples from each of three species of Iris (Iris setosa, Iris virginica and Iris versicolor). Four features were measured from each sample: the length and the width of the sepals and petals, in centimeters. Based on the combination of these four features, Fisher developed a linear discriminant model to distinguish each species. Fisher's paper was published in the Annals of Eugenics (today the Annals of Human Genetics). == Use of the data set == Originally used as an example data set on which Fisher's linear discriminant analysis was applied, it became a typical test case for many statistical classification techniques in machine learning such as support vector machines. The use of this data set in cluster analysis however is not common, since the data set only contains two clusters with rather obvious separation. One of the clusters contains Iris setosa, while the other cluster contains both Iris virginica and Iris versicolor and is not separable without the species information Fisher used. This makes the data set a good example to explain the difference between supervised and unsupervised techniques in data mining: Fisher's linear discriminant model can only be obtained when the object species are known: class labels and clusters are not necessarily the same. Nevertheless, all three species of Iris are separable in the projection on the nonlinear and branching principal component. The data set is approximated by the closest tree with some penalty for the excessive number of nodes, bending and stretching. Then the so-called "metro map" is constructed. The data points are projected into the closest node. For each node the pie diagram of the projected points is prepared. The area of the pie is proportional to the number of the projected points. It is clear from the diagram (left) that the absolute majority of the samples of the different Iris species belong to the different nodes. Only a small fraction of Iris-virginica is mixed with Iris-versicolor (the mixed blue-green nodes in the diagram). Therefore, the three species of Iris (Iris setosa, Iris virginica and Iris versicolor) are separable by the unsupervising procedures of nonlinear principal component analysis. To discriminate them, it is sufficient just to select the corresponding nodes on the principal tree. == Data set == The data set contains a set of 150 records under five attributes: sepal length, sepal width, petal length, petal width and species. The iris data set is widely used as a beginner's data set for machine learning purposes. The data set is included in R base and Python in the machine learning library scikit-learn, so that users can access it without having to find a source for it. Several versions of the data set have been published. === R code illustrating usage === The example R code shown below reproduce the scatterplot displayed at the top of this article: === Python code illustrating usage === This code gives:
Read more →