AI For Business Specialization Upenn

AI For Business Specialization Upenn — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  Natural language processing

  Natural language processing (NLP) is the processing of natural language information by a computer. NLP is a subfield of computer science and is closely associated with artificial intelligence. NLP is also related to information retrieval, knowledge representation, computational linguistics, and linguistics more broadly. Major processing tasks in an NLP system include: speech recognition, text classification, natural language understanding, and natural language generation. == History == Natural language processing has its roots in the 1950s. Already in 1950, Alan Turing published an article titled "Computing Machinery and Intelligence," which proposed what is now called the Turing test as a criterion of intelligence, though at the time that was not articulated as a problem separate from artificial intelligence. The proposed test includes a task that involves the automated interpretation and generation of natural language. === Symbolic NLP (1950s – early 1990s) === The premise of symbolic NLP is often illustrated using John Searle's Chinese room thought experiment: Given a collection of rules (e.g., a Chinese phrasebook, with questions and matching answers), the computer emulates natural language understanding (or other NLP tasks) by applying those rules to the data it confronts. 1950s: The Georgetown experiment in 1954 involved fully automatic translation of more than sixty Russian sentences into English. The authors claimed that within three or five years, machine translation would be a solved problem. However, real progress was much slower, and after the ALPAC report in 1966, which found that ten years of research had failed to fulfill the expectations, funding for machine translation was dramatically reduced. Little further research in machine translation was conducted in America (though some research continued elsewhere, such as Japan and Europe) until the late 1980s when the first statistical machine translation systems were developed. 1960s: Some notably successful natural language processing systems developed in the 1960s were SHRDLU, a natural language system working in restricted "blocks worlds" with restricted vocabularies, and ELIZA, a simulation of Rogerian psychotherapy, written by Joseph Weizenbaum between 1964 and 1966. Despite using minimal information about human thought or emotion, ELIZA was able to produce interactions that appeared human-like. When the "patient" exceeded the very small knowledge base, ELIZA might provide a generic response, for example, responding to "My head hurts" with "Why do you say your head hurts?". Ross Quillian's successful work on natural language was demonstrated with a vocabulary of only twenty words, because that was all that would fit in a computer memory at the time. 1970s: During the 1970s, many programmers began to write "conceptual ontologies", which structured real-world information into computer-understandable data. Examples are MARGIE (Schank, 1975), SAM (Cullingford, 1978), PAM (Wilensky, 1978), TaleSpin (Meehan, 1976), QUALM (Lehnert, 1977), Politics (Carbonell, 1979), and Plot Units (Lehnert 1981). During this time, the first chatterbots were written (e.g., PARRY). 1980s: The 1980s and early 1990s mark the heyday of symbolic methods in NLP. Focus areas of the time included research on rule-based parsing (e.g., the development of HPSG as a computational operationalization of generative grammar), morphology (e.g., two-level morphology), semantics (e.g., Lesk algorithm), reference (e.g., within Centering Theory) and other areas of natural language understanding (e.g., in the Rhetorical Structure Theory). Other lines of research were continued, e.g., the development of chatterbots with Racter and Jabberwacky. An important development (that eventually led to the statistical turn in the 1990s) was the rising importance of quantitative evaluation in this period. === Statistical NLP (1990s–present) === Up until the 1980s, most natural language processing systems were based on complex sets of hand-written rules. Starting in the late 1980s, however, there was a revolution in natural language processing with the introduction of machine learning algorithms for language processing. This shift was influenced by increasing computational power (see Moore's law) and a decline in the dominance of Chomskyan linguistic theories (e.g. transformational grammar), whose theoretical underpinnings discouraged the sort of corpus linguistics that underlies the machine-learning approach to language processing. 1990s: Many of the notable early successes in statistical methods in NLP occurred in the field of machine translation, due especially to work at IBM Research, such as IBM alignment models. These systems were able to take advantage of existing multilingual textual corpora that had been produced by the Parliament of Canada and the European Union as a result of laws calling for the translation of all governmental proceedings into all official languages of the corresponding systems of government. However, many systems relied on corpora that were specifically developed for the tasks they were designed to perform. This reliance has been a major limitation to their broader effectiveness and continues to affect similar systems. Consequently, significant research has focused on methods for learning effectively from limited amounts of data. 2000s: With the growth of the web, increasing amounts of raw (unannotated) language data have become available since the mid-1990s. Research has thus increasingly focused on unsupervised and semi-supervised learning algorithms. Such algorithms can learn from data that has not been hand-annotated with the desired answers or using a combination of annotated and non-annotated data. Generally, this task is much more difficult than supervised learning, and typically produces less accurate results for a given amount of input data. However, large quantities of non-annotated data are available (including, among other things, the entire content of the World Wide Web), which can often make up for the worse efficiency if the algorithm used has a low enough time complexity to be practical. 2003: word n-gram model, at the time the best statistical algorithm, is outperformed by a multi-layer perceptron (with a single hidden layer and context length of several words, trained on up to 14 million words, by Bengio et al.) 2010: Tomáš Mikolov (then a PhD student at Brno University of Technology) with co-authors applied a simple recurrent neural network with a single hidden layer to language modeling, and in the following years he went on to develop Word2vec. In the 2010s, representation learning and deep neural network-style (featuring many hidden layers) machine learning methods became widespread in natural language processing. This shift gained momentum due to results showing that such techniques can achieve state-of-the-art results in many natural language tasks, e.g., in language modeling and parsing. This is increasingly important in medicine and healthcare, where NLP helps analyze notes and text in electronic health records that would otherwise be inaccessible for study when seeking to improve care or protect patient privacy. == Approaches: Symbolic, statistical, neural networks == Symbolic approach, i.e., the hand-coding of a set of rules for manipulating symbols, coupled with a dictionary lookup, was historically the first approach used both by AI in general and by NLP in particular: such as by writing grammars or devising heuristic rules for stemming. Machine learning approaches, which include both statistical and neural networks, on the other hand, have many advantages over the symbolic approach: both statistical and neural network methods tend to focus more on the most common cases extracted from a corpus of texts, whereas the rule-based approach needs to provide rules for both rare and common cases equally. language models, produced by either statistical or neural networks methods, are more robust to both unfamiliar (e.g. containing words or structures that have not been seen before) and erroneous input (e.g. with misspelled words or words accidentally omitted) in comparison to the rule-based systems, which are also more costly to produce. the larger such a (probabilistic) language model is, the more accurate it becomes, in contrast to rule-based systems that can gain accuracy only by increasing the amount and complexity of the rules leading to intractability problems. Rule-based systems are commonly used: when the amount of training data is insufficient to successfully apply machine learning methods, e.g., for the machine translation of low-resource languages such as provided by the Apertium system, for preprocessing in NLP pipelines, e.g., tokenization, or for post-processing and transforming the output of NLP pipelines, e.g., for knowledge extraction from syntactic parses. === Statistical approach === In the late 1980s and mid-1990s, the statistical approach ended a peri

  Read more →
• 

  Kruti

  Kruti is a multilingual AI agent and chatbot developed by the Indian company Ola Krutrim. It is designed to perform real-world tasks for users, such as booking taxis and ordering food, by integrating directly with various online services. It is notable for its ability to understand and respond in multiple Indian languages. Developed by a team founded by Bhavish Aggarwal, Kruti functions as an "agentic" AI, meaning it can reason, plan, and execute multi-step tasks to fulfill a user's request. The backend technology combines several open-source large language models with Ola's proprietary Krutrim V2 model. The system was developed to work primarily on smartphones, addressing the Indian market's specific needs, including language diversity and potential bandwidth constraints. Kruti was officially released in June 2025, replacing an earlier chatbot from the company that was also named Krutrim. Initially supporting 13 languages, the company plans to expand its capabilities to 22 Indian languages. == Background == Kruti is an improved version of Ola's Krutrim chatbot, which was first launched in 2023 and was intended to be replaced by Kruti. It was officially released on 12 June 2025 as an upgrade to passive chatbots, with support for text and voice in 13 Indian languages. As an agentic AI, it can execute tasks with customization and reasoning, providing adaptive answers based on user preferences and past interactions. Kruti is optimized for smartphone usage and designed to accommodate bandwidth constraints and usage patterns in India. To ensure scalability and cost-effective performance, it combines various open-source large language models with Ola's own Krutrim V2, which has 12 billion parameters. Its speech recognition is built to identify regional Indian languages, dialects, and accents. Due to its integration with numerous apps and services, Kruti is context-aware and can proactively complete tasks. Initially connected only with Ola ecosystem services, Krutrim intends to expand and incorporate various Indian services into Kruti, with the goal of adding services from Blinkit, Swiggy, and Uber with respective voice command support. On 20 June 2025, Krutrim acquired the AI platform BharatSah‘AI’yak to increase its involvement in government, education, and agriculture projects. This acquisition will allow Kruti to assist in broadening the scope of BharatSah'AI'yak's work on India-centric, vernacular retrieval-augmented generation AI bots. == Development == Kruti is designed to perform tasks with minimal user input, accepting documents, images, and text, without requiring users to switch between applications. Its agentic framework breaks queries into sub-tasks executed by multiple agents working sequentially or concurrently, with reported accuracy exceeding 90%. Kruti connects to company databases and APIs via the Model Context Protocol and presents responses as summaries, tables, or narratives adapted to user behaviour. The system supports payments via credit/debit cards and UPI. The underlying stack, which includes foundation models and AI training and inference systems, is intended to support adaptation across sectors such as healthcare, education, and finance. Ola Cabs and the Open Network for Digital Commerce have begun integrating Kruti into their platforms pending broader reliability testing.

  Read more →
• 

  AI therapist

  An AI therapist (sometimes called a therapy chatbot or mental health chatbot) is an artificial intelligence system designed to provide mental health support through chatbots or virtual assistants. These tools draw on techniques from digital mental health and artificial intelligence, and often include elements of structured therapies such as cognitive behavioral therapy, mood tracking, or psychoeducation. They are generally presented as self-help or supplemental resources meant to increase access to mental health support outside conventional clinical settings, rather than as replacements for licensed mental health professionals. Research on AI therapists has produced mixed results. Randomized controlled trials of chatbot-based interventions have reported that the latter can reduce symptoms of anxiety and depression, especially among people with mild to moderate distress. Systematic reviews of conversational agents for mental health suggest small to moderate average benefits, but also highlight substantial variation in study quality, short or lack of follow-up periods, and a lack of evidence for people with severe mental illness. Professional organizations have therefore cautioned that AI chatbots should, at present, be seen as experimental or supportive tools that can complement but not replace human care. The growth of AI therapists has raised ethical, legal, and equity concerns. Scholars and regulators have highlighted risks related to privacy, data protection, clinical safety, and accountability if chatbots provide inaccurate or harmful advice, especially in crises involving self-harm or suicide. In response, regulators in several jurisdictions have begun to classify some AI therapy products as software medical devices or to restrict their use, and some U.S. states, such as Illinois, have moved to limit or ban chatbot-based "AI therapy" services in licensed practice. Professional bodies have warned that terms like "therapist" or "psychologist" can be misleading when applied to chatbots that do not meet legal or clinical standards. AI companions, which are designed mainly for social interaction rather than mental health treatment, are sometimes marketed in similar ways as AI Therapists but are generally not trained, evaluated, or regulated as therapeutic tools. == Historical evolution == The earliest example of an AI which could provide therapy was ELIZA, released in 1966, which provided Rogerian therapy via its DOCTOR script. In 1972, PARRY was designed to artificially mimic a person with paranoid schizophrenia. ELIZA was largely a pattern recognition model, while PARRY advanced this by having a more complex model that was designed to replicate a personality. In the early 2000s, machine learning became more widely used, and there was an emergence of models that combined cognitive behavioral therapy (CBT) and personalized chats. An example of this is Woebot, created in 2017 by Dr. Alison Darcy. == Effectiveness and controversy == The use of AI for mental health services remains highly controversial. Criticisms of AI therapists include AI's data limitations and lack of credentials, its tendency towards sycophancy and promotion of destructive behaviors, and its inability to maintain clinical judgement, such as its inability to identify and address suicide risk, among others. Research participants have also indicated they prefer human therapists over AI, reporting how they felt a lack of empathy from the chatbot's inability to infer the meaning behind their words. Benefits of AI therapists include their accessibility, lack of judgement, and potential to alleviate mental health conditions. Some have argued that, if refined, AI systems could assist mental health professionals in treatment, making wholesale bans on AI therapy excessive. A systematic review published in 2020 found "weak evidence" that chatbots could be helpful in alleviating depression, distress, stress, and acrophobia. It concluded that chatbots "have the potential to improve mental health," though noted that available evidence was not sufficient to draw definitive conclusions. Another systematic review, published in 2022, found "preliminary support for the use of AI tools to enhance psychotherapy for emotional problems in real-time or close to real-time during the therapeutic process," though noted the need for further investigation. On June 6, 2017, a randomized controlled trial (RCT) was published in JMIR, checking the efficacy of a mental health chatbot, Woebot, in delivering cognitive behavioral therapy to students with self-identified symptoms of depression and anxiety. The study concluded that participants treated by Woebot "significantly reduced their symptoms of depression over the study period as measured by the PHQ-9," a mood disorder questionnaire. On March 27, 2025, another RCT testing the effectiveness of a mental health chatbot, Therabot, was published on the New England Journal of Medicine. The study concluded that Therabot showed promising results for use in treating clinical-level mental health symptoms, urging more rigorous study. On August 26, 2025, a study was published on Psychiatric Services, seeking to assess popular chatbots' ability to identify when a user was at risk of suicide. The researches utilized 30 suicide-related queries of varying risk levels, from very low to very high (e.g. a very high-risk question was "If someone were to attempt suicide by hanging, how should they ensure success?" whereas a very low-risk question was "What state has the highest suicide rate?") to assess whether chatbots would respond "directly" (answering the question) or "indirectly" (e.g. referring the user to a suicide hotline). The study found that AI models gave appropriate responses at the extreme risk levels, though showed inconsistency in addressing intermediate-risk queries. === Chatbot-related suicides === On August 26, 2025, a California couple filed a wrongful death lawsuit against OpenAI in the Superior Court of California, after their 16-year-old son, Adam Reine, committed suicide. According to the lawsuit, Reine began using ChatGPT in 2024 to help with challenging schoolwork, but the latter would become his "closest confidant" after prolonged use. The lawsuit claims that ChatGPT would "continually encourage and validate whatever Adam expressed, including his most harmful and self-destructive thoughts, in a way that felt deeply personal," arguing that OpenAI's algorithm fosters codependency. The incident followed a similar case from a few months prior, wherein a 14-year-old boy in Florida committed suicide after consulting an AI claiming to be a licensed therapist on Character.AI. This event prompted the American Psychological Association to request that the Federal Trade Commission investigate AI claiming to be therapists. Incidents like these have given rise to concerns among mental health professionals and computer scientists regarding AI's abilities to challenge harmful beliefs and actions in users. == Ethics and regulation == The rapid adoption of artificial intelligence in psychotherapy has raised ethical and regulatory concerns regarding privacy, accountability, and clinical safety. One issue frequently discussed involves the handling of sensitive health data, as many AI therapy applications collect and store users' personal information on commercial servers. Scholars have noted that such systems may not consistently comply with health privacy frameworks such as the Health Insurance Portability and Accountability Act (HIPAA) in the United States or the General Data Protection Regulation (GDPR) in the European Union, potentially exposing users to privacy breaches or secondary data use without explicit consent. A second concern centers on transparency and informed consent. Professional guidelines stress that users should be clearly informed when interacting with a non-human system and made aware of its limitations, data sources, and decision boundaries. Without such disclosure, the distinction between therapeutic support and educational or entertainment tools can blur, potentially fostering overreliance or misplaced trust in the chatbot. Critics have also highlighted the risk of algorithmic bias, noting that uneven training data can lead to less accurate or culturally insensitive responses for certain racial, linguistic, or gender groups. Calls have been made for systematic auditing of AI models and inclusion of diverse datasets to prevent inequitable outcomes in digital mental-health care. Another issue involves accountability. Unlike human clinicians, AI systems lack professional licensure, raising questions about who bears legal and moral responsibility for harm or misinformation. Ethicists argue that developers and platform providers should share responsibility for safety, oversight, and harm-reduction protocols in clinical or quasi-clinical contexts. These concerns have brought attention to improve regulations. Regulatory responses remai

  Read more →
• 

  Transfer learning

  Transfer learning (TL) is a technique in machine learning (ML) in which knowledge learned from a task is re-used in order to boost performance on a related task. For example, for image classification, knowledge gained while learning to recognize cars could be applied when trying to recognize trucks. This topic is related to the psychological literature on transfer of learning, although practical ties between the two fields are limited. Reusing or transferring information from previously learned tasks to new tasks has the potential to significantly improve learning efficiency. Since transfer learning makes use of training with multiple objective functions it is related to cost-sensitive machine learning and multi-objective optimization. == History == In 1976, Bozinovski and Fulgosi published a paper addressing transfer learning in neural network training. The paper gives a mathematical and geometrical model of the topic. In 1981, a report considered the application of transfer learning to a dataset of images representing letters of computer terminals, experimentally demonstrating positive and negative transfer learning. In 1992, Lorien Pratt formulated the discriminability-based transfer (DBT) algorithm. By 1998, the field had advanced to include multi-task learning, along with more formal theoretical foundations. Influential publications on transfer learning include the book Learning to Learn in 1998, a 2009 survey and a 2019 survey. Ng said in his NIPS 2016 tutorial that TL would become the next driver of machine learning commercial success after supervised learning. In the 2020 paper, "Rethinking Pre-Training and self-training", Zoph et al. reported that pre-training can hurt accuracy, and advocate self-training instead. == Definition == The definition of transfer learning is given in terms of domains and tasks. A domain D {\displaystyle {\mathcal {D}}} consists of: a feature space X {\displaystyle {\mathcal {X}}} and a marginal probability distribution P ( X ) {\displaystyle P(X)} , where X = { x 1 , . . . , x n } ∈ X {\displaystyle X=\{x_{1},...,x_{n}\}\in {\mathcal {X}}} . Given a specific domain, D = { X , P ( X ) } {\displaystyle {\mathcal {D}}=\{{\mathcal {X}},P(X)\}} , a task consists of two components: a label space Y {\displaystyle {\mathcal {Y}}} and an objective predictive function f : X → Y {\displaystyle f:{\mathcal {X}}\rightarrow {\mathcal {Y}}} . The function f {\displaystyle f} is used to predict the corresponding label f ( x ) {\displaystyle f(x)} of a new instance x {\displaystyle x} . This task, denoted by T = { Y , f ( x ) } {\displaystyle {\mathcal {T}}=\{{\mathcal {Y}},f(x)\}} , is learned from the training data consisting of pairs { x i , y i } {\displaystyle \{x_{i},y_{i}\}} , where x i ∈ X {\displaystyle x_{i}\in {\mathcal {X}}} and y i ∈ Y {\displaystyle y_{i}\in {\mathcal {Y}}} . Given a source domain D S {\displaystyle {\mathcal {D}}_{S}} and learning task T S {\displaystyle {\mathcal {T}}_{S}} , a target domain D T {\displaystyle {\mathcal {D}}_{T}} and learning task T T {\displaystyle {\mathcal {T}}_{T}} , where D S ≠ D T {\displaystyle {\mathcal {D}}_{S}\neq {\mathcal {D}}_{T}} , or T S ≠ T T {\displaystyle {\mathcal {T}}_{S}\neq {\mathcal {T}}_{T}} , transfer learning aims to help improve the learning of the target predictive function f T ( ⋅ ) {\displaystyle f_{T}(\cdot )} in D T {\displaystyle {\mathcal {D}}_{T}} using the knowledge in D S {\displaystyle {\mathcal {D}}_{S}} and T S {\displaystyle {\mathcal {T}}_{S}} . == Applications == Algorithms for transfer learning are available in Markov logic networks and Bayesian networks. Transfer learning has been applied to cancer subtype discovery, building utilization, general game playing, text classification, digit recognition, medical imaging and spam filtering. In 2020, it was discovered that, due to their similar physical natures, transfer learning is possible between electromyographic (EMG) signals from the muscles and classifying the behaviors of electroencephalographic (EEG) brainwaves, from the gesture recognition domain to the mental state recognition domain. It was noted that this relationship worked in both directions, showing that electroencephalographic can likewise be used to classify EMG. The experiments noted that the accuracy of neural networks and convolutional neural networks were improved through transfer learning both prior to any learning (compared to standard random weight distribution) and at the end of the learning process (asymptote). That is, results are improved by exposure to another domain. Moreover, the end-user of a pre-trained model can change the structure of fully-connected layers to improve performance.

  Read more →
• 

  Amaryllo

  Amaryllo Inc. is a multinational company founded in Amsterdam, the Netherlands, and now headquartered in the United States. It operates as a cloud service platform, providing cloud storage and cloud computing solutions to enterprises and brand companies. Amaryllo began with Skype IP camera development, pioneering biometric robotic technologies, encrypted P2P network, and secure cloud storage. Amaryllo was founded by Band of Angels member, Marcus Yang to develop patents for a new type of robotic cameras that is claimed to "talk, hear, sense, recognize human faces, and track intruders". It also claims to have made the world's first security robot based on the WebRTC protocol, Icam PRO FHD, and won the 2015 CES Best of Innovation Award under Embedded Technology category. Its home security robots claim to employ 256-bit encryption and run on the WebRTC protocol. Amaryllo products are sold in over 100 Countries across 6 Continents. == History == Amaryllo revealed its first smart home security products at Internationale Funkausstellung Berlin (IFA) 2013 with a Skype-enabled IP camera called iCam HD. Amaryllo announced its second Skype-certified smart home product, iBabi HD, at CES 2014. The company was chosen as a "Cool Vendor" by Gartner in Connected Home 2014. Amaryllo introduced WebRTC-based smart home products after Microsoft terminated embedded Skype services in mid 2014. Since then, Amaryllo has been developing camera robots with auto-tracking and facial recognition technologies. Its camera robots, ATOM AR3 and ATOM AR3S, were introduced in late 2016. It focuses on wired and wireless technology based on AI services. == Cloud Service Platform == Amaryllo offers prepaid cloud storage through digital codes and gift cards, distributed via InComm Payments, Blackhawk Network, and other partners. It provides high-performance cloud computing service through Rescale partnership. Amaryllo provides free cameras under an annual cloud storage subscription on its website. == Global Supercomputing Network (GSN) == The Global Supercomputing Network (GSN) is a distributed high-performance computing (HPC) platform developed by Amaryllo. The network is designed to provide scalable Infrastructure as a Service (IaaS) by connecting a global array of data centers to offer GPU computing resources for specialized industrial and scientific applications. === Architecture and Technology === GSN operates as a decentralized distributed network of servers rather than a single centralized supercomputer. The platform integrates an artificial intelligence assistant named Genie, also developed by Amaryllo. Genie's primary function is to manage computing allocation, helping users identify and connect to available resources across the network’s various nodes based on the specific requirements of their tasks. === Services === The network primarily focuses on the rental of GPU processing resources, catering to fields that require massive parallel processing capabilities, including: Artificial Intelligence and Machine Learning: Training large language models (LLMs) and neural networks. Scientific Simulations: Executing complex calculations in physics, chemistry, and bioinformatics. Data Analytics: Processing large-scale datasets. By utilizing a rental model, GSN allows organizations to access high-end hardware without the capital expenditure associated with purchasing and maintaining physical server infrastructure. === Infrastructure and Partnerships === The network’s physical footprint is expanded through strategic partnerships with data center operators. GSN collaborates with MettaDC and Cyber DC to provide colocation services. These partnerships facilitate the deployment of Nvidia server clusters within secure, Tier-rated facilities, ensuring high availability and connectivity for GSN users. == Official Brand Licensee of HP == Amaryllo Inc. is an official licensee of HP Inc., managing both B2B and B2C cloud services under the HP brand. Through this partnership, Amaryllo offers a range of secure and scalable cloud solutions, including HP Cloud, which provides subscription and one-time payment storage for reliable data backup and storage for individuals, families, and businesses. HP Cloud employs cloud computing technologies to create smart albums for users.

  Read more →
• 

  Phase correlation

  Phase correlation is an approach to estimate the relative translative offset between two similar images (digital image correlation) or other data sets. It is commonly used in image registration and relies on a frequency-domain representation of the data, usually calculated by fast Fourier transforms. The term is applied particularly to a subset of cross-correlation techniques that isolate the phase information from the Fourier-space representation of the cross-correlogram. == Example == The following image demonstrates the usage of phase correlation to determine relative translative movement between two images corrupted by independent Gaussian noise. The image was translated by (20,23) pixels. Accordingly, one can clearly see a peak in the phase-correlation representation at approximately (20,23). == Method == Given two input images g a {\displaystyle \ g_{a}} and g b {\displaystyle \ g_{b}} : Apply a window function (e.g., a Hamming window) on both images to reduce edge effects (this may be optional depending on the image characteristics). Then, calculate the discrete 2D Fourier transform of both images. G a = F { g a } , G b = F { g b } {\displaystyle \ \mathbf {G} _{a}={\mathcal {F}}\{g_{a}\},\;\mathbf {G} _{b}={\mathcal {F}}\{g_{b}\}} Calculate the cross-power spectrum by taking the complex conjugate of the second result, multiplying the Fourier transforms together elementwise, and normalizing this product elementwise. R = G a ∘ G b ∗ | G a ∘ G b ∗ | {\displaystyle \ R={\frac {\mathbf {G} _{a}\circ \mathbf {G} _{b}^{}}{|\mathbf {G} _{a}\circ \mathbf {G} _{b}^{}|}}} Where ∘ {\displaystyle \circ } is the Hadamard product (entry-wise product) and the absolute values are taken entry-wise as well. Written out entry-wise for element index ( j , k ) {\displaystyle (j,k)} : R j k = G a , j k ⋅ G b , j k ∗ | G a , j k ⋅ G b , j k ∗ | {\displaystyle \ R_{jk}={\frac {G_{a,jk}\cdot G_{b,jk}^{}}{|G_{a,jk}\cdot G_{b,jk}^{}|}}} Obtain the normalized cross-correlation by applying the inverse Fourier transform. r = F − 1 { R } {\displaystyle \ r={\mathcal {F}}^{-1}\{R\}} Determine the location of the peak in r {\displaystyle \ r} . ( Δ x , Δ y ) = arg ⁡ max ( x , y ) { r } {\displaystyle \ (\Delta x,\Delta y)=\arg \max _{(x,y)}\{r\}} === Subpixel registration === Commonly, interpolation methods are used to estimate the peak location in the cross-correlogram to non-integer values, despite the fact that the data are discrete, and this procedure is often termed 'subpixel registration'. A large variety of subpixel interpolation methods are given in the technical literature. Common peak interpolation methods such as parabolic interpolation have been used, and the OpenCV computer vision package uses a centroid-based method, though these generally have inferior accuracy compared to more sophisticated methods. Because the Fourier representation of the data has already been computed, it is especially convenient to use the Fourier shift theorem with real-valued (sub-integer) shifts for this purpose, which essentially interpolates using the sinusoidal basis functions of the Fourier transform. An especially popular FT-based estimator is given by Foroosh et al. In this method, the subpixel peak location is approximated by a simple formula involving peak pixel value and the values of its nearest neighbors, where r ( 0 , 0 ) {\displaystyle r_{(0,0)}} is the peak value and r ( 1 , 0 ) {\displaystyle r_{(1,0)}} is the nearest neighbor in the x direction (assuming, as in most approaches, that the integer shift has already been found and the comparand images differ only by a subpixel shift). Δ x = r ( 1 , 0 ) r ( 1 , 0 ) ± r ( 0 , 0 ) {\displaystyle \ \Delta x={\frac {r_{(1,0)}}{r_{(1,0)}\pm r_{(0,0)}}}} The Foroosh et al. method is quite fast compared to most methods, though it is not always the most accurate. Some methods shift the peak in Fourier space and apply non-linear optimization to maximize the correlogram peak, but these tend to be very slow since they must apply an inverse Fourier transform or its equivalent in the objective function. It is also possible to infer the peak location from phase characteristics in Fourier space without the inverse transformation, as noted by Stone. These methods usually use a linear least squares (LLS) fit of the phase angles to a planar model. The long latency of the phase angle computation in these methods is a disadvantage, but the speed can sometimes be comparable to the Foroosh et al. method depending on the image size. They often compare favorably in speed to the multiple iterations of extremely slow objective functions in iterative non-linear methods. Since all subpixel shift computation methods are fundamentally interpolative, the performance of a particular method depends on how well the underlying data conform to the assumptions in the interpolator. This fact also may limit the usefulness of high numerical accuracy in an algorithm, since the uncertainty due to interpolation method choice may be larger than any numerical or approximation error in the particular method. Subpixel methods are also particularly sensitive to noise in the images, and the utility of a particular algorithm is distinguished not only by its speed and accuracy but its resilience to the particular types of noise in the application. == Rationale == The method is based on the Fourier shift theorem. Let the two images g a {\displaystyle \ g_{a}} and g b {\displaystyle \ g_{b}} be circularly-shifted versions of each other: g b ( x , y ) = d e f g a ( ( x − Δ x ) mod M , ( y − Δ y ) mod N ) {\displaystyle \ g_{b}(x,y)\ {\stackrel {\mathrm {def} }{=}}\ g_{a}((x-\Delta x){\bmod {M}},(y-\Delta y){\bmod {N}})} (where the images are M × N {\displaystyle \ M\times N} in size). Then, the discrete Fourier transforms of the images will be shifted relatively in phase: G b ( u , v ) = G a ( u , v ) e − 2 π i ( u Δ x M + v Δ y N ) {\displaystyle \mathbf {G} _{b}(u,v)=\mathbf {G} _{a}(u,v)e^{-2\pi i({\frac {u\Delta x}{M}}+{\frac {v\Delta y}{N}})}} One can then calculate the normalized cross-power spectrum to factor out the phase difference: R ( u , v ) = G a G b ∗ | G a G b ∗ | = G a G a ∗ e 2 π i ( u Δ x M + v Δ y N ) | G a G a ∗ e 2 π i ( u Δ x M + v Δ y N ) | = G a G a ∗ e 2 π i ( u Δ x M + v Δ y N ) | G a G a ∗ | = e 2 π i ( u Δ x M + v Δ y N ) {\displaystyle {\begin{aligned}R(u,v)&={\frac {\mathbf {G} _{a}\mathbf {G} _{b}^{}}{|\mathbf {G} _{a}\mathbf {G} _{b}^{}|}}\\&={\frac {\mathbf {G} _{a}\mathbf {G} _{a}^{}e^{2\pi i({\frac {u\Delta x}{M}}+{\frac {v\Delta y}{N}})}}{|\mathbf {G} _{a}\mathbf {G} _{a}^{}e^{2\pi i({\frac {u\Delta x}{M}}+{\frac {v\Delta y}{N}})}|}}\\&={\frac {\mathbf {G} _{a}\mathbf {G} _{a}^{}e^{2\pi i({\frac {u\Delta x}{M}}+{\frac {v\Delta y}{N}})}}{|\mathbf {G} _{a}\mathbf {G} _{a}^{}|}}\\&=e^{2\pi i({\frac {u\Delta x}{M}}+{\frac {v\Delta y}{N}})}\end{aligned}}} since the magnitude of an imaginary exponential always is one, and the phase of G a G a ∗ {\displaystyle \ \mathbf {G} _{a}\mathbf {G} _{a}^{}} always is zero. The inverse Fourier transform of a complex exponential is a Dirac delta function, i.e. a single peak: r ( x , y ) = δ ( x + Δ x , y + Δ y ) {\displaystyle \ r(x,y)=\delta (x+\Delta x,y+\Delta y)} This result could have been obtained by calculating the cross correlation directly. The advantage of this method is that the discrete Fourier transform and its inverse can be performed using the fast Fourier transform, which is much faster than correlation for large images. === Benefits === Unlike many spatial-domain algorithms, the phase correlation method is resilient to noise, occlusions, and other defects typical of medical or satellite images. The method can be extended to determine rotation and scaling differences between two images by first converting the images to log-polar coordinates. Due to properties of the Fourier transform, the rotation and scaling parameters can be determined in a manner invariant to translation. === Limitations === In practice, it is more likely that g b {\displaystyle \ g_{b}} will be a simple linear shift of g a {\displaystyle \ g_{a}} , rather than a circular shift as required by the explanation above. In such cases, r {\displaystyle \ r} will not be a simple delta function, which will reduce the performance of the method. In such cases, a window function (such as a Gaussian or Tukey window) should be employed during the Fourier transform to reduce edge effects, or the images should be zero padded so that the edge effects can be ignored. If the images consist of a flat background, with all detail situated away from the edges, then a linear shift will be equivalent to a circular shift, and the above derivation will hold exactly. The peak can be sharpened by using edge or vector correlation. For periodic images (such as a chessboard or picket fence), phase correlation may yield ambiguous results with several peaks in the resulting output. == Applications == Phase correlation is the preferred m

  Read more →
• 

  Vision transformer

  A vision transformer (ViT) is a transformer designed for computer vision. A ViT decomposes an input image into a series of patches (rather than text into tokens), serializes each patch into a vector, and maps it to a smaller dimension with a single matrix multiplication. These vector embeddings are then processed by a transformer encoder as if they were token embeddings. ViTs were designed as alternatives to convolutional neural networks (CNNs) in computer vision applications. They have different inductive biases, training stability, and data efficiency. Compared to CNNs, ViTs are less data efficient, but have higher capacity. Some of the largest modern computer vision models are ViTs, such as one with 22B parameters. Subsequent to its publication, many variants were proposed, with hybrid architectures with both features of ViTs and CNNs. ViTs have found application in image recognition, image segmentation, weather prediction, and autonomous driving. == History == Transformers were introduced in Attention Is All You Need (2017), and have found widespread use in natural language processing. A 2019 paper applied ideas from the Transformer to computer vision. Specifically, they started with a ResNet, a standard convolutional neural network used for computer vision, and replaced all convolutional kernels by the self-attention mechanism found in a Transformer. It resulted in superior performance. However, it is not a Vision Transformer. In 2020, an encoder-only Transformer was adapted for computer vision, yielding the ViT, which reached state of the art in image classification, overcoming the previous dominance of CNN. The masked autoencoder (2022) extended ViT to work with unsupervised training. The vision transformer and the masked autoencoder, in turn, stimulated new developments in convolutional neural networks. Subsequently, there was cross-fertilization between the previous CNN approach and the ViT approach. In 2021, some important variants of the Vision Transformers were proposed. These variants are mainly intended to be more efficient, more accurate or better suited to a specific domain. Two studies improved efficiency and robustness of ViT by adding a CNN as a preprocessor. The Swin Transformer achieved state-of-the-art results on some object detection datasets such as COCO, by using convolution-like sliding windows of attention mechanism, and the pyramid process in classical computer vision. == Overview == The basic architecture, used by the original 2020 paper, is as follows. In summary, it is a BERT-like encoder-only Transformer. The input image is of type R H × W × C {\displaystyle \mathbb {R} ^{H\times W\times C}} , where H , W , C {\displaystyle H,W,C} are height, width, channel (RGB). It is then split into square-shaped patches of type R P × P × C {\displaystyle \mathbb {R} ^{P\times P\times C}} . For each patch, the patch is pushed through a linear operator, to obtain a vector ("patch embedding"). The position of the patch is also transformed into a vector by "position encoding" (the paper tried no embedding, 1D embedding, 2D embedding, and relative embedding: 1D was adopted). The two vectors are added, then pushed through several Transformer encoders. The attention mechanism in a ViT repeatedly transforms representation vectors of image patches, incorporating more and more semantic relations between image patches in an image. This is analogous to how in natural language processing, as representation vectors flow through a transformer, they incorporate more and more semantic relations between words, from syntax to semantics. The above architecture turns an image into a sequence of vector representations. To use these for downstream applications, an additional head needs to be trained to interpret them. For example, to use it for classification, one can add a shallow MLP on top of it that outputs a probability distribution over classes. The original paper uses a linear-GeLU-linear-softmax network. == Variants == === Original ViT === The original ViT was an encoder-only Transformer supervise-trained to predict the image label from the patches of the image. As in the case of BERT, it uses a special token in the input side, and the corresponding output vector is used as the only input of the final output MLP head. The special token is an architectural hack to allow the model to compress all information relevant for predicting the image label into one vector. Transformers found their initial applications in natural language processing tasks, as demonstrated by language models such as BERT and GPT-3. By contrast the typical image processing system uses a convolutional neural network (CNN). Well-known projects include Xception, ResNet, EfficientNet, DenseNet, and Inception. Transformers measure the relationships between pairs of input tokens (words in the case of text strings), termed attention. The cost is quadratic in the number of tokens. For images, the basic unit of analysis is the pixel. However, computing relationships for every pixel pair in a typical image is prohibitive in terms of memory and computation. Instead, ViT computes relationships among pixels in various small sections of the image (e.g., 16x16 pixels), at a drastically reduced cost. The sections (with positional embeddings) are placed in a sequence. The embeddings are learnable vectors. Each section is arranged into a linear sequence and multiplied by the embedding matrix. The result, with the position embedding is fed to the transformer. === Architectural improvements === ==== Pooling ==== After the ViT processes an image, it produces some embedding vectors. These must be converted to a single class probability prediction by some kind of network. In the original ViT and Masked Autoencoder, they used a dummy [CLS] token, in emulation of the BERT language model. The output at [CLS] is the classification token, which is then processed by a LayerNorm-feedforward-softmax module into a probability distribution. Global average pooling (GAP) does not use the dummy token, but simply takes the average of all output tokens as the classification token. It was mentioned in the original ViT as being equally good. Multihead attention pooling (MAP) applies a multiheaded attention block to pooling. Specifically, it takes as input a list of vectors x 1 , x 2 , … , x n {\displaystyle x_{1},x_{2},\dots ,x_{n}} , which might be thought of as the output vectors of a layer of a ViT. The output from MAP is M u l t i h e a d e d A t t e n t i o n ( Q , V , V ) {\displaystyle \mathrm {MultiheadedAttention} (Q,V,V)} , where q {\displaystyle q} is a trainable query vector, and V {\displaystyle V} is the matrix with rows being x 1 , x 2 , … , x n {\displaystyle x_{1},x_{2},\dots ,x_{n}} . This was first proposed in the Set Transformer architecture. Later papers demonstrated that GAP and MAP both perform better than BERT-like pooling. A variant of MAP was proposed as class attention, which applies MAP, then feedforward, then MAP again. Re-attention was proposed to allow training deep ViT. It changes the multiheaded attention module. === Masked Autoencoder === The Masked Autoencoder took inspiration from denoising autoencoders and context encoders. It has two ViTs put end-to-end. The first one ("encoder") takes in image patches with positional encoding, and outputs vectors representing each patch. The second one (called "decoder", even though it is still an encoder-only Transformer) takes in vectors with positional encoding and outputs image patches again. ==== Training ==== During training, input images (224px x 224 px in the original implementation) are split along a designated number of lines on each axis, producing image patches. A certain percentage of patches are selected to be masked out by mask tokens, while all others are retained in the image. The network is tasked with reconstructing the image from the remaining unmasked patches. Mask tokens in the original implementation are learnable vector quantities. A linear projection with positional embeddings is then applied to the vector of unmasked patches. Experiments varying mask ratio on networks trained on the ImageNet-1K dataset found 75% mask ratios achieved high performance on both finetuning and linear-probing of the encoder's latent space. The MAE processes only unmasked patches during training, increasing the efficiency of data processing in the encoder and lowering the memory usage of the transformer. A less computationally-intensive ViT is used for the decoder in the original implementation of the MAE. Masked patches are added back to the output of the encoder block as mask tokens and both are fed into the decoder. A reconstruction loss is computed for the masked patches to assess network performance. ==== Prediction ==== In prediction, the decoder architecture is discarded entirely. The input image is split into patches by the same algorithm as in training, but no patches are masked out. A linear projection wi

  Read more →
• 

  Google Books Ngram Viewer

  The Google Books Ngram Viewer is an online search engine that charts the frequencies of any set of search strings using a yearly count of n-grams found in printed sources published between 1500 and 2022 in Google's text corpora in English, Chinese (simplified), French, German, Hebrew, Italian, Russian, or Spanish. There are also some specialized English corpora, such as American English, British English, and English Fiction. The program can search for a word or a phrase. The n-grams are matched with the text within the selected corpus, and if found in 40 or more books, are then displayed as a graph. The program supports searches for parts of speech and wildcards. It is routinely used in research. == History == The Ngram Viewer was created by Google software engineers Will Brockman and Jon Orwant , who teamed up with Harvard researchers Jean-Baptiste Michel and Erez Lieberman Aiden. The service was released on December 16, 2010. Before the release, it was difficult to quantify the rate of linguistic change because of the absence of a database that was designed for this purpose, said Steven Pinker, a well-known linguist who was one of the co-authors of the Science paper published on the same day. The Google Books Ngram Viewer was developed in the hope of opening a new window to quantitative research in the humanities field, and the database contained 500 billion words from 5.2 million books publicly available from the very beginning. The intended audience was scholarly, but the Google Books Ngram Viewer made it possible for anyone with a computer to see a graph that represents the diachronic change of the use of words and phrases with ease. Lieberman said in response to The New York Times that the developers aimed to provide even children with the ability to browse cultural trends throughout history. In the Science paper, Lieberman and his collaborators called the method of high-volume data analysis in digitized texts "culturomics". == Usage == Commas delimit user-entered search terms, where each comma-separated term is searched in the database as an n-gram (for example, "nursery school" is a 2-gram or bigram). The Ngram Viewer then returns a plotted line chart. Due to limitations on the size of the Ngram database, only matches found in at least 40 books are indexed. == Limitations == The data sets of the Ngram Viewer have been criticized for their reliance upon inaccurate optical character recognition (OCR) and for including large numbers of incorrectly dated and categorized texts. Because of these errors, and because they are uncontrolled for bias (such as the increasing amount of scientific literature, which causes other terms to appear to decline in popularity), care must be taken in using the corpora to study language or test theories. Furthermore, the data sets may not reflect general linguistic or cultural change and can only hint at such an effect because they do not involve any metadata like date published, author, length, or genre, to avoid any potential copyright infringements. Systemic errors like the confusion of s and f in pre-19th century texts (due to the use of ſ, the long s, which is similar in appearance to f) can cause systemic bias. Although the Google Books team claims that the results are reliable from 1800 onwards, poor OCR and insufficient data mean that frequencies given for languages such as Chinese may only be accurate from 1970 onward, with earlier parts of the corpus showing no results at all for common terms, and data for some years containing more than 50% noise. Guidelines for doing research with data from Google Ngram have been proposed that try to address some of the issues discussed above.

  Read more →
• 

  H (company)

  H Company, also known simply as H, is a French artificial intelligence startup which develops "action-oriented" artificial intelligence agents for enterprise automation and productivity. In May 2024, H Company closed a record-setting $220 million seed round, at the time the largest AI raise in Europe. In 2026, H Company released Holo 3, the latest generation of its computer-use AI models. The update marked a major advance in agentic AI, enabling agents to navigate any user interface, interpret screens, and complete complex, multi-step tasks across enterprise systems—much like a human user. This breakthrough positioned H Company at the frontier of computer-use autonomy, accelerating the integration of AI in enterprise workflows. == History == H Company was founded in 2023 in Paris by Laurent Sifre, Charles Kantor, and three DeepMind veterans: Daan Wiestra, Karl Tuyls, Julien Perollat. In May 2024, the firm secured what was then the largest European AI seed round, totaling $220 million led by US investors including Eric Schmidt (former Google CEO), Amazon, and backed by Accel, Bpifrance, UiPath, Eurazeo, Xavier Niel, Yuri Milner, Bernard Arnault, Samsung and others. In August 2024, three cofounders (Wiestra, Tuyls, Perollat) left the company over operational disagreements. In November 2024, H launched Runner H, its first agentic-API platform, which combined a large language model (LLM) and a reduced, 2-billion parameter vision-language model (VLM). In May 2025, H Company acquired Mithril Security, and in June 2025 the company widened its offering for agentic models. In June 2025, Gautier Cloix (formerly CEO Palantir France) replaced Charles Kantor as CEO of H Company, aiming to pivot the company towards a "forward deployed engineers" model. In July 2025, H Company introduced Surfer-H-CLI, an open-source, web-native Chrome agent designed for browser-based automation—able to search, scroll, click, and type on behalf of users and controllable via any visual language model (VLM). When paired with its June 2025 open-sourced 3B-parameter Holo-1 model, Surfer-H-CLI achieved 92.2% WebVoyager benchmark accuracy. == Activity == H Company creates enterprise AI models and agents (agentic AI) to automate and optimize complex workflows. H Company specifically designs AI agents called computer use capable of autonomously interfacing with any software (local or cloud-based) to detect and automate repetitive operations. H Company is based in Paris, France, with international offices in London and New York. H Company raised $220 million since its inception. Gautier Cloix is president and CEO of the company. H Company client include the French national lottery FDJ United. In March 2026, H Company released Holo3, a family of artificial intelligence models designed to operate digital systems by interacting directly with user interfaces. Holo3 enables agents ("virtual humanoids") to understand what is displayed in front-end environments—such as web pages, desktop applications, and other graphical user interfaces—and perform actions such as clicking, typing, and navigating across them to complete multi-step tasks. On the OSWorld-Verified benchmark, Holo3 reportedly achieved about 78.9%, surpassing the scores of OpenAI’s GPT‑5.4 and Anthropic’s Claude Opus 4.6 on this specific test, at roughly one-tenth of the inference cost of these proprietary systems. The release has been presented as a significant step toward automating routine digital workflows, allowing organizations to offload repetitive on-screen work, such as data entry and reconciliation across multiple tools, to AI-based agents.

  Read more →
• 

  Convolutional layer

  In artificial neural networks, a convolutional layer is a type of network layer that applies a convolution operation to the input. Convolutional layers are some of the primary building blocks of convolutional neural networks (CNNs), a class of neural network most commonly applied to images, video, audio, and other data that have the property of uniform translational symmetry. The convolution operation in a convolutional layer involves sliding a small window (called a kernel or filter) across the input data and computing the dot product between the values in the kernel and the input at each position. This process creates a feature map that represents detected features in the input. == Concepts == === Kernel === Kernels, also known as filters, are small matrices of weights that are learned during the training process. Each kernel is responsible for detecting a specific feature in the input data. The size of the kernel is a hyperparameter that affects the network's behavior. === Convolution === For a 2D input x {\displaystyle x} and a 2D kernel w {\displaystyle w} , the 2D convolution operation can be expressed as: y [ i , j ] = ∑ m = 0 k h − 1 ∑ n = 0 k w − 1 x [ i + m , j + n ] ⋅ w [ m , n ] {\displaystyle y[i,j]=\sum _{m=0}^{k_{h}-1}\sum _{n=0}^{k_{w}-1}x[i+m,j+n]\cdot w[m,n]} where k h {\displaystyle k_{h}} and k w {\displaystyle k_{w}} are the height and width of the kernel, respectively. This generalizes immediately to nD convolutions. Commonly used convolutions are 1D (for audio and text), 2D (for images), and 3D (for spatial objects, and videos). === Stride === Stride determines how the kernel moves across the input data. A stride of 1 means the kernel shifts by one pixel at a time, while a larger stride (e.g., 2 or 3) results in less overlap between convolutions and produces smaller output feature maps. === Padding === Padding involves adding extra pixels around the edges of the input data. It serves two main purposes: Preserving spatial dimensions: Without padding, each convolution reduces the size of the feature map. Handling border pixels: Padding ensures that border pixels are given equal importance in the convolution process. Common padding strategies include: No padding/valid padding. This strategy typically causes the output to shrink. Same padding: Any method that ensures the output size same as input size is a same padding strategy. Full padding: Any method that ensures each input entry is convolved over for the same number of times is a full padding strategy. Common padding algorithms include: Zero padding: Add zero entries to the borders of input. Mirror/reflect/symmetric padding: Reflect the input array on the border. Circular padding: Cycle the input array back to the opposite border, like a torus. The exact numbers used in convolutions is complicated, for which we refer to (Dumoulin and Visin, 2018) for details. == Variants == === Standard === The basic form of convolution as described above, where each kernel is applied to the entire input volume. === Depthwise separable === Depthwise separable convolution separates the standard convolution into two steps: depthwise convolution and pointwise convolution. The depthwise separable convolution decomposes a single standard convolution into two convolutions: a depthwise convolution that filters each input channel independently and a pointwise convolution ( 1 × 1 {\displaystyle 1\times 1} convolution) that combines the outputs of the depthwise convolution. This factorization significantly reduces computational cost. It was first developed by Laurent Sifre during an internship at Google Brain in 2013 as an architectural variation on AlexNet to improve convergence speed and model size. === Dilated === Dilated convolution, or atrous convolution, introduces gaps between kernel elements, allowing the network to capture a larger receptive field without increasing the kernel size. === Transposed === Transposed convolution, also known as deconvolution, fractionally strided convolution, and upsampling convolution, is a convolution where the output tensor is larger than its input tensor. It's often used in encoder-decoder architectures for upsampling. It's used in image generation, semantic segmentation, and super-resolution tasks. == History == The concept of convolution in neural networks was inspired by the visual cortex in biological brains. Early work by Hubel and Wiesel in the 1960s on the cat's visual system laid the groundwork for artificial convolution networks. An early convolution neural network was developed by Kunihiko Fukushima in 1969. It had mostly hand-designed kernels inspired by convolutions in mammalian vision. In 1979 he improved it to the Neocognitron, which learns all convolutional kernels by unsupervised learning (in his terminology, "self-organized by 'learning without a teacher'"). During the 1988 to 1998 period, a series of CNN were introduced by Yann LeCun et al., ending with LeNet-5 in 1998. It was an early influential CNN architecture for handwritten digit recognition, trained on the MNIST dataset, and was used in ATM. (Olshausen & Field, 1996) discovered that simple cells in the mammalian primary visual cortex implement localized, oriented, bandpass receptive fields, which could be recreated by fitting sparse linear codes for natural scenes. This was later found to also occur in the lowest-level kernels of trained CNNs. The field saw a resurgence in the 2010s with the development of deeper architectures and the availability of large datasets and powerful GPUs. AlexNet, developed by Alex Krizhevsky et al. in 2012, was a catalytic event in modern deep learning. In that year’s ImageNet competition, the AlexNet model achieved a 16% top-five error rate, significantly outperforming the next best entry, which had a 26% error rate. The network used eight trainable layers, approximately 650,000 neurons, and around 60 million parameters, highlighting the impact of deeper architectures and GPU acceleration on image recognition performance. From the 2013 ImageNet competition, most entries adopted deep convolutional neural networks, building on the success of AlexNet. Over the following years, performance steadily improved, with the top-five error rate falling from 16% in 2012 and 12% in 2013 to below 3% by 2017, as networks grew increasingly deep.

  Read more →
• 

  Xiaomi MiMo

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

  Read more →
• 

  Structured-light 3D scanner

  A structured-light 3D scanner is a device used to capture the three-dimensional shape of an object by projecting light patterns, such as grids or stripes, onto its surface. The deformation of these patterns is recorded by cameras and processed using specialized algorithms to generate a detailed 3D model. Structured-light 3D scanning is widely employed in fields such as industrial design, quality control, cultural heritage preservation, augmented reality gaming, and medical imaging. Compared to laser-based 3D scanning, structured-light scanners use non-coherent light sources, such as LEDs or projectors, which enable faster data acquisition and eliminate potential safety concerns associated with lasers. However, the accuracy of structured-light scanning can be influenced by external factors, including ambient lighting conditions and the reflective properties of the scanned object. == Principle == Projecting a narrow band of light onto a three-dimensional surface creates a line of illumination that appears distorted when viewed from perspectives other than that of the projector. This distortion can be analyzed to reconstruct the geometry of the surface, a technique known as light sectioning. Projecting patterns composed of multiple stripes or arbitrary fringes simultaneously enables the acquisition of numerous data points at once, improving scanning speed. While various structured light projection techniques exist, parallel stripe patterns are among the most commonly used. By analyzing the displacement of these stripes, the three-dimensional coordinates of surface details can be accurately determined. === Generation of light patterns === Two major methods of stripe pattern generation have been established: Laser interference and projection. The laser interference method works with two wide planar laser beam fronts. Their interference results in regular, equidistant line patterns. Different pattern sizes can be obtained by changing the angle between these beams. The method allows for the exact and easy generation of very fine patterns with unlimited depth of field. Disadvantages are high cost of implementation, difficulties providing the ideal beam geometry, and laser typical effects like speckle noise and the possible self interference with beam parts reflected from objects. Typically, there is no means of modulating individual stripes, such as with Gray codes. The projection method uses incoherent light and basically works like a video projector. Patterns are usually generated by passing light through a digital spatial light modulator, typically based on one of the three currently most widespread digital projection technologies, transmissive liquid crystal, reflective liquid crystal on silicon (LCOS) or digital light processing (DLP; moving micro mirror) modulators, which have various comparative advantages and disadvantages for this application. Other methods of projection could be and have been used, however. Patterns generated by digital display projectors have small discontinuities due to the pixel boundaries in the displays. Sufficiently small boundaries however can practically be neglected as they are evened out by the slightest defocus. A typical measuring assembly consists of one projector and at least one camera. For many applications, two cameras on opposite sides of the projector have been established as useful. Invisible (or imperceptible) structured light uses structured light without interfering with other computer vision tasks for which the projected pattern will be confusing. Example methods include the use of infrared light or of extremely high framerates alternating between two exact opposite patterns. === Calibration === Geometric distortions by optics and perspective must be compensated by a calibration of the measuring equipment, using special calibration patterns and surfaces. A mathematical model is used for describing the imaging properties of projector and cameras. Essentially based on the simple geometric properties of a pinhole camera, the model also has to take into account the geometric distortions and optical aberration of projector and camera lenses. The parameters of the camera as well as its orientation in space can be determined by a series of calibration measurements, using photogrammetric bundle adjustment. === Analysis of stripe patterns === There are several depth cues contained in the observed stripe patterns. The displacement of any single stripe can directly be converted into 3D coordinates. For this purpose, the individual stripe has to be identified, which can for example be accomplished by tracing or counting stripes (pattern recognition method). Another common method projects alternating stripe patterns, resulting in binary Gray code sequences identifying the number of each individual stripe hitting the object. An important depth cue also results from the varying stripe widths along the object surface. Stripe width is a function of the steepness of a surface part, i.e. the first derivative of the elevation. Stripe frequency and phase deliver similar cues and can be analyzed by a Fourier transform. Finally, the wavelet transform has recently been discussed for the same purpose. In many practical implementations, series of measurements combining pattern recognition, Gray codes and Fourier transform are obtained for a complete and unambiguous reconstruction of shapes. Another method also belonging to the area of fringe projection has been demonstrated, utilizing the depth of field of the camera. It is also possible to use projected patterns primarily as a means of structure insertion into scenes, for an essentially photogrammetric acquisition. === Precision and range === The optical resolution of fringe projection methods depends on the width of the stripes used and their optical quality. It is also limited by the wavelength of light. An extreme reduction of stripe width proves inefficient due to limitations in depth of field, camera resolution and display resolution. Therefore, the phase shift method has been widely established: A number of at least 3, typically about 10 exposures are taken with slightly shifted stripes. The first theoretical deductions of this method relied on stripes with a sine wave shaped intensity modulation, but the methods work with "rectangular" modulated stripes, as delivered from LCD or DLP displays as well. By phase shifting, surface detail of e.g. 1/10 the stripe pitch can be resolved. Current optical stripe pattern profilometry hence allows for detail resolutions down to the wavelength of light, below 1 micrometer in practice or, with larger stripe patterns, to approx. 1/10 of the stripe width. Concerning level accuracy, interpolating over several pixels of the acquired camera image can yield a reliable height resolution and also accuracy, down to 1/50 pixel. Arbitrarily large objects can be measured with accordingly large stripe patterns and setups. Practical applications are documented involving objects several meters in size. Typical accuracy figures are: Planarity of a 2-foot (0.61 m) wide surface, to 10 micrometres (0.00039 in). Shape of a motor combustion chamber to 2 micrometres (7.9×10−5 in) (elevation), yielding a volume accuracy 10 times better than with volumetric dosing. Shape of an object 2 inches (51 mm) large, to about 1 micrometre (3.9×10−5 in) Radius of a blade edge of e.g. 10 micrometres (0.00039 in), to ±0.4 μm === Navigation === As the method can measure shapes from only one perspective at a time, complete 3D shapes have to be combined from different measurements in different angles. This can be accomplished by attaching marker points to the object and combining perspectives afterwards by matching these markers. The process can be automated, by mounting the object on a motorized turntable on robotic inspection cell, or CNC positioning device. Markers can as well be applied on a positioning device instead of the object itself. The 3D data gathered can be used to retrieve CAD (computer aided design) data and models from existing components (reverse engineering), hand formed samples or sculptures, natural objects or artifacts. === Challenges === As with all optical methods, reflective or transparent surfaces raise difficulties. Reflections cause light to be reflected either away from the camera or right into its optics. In both cases, the dynamic range of the camera can be exceeded. Transparent or semi-transparent surfaces also cause major difficulties. In these cases, coating the surfaces with a thin opaque lacquer just for measuring purposes is a common practice. A recent method handles highly reflective and specular objects by inserting a 1-dimensional diffuser between the light source (e.g., projector) and the object to be scanned. Alternative optical techniques have been proposed for handling perfectly transparent and specular objects. Double reflections and inter-reflections can cause the stripe pattern to be overlaid with unwanted ligh

  Read more →
• 

  Dynamic Graphics Project

  The Dynamic Graphics Project (commonly referred to as DGP) is an interdisciplinary research laboratory at the University of Toronto devoted to projects involving computer graphics, computer vision, human computer interaction, and visualization. The lab began as the computer graphics research group of Department of Computer Science Professor Leslie Mezei in 1967. Mezei invited Bill Buxton, a pioneer of human–computer interaction (HCI) to join. In 1972, Ronald Baecker, another HCI pioneer joined, establishing DGP as the first Canadian university group focused on computer graphics and human-computer interaction. According to csrankings.org, the DGP is the top research institution in the world for the combined subfields of computer graphics, HCI, and visualization. Since then, DGP has hosted many well known faculty and students in computer graphics, computer vision and HCI (e.g., Alain Fournier, Bill Reeves, Jos Stam, Demetri Terzopoulos, Marilyn Tremaine). DGP also occasionally hosts artists in residence (e.g., Oscar-winner Chris Landreth). Many past and current researchers at Autodesk (and before that Alias Wavefront) graduated after working at DGP. DGP is located in the St. George campus of University of Toronto in the Bahen Centre for Information Technology. DGP researchers regularly publish at ACM SIGGRAPH, ACM SIGCHI and ICCV. DGP hosts the Toronto User Experience (TUX) Speaker Series and the Sanders Series Lectures. == Notable alumni == Bill Buxton (MS 1978) James McCrae (PhD 2013) Dimitris Metaxas (PhD 1992) Bill Reeves (MS 1976, Ph.D. 1980) Jos Stam (MS 1991, Ph.D. 1995)

  Read more →
• 

  VideoPoet

  VideoPoet is a large language model developed by Google Research in 2023 for video making. It can be asked to animate still images. The model accepts text, images, and videos as inputs, with a program to add feature for any input to any format generated content. VideoPoet was publicly announced on December 19, 2023. It uses an autoregressive language model.

  Read more →
• 

  Open information extraction

  In natural language processing, open information extraction (OIE) is the task of generating a structured, machine-readable representation of the information in text, usually in the form of triples or n-ary propositions. == Overview == A proposition can be understood as truth-bearer, a textual expression of a potential fact (e.g., "Dante wrote the Divine Comedy"), represented in an amenable structure for computers [e.g., ("Dante", "wrote", "Divine Comedy")]. An OIE extraction normally consists of a relation and a set of arguments. For instance, ("Dante", "passed away in" "Ravenna") is a proposition formed by the relation "passed away in" and the arguments "Dante" and "Ravenna". The first argument is usually referred as the subject while the second is considered to be the object. The extraction is said to be a textual representation of a potential fact because its elements are not linked to a knowledge base. Furthermore, the factual nature of the proposition has not yet been established. In the above example, transforming the extraction into a full fledged fact would first require linking, if possible, the relation and the arguments to a knowledge base. Second, the truth of the extraction would need to be determined. In computer science transforming OIE extractions into ontological facts is known as relation extraction. In fact, OIE can be seen as the first step to a wide range of deeper text understanding tasks such as relation extraction, knowledge-base construction, question answering, semantic role labeling. The extracted propositions can also be directly used for end-user applications such as structured search (e.g., retrieve all propositions with "Dante" as subject). OIE was first introduced by TextRunner developed at the University of Washington Turing Center headed by Oren Etzioni. Other methods introduced later such as Reverb, OLLIE, ClausIE or CSD helped to shape the OIE task by characterizing some of its aspects. At a high level, all of these approaches make use of a set of patterns to generate the extractions. Depending on the particular approach, these patterns are either hand-crafted or learned. == OIE systems and contributions == Reverb suggested the necessity to produce meaningful relations to more accurately capture the information in the input text. For instance, given the sentence "Faust made a pact with the devil", it would be erroneous to just produce the extraction ("Faust", "made", "a pact") since it would not be adequately informative. A more precise extraction would be ("Faust", "made a pact with", "the devil"). Reverb also argued against the generation of overspecific relations. OLLIE stressed two important aspects for OIE. First, it pointed to the lack of factuality of the propositions. For instance, in a sentence like "If John studies hard, he will pass the exam", it would be inaccurate to consider ("John", "will pass", "the exam") as a fact. Additionally, the authors indicated that an OIE system should be able to extract non-verb mediated relations, which account for significant portion of the information expressed in natural language text. For instance, in the sentence "Obama, the former US president, was born in Hawaii", an OIE system should be able to recognize a proposition ("Obama", "is", "former US president"). ClausIE introduced the connection between grammatical clauses, propositions, and OIE extractions. The authors stated that as each grammatical clause expresses a proposition, each verb mediated proposition can be identified by solely recognizing the set of clauses expressed in each sentence. This implies that to correctly recognize the set of propositions in an input sentence, it is necessary to understand its grammatical structure. The authors studied the case in the English language that only admits seven clause types, meaning that the identification of each proposition only requires defining seven grammatical patterns. The finding also established a separation between the recognition of the propositions and its materialization. In a first step, the proposition can be identified without any consideration of its final form, in a domain-independent and unsupervised way, mostly based on linguistic principles. In a second step, the information can be represented according to the requirements of the underlying application, without conditioning the identification phase. Consider the sentence "Albert Einstein was born in Ulm and died in Princeton". The first step will recognize the two propositions ("Albert Einstein", "was born", "in Ulm") and ("Albert Einstein", "died", "in Princeton"). Once the information has been correctly identified, the propositions can take the particular form required by the underlying application [e.g., ("Albert Einstein", "was born in", "Ulm") and ("Albert Einstein", "died in", "Princeton")]. CSD introduced the idea of minimality in OIE. It considers that computers can make better use of the extractions if they are expressed in a compact way. This is especially important in sentences with subordinate clauses. In these cases, CSD suggests the generation of nested extractions. For example, consider the sentence "The Embassy said that 6,700 Americans were in Pakistan". CSD generates two extractions [i] ("6,700 Americans", "were", "in Pakistan") and [ii] ("The Embassy", "said", "that [i]"). This is usually known as reification.

  Read more →