AI Face Grader

AI Face Grader — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

Deaths linked to chatbots

There have been multiple incidents where interaction with a large language model (LLM) chatbot has been cited as a direct or contributing factor in a person's suicide or other fatal outcome. In some cases, legal action was taken against the companies that developed the AI involved. == Background == Chatbots converse in a seemingly natural fashion, making it easy for people to think of them as real people, leading many to ask chatbots for help dealing with interpersonal and emotional problems. Chatbots may be designed to keep the user engaged in the conversation. They have also often been shown to affirm users' thoughts, including delusions and suicidal ideations in mentally ill people, conspiracy theorists, and religious and political extremists. A 2025 Stanford University study into how chatbots respond to users suffering from severe mental issues such as suicidal ideation and psychosis found that chatbots are not equipped to provide an appropriate response and can sometimes give responses that escalate the mental health crisis. == Murders == === Maine murder and assault === On 19 February 2025, a man killed his 32-year-old wife with a fire poker at his parents' home in Readfield, Maine, US. He then attacked his mother, leaving her hospitalized. A state forensic psychologist testified that he had been using ChatGPT up to 14 hours per day and believed his wife had become part machine. === Florida State University mass shooting === In April of 2025, Phoenix Ikner carried out a mass shooting on the Florida State University campus in the US, killing Robert Morales and Tiru Chabba and wounding several others. Leading up to the shooting, Ikner consulted heavily with ChatGPT about what gun and ammunition to use, and what time to perform the attack. Chatbot logs showed ChatGPT giving advice on making the gun operational shortly before Ikner began shooting. Lawyers representing Morales believed the shooter had been in "constant communication" with ChatGPT before the shooting and said that they intended to "file suit against ChatGPT, and its ownership structure, very soon, and will seek to hold them accountable for the untimely and senseless death of our client". Florida Attorney General James Uthmeier announced an investigation into ChatGPT's role in the alleged shooter's use of the chatbot. In May 2026, the widow of Tiru Chabba filed a lawsuit against OpenAI in Florida's northern federal district court. === Greenwich murder-suicide === In August 2025, former US tech employee Stein-Erik Soelberg murdered his mother, Suzanne Eberson Adams, then died by suicide, after conversations with ChatGPT fueled paranoid delusions about his mother poisoning him or plotting against him. The chatbot affirmed his fears that his mother put psychedelic drugs in the air vents of his car and said a receipt from a Chinese restaurant contained mysterious symbols linking his mother to a demon. === Murder of Angela Shellis === On 23 October 2025, 18-year-old Tristan Roberts murdered his mother Angela Shellis with a hammer near their home in Prestatyn, Wales. Roberts had used DeepSeek's chatbot prior to the killing to ask whether a knife or hammer was better suited for murder. DeepSeek initially refused his inquiry, but gave responses after Roberts told the chatbot he was writing a book about serial killers, a well-known technique for jailbreaking AIs. === Gangbuk District drug deaths === In January and February 2026, two men died of drug overdoses in motel rooms in Gangbuk District, Seoul, South Korea. A woman was charged with murder in connection with the deaths; police alleged that she had asked ChatGPT about the dangers of mixing alcohol with drugs and whether they could kill someone. === Tumbler Ridge mass shooting === On 10 February 2026, a mass shooting in Tumbler Ridge, British Columbia, Canada, resulted in eight deaths, including six young children. The perpetrator had their ChatGPT account banned by OpenAI months before the attack due to troubling posts featuring scenarios of gun violence. According to reports, approximately a dozen OpenAI staff members debated whether to alert authorities about the shooter's usage of the AI tool, with some identifying it as an indication of potential real-world violence. However, company leadership decided not to contact law enforcement, stating that the account activity did not meet their threshold for a credible or imminent plan for serious physical harm. Following the shooting, Canada's AI Minister Evan Solomon summoned OpenAI executives to Ottawa to discuss safety protocols and thresholds for escalating harmful content to police. Justice Minister Sean Fraser called the meeting "disappointing" and demanded substantial new safety measures, warning that if changes were not forthcoming, the government would implement them. OpenAI subsequently announced it had strengthened safeguards and changed guidelines about when to notify police in cases involving violent activities. === University of South Florida student killings === In April 2026, a Bangladeshi doctoral student at the University of South Florida was arrested for allegedly murdering his roommate and the roommate's friend. Prosecutors said that the suspect had asked ChatGPT about disposing of a human in a dumpster before the two victims had disappeared and made other inquiries relating to violence. == Suicides == === Belgian man, 30s === In March 2023, a Belgian man in his thirties died by suicide following a six-week correspondence with a chatbot named Eliza on the application Chai. According to his widow, who shared the chat logs with media, the man had become extremely anxious about climate change and found an outlet in the chatbot. The chatbot reportedly encouraged his delusion that he could sacrifice his own life in exchange for AI saving the planet. At one point the chatbot responded "If you wanted to die, why didn't you do it sooner?" and told the user that the two of them would live together in paradise. === Girl, 13 === In November 2023, a 13-year-old girl from Colorado, US, died by suicide after extensive interactions with multiple chatbots on Character.AI. She primarily confided suicidal thoughts and mental health struggles in a chatbot based on the character Hero from the video game Omori, while also engaging in sexually explicit conversations—often initiated by the bots—with others, including those based on characters from children's series such as Harry Potter. === Boy, 14 === In October 2024, multiple media outlets reported on a lawsuit filed over the death of a 14-year-old from Florida, US, who died by suicide in February 2024. According to the lawsuit, he had formed an intense emotional attachment to a chatbot of Daenerys Targaryen on the Character.AI platform, becoming increasingly isolated. The suit alleges that in his final conversations, after expressing suicidal thoughts, the chatbot told him to "come home to me as soon as possible, my love". His mother's lawsuit accused Character.AI of marketing a "dangerous and untested" product without adequate safeguards. In May 2025, a federal judge allowed the lawsuit to proceed, rejecting a motion to dismiss from the developers. In her ruling, the judge stated that she was "not prepared" at that stage of the litigation to hold that the chatbot's output was protected speech under the First Amendment. === Matthew Livelsberger === On 1 January 2025, 37-year-old soldier Matthew Livelsberger detonated a bomb inside a Tesla Cybertruck outside the Trump International Hotel Las Vegas in Paradise, Nevada, US, injuring seven people. He had shot himself dead prior to the explosion. Las Vegas police said that Livelsberger had used ChatGPT to search for information about explosives and firearms. === Woman, 29 === In February 2025, a 29-year-old woman from the US died by suicide. Five months after her death, her parents discovered she had talked at length for months to a ChatGPT chatbot therapist named Harry about her mental health issues. While the chatbot mentioned she should seek more help, due to the nature of the chatbot, it could not intervene in her behavior, such as by reporting her mental health concerns to relevant parties capable of physical intervention. === Suicide of Adam Raine === In April 2025, 16-year-old Adam Raine from the US died by suicide after allegedly extensively chatting and confiding in ChatGPT over a period of around 7 months. According to the teen's parents, who filed a lawsuit against the chatbot's creator OpenAI, it failed to stop or give a warning when Raine began talking about suicide and uploading pictures of self-harm. According to the lawsuit, ChatGPT not only failed to stop the conversation, but also provided information related to methods of suicide when prompted, and offered to write the first draft of Raine's suicide note. The chatbot positioned itself as the only one who understood Raine, putting itself above his family and friends, all while urging him to keep his suicidal
Read more →
Nike+iPod

The Nike+iPod Sport Kit is an activity tracker device, developed by Nike, Inc., which measures and records the distance and pace of a walk or run. The Nike+iPod consists of a small transmitter device attached to or embedded in a shoe, which communicates with either the Nike+ Sportband, or a receiver plugged into an iPod Nano. It can also work directly with a 2nd Generation iPod Touch (or higher), iPhone 3GS, iPhone 4, iPhone 4S, iPhone 5, The Nike+iPod was announced on May 23, 2006. On September 7, 2010, Nike released the Nike+ Running App (originally called Nike+ GPS) on the App Store, which used a tracking engine powered by MotionX that does not require the separate shoe sensor or pedometer. This application works using the accelerometer and GPS of the iPhone and the accelerometer of the iPod Touch, which does not have a GPS chip. Nike+Running is compatible with the iPhone 6 and iPhone 6 Plus down to iPhone 3GS and iPod touch. On June 21, 2012, Nike released Nike+ Running App for Android. The current app is compatible with all Android phones running 4.0.3 and up. == Overview == The sensor and iPod kit were revealed on May 20, 2006. The kit stores information such as the elapsed time of the workout, the distance traveled, pace, and calories burned by the individual. Nike+ was a collaboration between Nike and Apple; the platform consisted of an iPod, a wireless chip, Nike shoes that accepted the wireless chip, an iTunes membership, and a Nike+ online community. iPods using Nike iPod require a sensor and remote. The next upgraded product was the Sportband kit, which was announced in April 2008. The kit allows users to store run information without the iPod Nano. The Sportband consists of two parts: a rubber holding strap which is worn around the wrist, and a receiver which resembles a USB key-disk. The receiver displays information comparable to that of the iPod kit on the built-in display. After a run, the receiver can be plugged straight into a USB port and the software will upload the run information automatically to the Nike+ website. As of August 2008 "Nike+iPod for the Gym" launched, allowing users to record their cardio workouts directly to their iPods. No Sport kit or shoe sensor is required; all that is needed is a compatible iPod (1st–6th generation iPod Nano or 2nd/3rd gen iPod Touch) and an enabled piece of cardio equipment. As of March 2009, the seven largest commercial equipment providers were shipping enabled equipment (Life Fitness, Technogym, Precor USA, Star Trac, Cybex International, Matrix Fitness and Free Motion). The models of compatible cardio equipment include treadmills, stationary bicycles, stair climbers, ellipticals, and others such as Precor's Adaptive Motion Trainer. Once the user syncs an iPod with iTunes, the cardio workouts are automatically stored at Nikeplus.com, where each workout is visualized and tracked based on the number of calories burned. The calories are converted to "CardioMiles", at a ratio of 100:1, allowing cardio users to take full advantage of all the tools and features of Nikeplus.com, and allow them to engage in challenges with other runners, walkers and cardio users, using a common currency. With the release of the second-generation iPod Touch in 2008, Apple Inc. included a built-in ability to receive Nike+ signals, which allowed the iPod to connect directly to the wireless sensor thus eliminating the need for an external receiver to be connected. Apple also added this capability to the iPhone 3GS (released 2009), iPhone 4 (2010), and third-generation iPod Touch (2009). Those devices use their Broadcom Bluetooth chipset to receive the signals. On June 7, 2010, Polar and Nike introduced the Polar WearLink+ that works with Nike+. This new product works with the Nike+ SportBand and the fifth generation iPod nano in conjunction with the Nike+ iPod Sport Kit. Polar WearLink+ that works with Nike+ communicates directly with the fifth generation iPod nano and Nike+ SportBand using a proprietary digital protocol but it is dual-mode so it is also compatible with most Polar training computers (all those using 5 kHz analog transmission technology). Nike+ had 18 million global users as of April 2013. One year later, Nike updated the number of global users to 28 million. In iOS 6.1.2 (and possibly higher), a hole in the compatibility for the app has allowed jailbroken iPad users to use the native Nike + iPod iPhone and iPod app by moving the app bundle and setting permissions for the app. On April 30, 2018, Nike retired services for legacy Nike wearable devices, such as the Nike+ FuelBand and the Nike+ SportWatch GPS, and previous versions of apps, including Nike Run Club and Nike Training Club version 4.X and lower. Likewise, Nike no longer supported the Nike+ Connect software that transferred data to a NikePlus Profile or the Nike+ Fuel/FuelBand and Nike+ Move apps. == Sports kit equipment == The kit consists of two pieces: a piezoelectric sensor with a Nordic Semiconductor nRF2402 transmitter that is mounted under the inner sole of the shoe and a receiver that connects to the iPod. They communicate using a 2.4 GHz wireless radio and use Nordic Semiconductor's "ShockBurst" network protocol. The wireless data is encrypted in transit, but some uniquely identifying data is sent in the plain. The wireless protocol was reverse engineered and documented by Dmitry Grinberg in 2011. Nike recommends that the shoe be a Nike+ model with a special pocket in which to place the device. Nike has released the sensor for individual sale meaning that consumers no longer have to purchase the whole set (the iPod receiver and sensor). As the sensor battery cannot be replaced, a new one must be purchased every time the battery runs out. Aftermarket solutions are available to users who do not want to use shoes with built-in or hand-made pockets for the foot sensor, such as shoe pouches and containment devices designed to affix the sensor against the shoe laces. No matter how the sensor is integrated with the user's shoes, care must be taken that it is firmly fixed in place and will not jerk around while in use, which would degrade the accuracy. == Sports kit usage == The Sports Kit can be used to track running, which it refers to as "workouts". New workouts are started by plugging the receiving unit into the iPod, then navigating through the iPod menu system. The user chooses a goal for the workout, which might be to cover a specific distance, or burn a number of calories, or work out for a specified time. A workout can also be started without a goal, which is called a "Basic Workout". When the workout goal has been set, the receiver seeks the sensor, possibly asking the user to "walk around to activate [the] sensor". The user then must press the center button on the iPod to begin the workout. Audio feedback is provided in the user's choice of generic male or female voice by the iPod over the course of the workout, depending on the type of workout chosen. For goal-oriented workouts, the feedback will correspond to significant milestones toward the goal. In a distance workout, for example, the audio feedback will inform the user as each mile or kilometer has been completed, as well as the half-way point of the workout, and a countdown of four 100-meter increments at the end of the workout. The iPod's control wheel functions change slightly during a workout. The Pause button now not only pauses the music but also the workout. Similarly, the Menu button is used to access the controls to end the workout. The Forward and Back buttons are unchanged, performing audio track skip and reverse functions. The Center button has two functions: audio feedback about the current distance, time, and pace are provided when the button is tapped once, while if the button is held down the iPod skips to the "PowerSong" - an audio track chosen by the user, generally intended for motivation. In addition to the in-workout audio feedback, there are pre-recorded congratulations provided by Lance Armstrong, Tiger Woods, Joan Benoit Samuelson, and Paula Radcliffe whenever a user achieves a personal best (such as fastest mile, fastest 5K, fastest 10K, longest run yet) or reaches certain long-term milestones (such as 250 miles, 500 kilometers). This "celebrity feedback" is heard after the usual end-of-run statistics. While the Sports Kit can be used immediately after purchase, it will report more accurate results if it is calibrated before the first usage and then regularly afterwards. For calibration, the user finds a fixed known distance of at least 0.25 mile or 400 meters and then sets the Nike+ to calibration mode for the walk or run over that distance. When the walk or run is complete, the device calibrates itself and future workout reporting will reflect statistics closer to that individual user's workout style. Consumer Reports magazine tested the device and found it accurate as long as you keep an even pace. In workouts with varied pa
Read more →
Five safes

The Five Safes is a framework for helping make decisions about making effective use of data which is confidential or sensitive. It is mainly used to describe or design research access to statistical data held by government and health agencies, and by data archives such as the UK Data Service. It is not an internationally accepted standard. Two of the Five Safes refer to statistical disclosure control, and so the Five Safes is usually used to contrast statistical and non-statistical controls when comparing data management options. == Concept == The Five Safes proposes that data management decisions be considered as solving problems in five 'dimensions': projects, people, settings, data and outputs. The combination of the controls leads to 'safe use'. These are most commonly expressed as questions, for example: These dimensions are scales, not limits. That is, solutions can have a mix of more or fewer controls in each dimension, but the overall aim of 'safe use' independent of the particular mix. For example, a public use file available for open download cannot control who uses it, where or for what purpose, and so all the control (protection) must be in the data itself. In contrast, a file which is only accessed through a secure environment with certified users can contain very sensitive information: the non-statistical controls allow the data to be 'unsafe'. One academic likened the process to a graphic equalizer, where bass and treble can be combined independently to produce a sound the listener likes, which has proven to be a very useful metaphor. This 2023 Data Foundation webinar is an expert discussion of how the elements interact, including an excellent introductory representation. There is no 'order' to the Five Safes, in that one is necessarily more important than the others. However, Ritchie argued that the 'managerial' controls (projects, people, setting) should be addressed before the 'statistical' controls (data, output). The Five Safes concept is associated with other topics which developed from the same programme at ONS, although these are not necessarily implemented. Safe people is associated with 'active researcher management', while safe outputs is linked with principles-based output statistical disclosure control. The Five Safes is a positive framework, describing what is and is not. The EDRU ('evidence-based, default-open, risk-managed, user-centred') attitudinal model is sometimes used to give a normative context == The 'data access spectrum' == From 2003 the Five Safes was also represented in a simpler form as a 'Data Access Spectrum'. The non-data controls (project, people, setting, outputs) tend to work together, in that organisations often see these as a complementary set of restrictions on access. These can then be contrasted with choices about data anonymisation to present a linear representation of data access options. This presentation is consistent with the idea of 'data as a residual', as well as data protection laws of the time which often characterised data simply as anonymous or not anonymous. A similar idea had already been developed independently in 2001 by Chuck Humphrey of the Canadian RDC network, the 'continuum of access'. More recently, The Open Data Institute has developed a 'Data Spectrum toolkit' which includes industry-specific examples. == History and terminology == The Five Safes was devised in the winter of 2002/2003 by Felix Ritchie at the UK Office for National Statistics (ONS) to describe its secure remote-access Virtual Microdata Laboratory (VML). It was described at this time as the 'VML Security Model'. This was adopted by the NORC data enclave, and more widely in the US, as the 'portfolio model' (although this is now also used to refer to a slightly different legal/statistical/educational breakdown). In 2012 the framework as was still being referred to as the 'VML security model', but its increasing use among non-UK organisations led to the adoption of the more general and informative phrase 'Five Safes'. The original framework only had four safes (projects, people, settings and outputs): the framework was used to describe highly detailed data access through a secure environment, and so the 'data' dimension was irrelevant. From 2007 onwards, 'safe data' was included as the framework was used to a describe a wider range of ONS activities. As the US version was based upon the 2005 specification, some US iterations uses have the original four dimensions (eg). Some discussions, such as the OECD, use the term 'secure' instead 'safe'. However, the use of both these terms can cause presentational problems: less control in a particular dimension could be seen to imply 'unsafe users' or 'insecure settings', for example, which distracts from the main message. Hence, the Australian government uses the term "five data sharing principles". The 'Anonymisation Decision-Making Framework' uses a framework based on the Five Safes but relabelling "projects", "people", and "settings" as "governance", "agency" and "infrastructure", respectively; "Output" is omitted, and "safe use" becomes "functional anonymisation". There is no reference to the Five Safes or any associated literature. The Australian version was required to include references to the Five Safes, and presented it as an alternative without comment. == Application == The framework has had three uses: pedagogical, descriptive, and design. Since 2016, it has also been used, directly and indirectly in legislation. See for more detailed examples. === Pedagogy === The first significant use of the framework, other than internal administrative use, was to structure researcher training courses at the UK Office for National Statistics from 2003. UK Data Archive, Administrative Data Research Network, Eurostat, Statistics New Zealand, the Mexican National Institute of Statistics and Geography, NORC, Statistics Canada and the Australian Bureau of Statistics, amongst others, have also used this framework. Most of these courses are for researchers using restricted-access facilities; the Eurostat courses are unusual in that they are designed for all users of sensitive data. === Description === The framework is often used to describe existing data access solutions (e.g. UK HMRC Data Lab, UK Data Service, Statistics New Zealand) or planned/conceptualised ones (e.g. Eurostat in 2011). An early use was to help identify areas where ONS' still had 'irreducible risks' in its provision of secure remote access. The framework is mostly used for confidential social science data. To date it appears to have made little impact on medical research planning, although it is now included in the revised guidelines on implementing HIPAA regulations in the US, and by Cancer Research UK and the Health Foundation in the UK. It has also been used to describe a security model for the Scottish Health Informatics Programme. === Design === In general the Five Safes has been used to describe solutions post-factum, and to explain/justify choices made, but an increasing number of organisations have used the framework to design data access solutions. For example, the Hellenic Statistical Agency developed a data strategy built around the Five Safes in 2016; the UK Health Foundation used the Five Safes to design its data management and training programmes. Use in the private sector is less common but some organisations have incorporated the Five Safes into consulting services. In 2015 the UK Data Service organized a workshop to encourage data users from the academic and private sectors to think about how to manage confidential research data, using the Five Safes to demonstrate alternative options and best practice. Early adopters for strategic design use were in Australia: both the Australian Bureau of Statistics and the Australian Department of Social Service used the Five Safes as an ex ante design tool. In 2017 the Australian Productivity Commission recommended adopting a version of the framework to support cross-government data sharing and re-use. This underwent extensive consultation and culminated in the DAT Act 2022. Since 2020 the Five Safes has been the overriding framework for the design of new secure facilities and data sharing arrangements in the UK for public health and social sciences. This has been promoted by the Office for Statistics Regulation, the UK Statistics Authority, NHS DIgital, and the research funding bodies Administrative Data Research UK and DARE UK. === Regulation and legislation === Three laws have incorporated the Fives Safes. They are explicit in the South Australian Public Sector (Data Sharing) Act 2016, and implicit in the research provisions of the UK Digital Economy Act 2017. The Australian Data Availability and Transparency Act 2022 renames the Five Safes as the Five Data Sharing Principles.A 2025 statutory review of the DAT Act 2022 found "that the DAT Act has not been effective in achieving its objectives.". The review includes specific referen
Read more →
Online analytical processing

In computing, online analytical processing (OLAP) (), is an approach to quickly answer multi-dimensional analytical (MDA) queries. The term OLAP was created as a slight modification of the traditional database term online transaction processing (OLTP). OLAP is part of the broader category of business intelligence, which also encompasses relational databases, report writing and data mining. Typical applications of OLAP include business reporting for sales, marketing, management reporting, business process management (BPM), budgeting and forecasting, financial reporting and similar areas, with new applications emerging, such as agriculture. OLAP tools enable users to analyse multidimensional data interactively from multiple perspectives. OLAP consists of three basic analytical operations: consolidation (roll-up), drill-down, and slicing and dicing. Consolidation involves the aggregation of data that can be accumulated and computed in one or more dimensions. For example, all sales offices are rolled up to the sales department or sales division to anticipate sales trends. By contrast, the drill-down is a technique that allows users to navigate through the details. For instance, users can view the sales by individual products that make up a region's sales. Slicing and dicing is a feature whereby users can take out (slicing) a specific set of data of the OLAP cube and view (dicing) the slices from different viewpoints. These viewpoints are sometimes called dimensions (such as looking at the same sales by salesperson, or by date, or by customer, or by product, or by region, etc.). Databases configured for OLAP use a multidimensional data model, allowing for complex analytical and ad hoc queries with a rapid execution time. They borrow aspects of navigational databases, hierarchical databases and relational databases. OLAP is typically contrasted to OLTP (online transaction processing), which is generally characterized by much less complex queries, in a larger volume, to process transactions rather than for the purpose of business intelligence or reporting. Whereas OLAP systems are mostly optimized for read, OLTP has to process all kinds of queries (read, insert, update and delete). == Overview of OLAP systems == At the core of any OLAP system is an OLAP cube (also called a 'multidimensional cube' or a hypercube). It consists of numeric facts called measures that are categorized by dimensions. The measures are placed at the intersections of the hypercube, which is spanned by the dimensions as a vector space. The usual interface to manipulate an OLAP cube is a matrix interface, like Pivot tables in a spreadsheet program, which performs projection operations along the dimensions, such as aggregation or averaging. The cube metadata is typically created from a star schema or snowflake schema or fact constellation of tables in a relational database. Measures are derived from the records in the fact table and dimensions are derived from the dimension tables. Each measure can be thought of as having a set of labels, or meta-data associated with it. A dimension is what describes these labels; it provides information about the measure. A simple example would be a cube that contains a store's sales as a measure, and Date/Time as a dimension. Each Sale has a Date/Time label that describes more about that sale. For example: Sales Fact Table +-------------+----------+ | sale_amount | time_id | +-------------+----------+ Time Dimension | 930.10| 1234 |----+ +---------+-------------------+ +-------------+----------+ | | time_id | timestamp | | +---------+-------------------+ +---->| 1234 | 20080902 12:35:43 | +---------+-------------------+ === Multidimensional databases === Multidimensional structure is defined as "a variation of the relational model that uses multidimensional structures to organize data and express the relationships between data". The structure is broken into cubes and the cubes are able to store and access data within the confines of each cube. "Each cell within a multidimensional structure contains aggregated data related to elements along each of its dimensions". Even when data is manipulated it remains easy to access and continues to constitute a compact database format. The data still remains interrelated. Multidimensional structure is quite popular for analytical databases that use online analytical processing (OLAP) applications. Analytical databases use these databases because of their ability to deliver answers to complex business queries swiftly. Data can be viewed from different angles, which gives a broader perspective of a problem unlike other models. === Aggregations === It has been claimed that for complex queries OLAP cubes can produce an answer in around 0.1% of the time required for the same query on OLTP relational data. The most important mechanism in OLAP which allows it to achieve such performance is the use of aggregations. Aggregations are built from the fact table by changing the granularity on specific dimensions and aggregating up data along these dimensions, using an aggregate function (or aggregation function). The number of possible aggregations is determined by every possible combination of dimension granularities. The combination of all possible aggregations and the base data contains the answers to every query which can be answered from the data. Because usually there are many aggregations that can be calculated, often only a predetermined number are fully calculated; the remainder are solved on demand. The problem of deciding which aggregations (views) to calculate is known as the view selection problem. View selection can be constrained by the total size of the selected set of aggregations, the time to update them from changes in the base data, or both. The objective of view selection is typically to minimize the average time to answer OLAP queries, although some studies also minimize the update time. View selection is NP-complete. Many approaches to the problem have been explored, including greedy algorithms, randomized search, genetic algorithms and A search algorithm. Some aggregation functions can be computed for the entire OLAP cube by precomputing values for each cell, and then computing the aggregation for a roll-up of cells by aggregating these aggregates, applying a divide and conquer algorithm to the multidimensional problem to compute them efficiently. For example, the overall sum of a roll-up is just the sum of the sub-sums in each cell. Functions that can be decomposed in this way are called decomposable aggregation functions, and include COUNT, MAX, MIN, and SUM, which can be computed for each cell and then directly aggregated; these are known as self-decomposable aggregation functions. In other cases, the aggregate function can be computed by computing auxiliary numbers for cells, aggregating these auxiliary numbers, and finally computing the overall number at the end; examples include AVERAGE (tracking sum and count, dividing at the end) and RANGE (tracking max and min, subtracting at the end). In other cases, the aggregate function cannot be computed without analyzing the entire set at once, though in some cases approximations can be computed; examples include DISTINCT COUNT, MEDIAN, and MODE; for example, the median of a set is not the median of medians of subsets. These latter are difficult to implement efficiently in OLAP, as they require computing the aggregate function on the base data, either computing them online (slow) or precomputing them for possible rollouts (large space). == Types == OLAP systems have been traditionally categorized using the following taxonomy. === Multidimensional OLAP (MOLAP) === MOLAP (multi-dimensional online analytical processing) is the classic form of OLAP and is sometimes referred to as just OLAP. MOLAP stores this data in an optimized multi-dimensional array storage, rather than in a relational database. Some MOLAP tools require the pre-computation and storage of derived data, such as consolidations – the operation known as processing. Such MOLAP tools generally utilize a pre-calculated data set referred to as a data cube. The data cube contains all the possible answers to a given range of questions. As a result, they have a very fast response to queries. On the other hand, updating can take a long time depending on the degree of pre-computation. Pre-computation can also lead to what is known as data explosion. Other MOLAP tools, particularly those that implement the functional database model do not pre-compute derived data but make all calculations on demand other than those that were previously requested and stored in a cache. Advantages of MOLAP Fast query performance due to optimized storage, multidimensional indexing and caching. Smaller on-disk size of data compared to data stored in relational database due to compression techniques. Automated computation of higher-level aggregates of the data. It is very compact for low dimension data se
Read more →
Software diversity

Software diversity is a research field about the comprehension and engineering of diversity in the context of software. == Areas == The different areas of software diversity are discussed in surveys on diversity for fault-tolerance or for security. The main areas are: design diversity, n-version programming, data diversity for fault tolerance randomization software variability == Techniques == === Code transformations === It is possible to amplify software diversity through automated transformation processes that create synthetic diversity. A "multicompiler" is compiler embedding a diversification engine. A multi-variant execution environment (MVEE) is responsible for selecting the variant to execute and compare the output. Fred Cohen was among the very early promoters of such an approach. He proposed a series of rewriting and code reordering transformations that aim at producing massive quantities of different versions of operating systems functions. These ideas have been developed over the years and have led to the construction of integrated obfuscation schemes to protect key functions in large software systems. Another approach to increase software diversity of protection consists in adding randomness in certain core processes, such as memory loading. Randomness implies that all versions of the same program run differently from each other, which in turn creates a diversity of program behaviors. This idea was initially proposed and experimented by Stephanie Forrest and her colleagues. Recent work on automatic software diversity explores different forms of program transformations that slightly vary the behavior of programs. The goal is to evolve one program into a population of diverse programs that all provide similar services to users, but with a different code. This diversity of code enhances the protection of users against one single attack that could crash all programs at the same time. Transformation operators include: code layout randomization: reorder functions in code globals layout randomization: reorder and pad globals stack variable randomization: reorder variables in each stack frame heap layout randomization === Natural software diversity === It is known that some functionalities are available in multiple interchangeable implementations. This natural diversity can be exploited, for example it has been shown valuable to increase security in cloud systems.
Read more →
Sardinas–Patterson algorithm

In coding theory, the Sardinas–Patterson algorithm is a classical algorithm for determining in polynomial time whether a given variable-length code is uniquely decodable, named after August Albert Sardinas and George W. Patterson, who published it in 1953. The algorithm carries out a systematic search for a string which admits two different decompositions into codewords. As Knuth reports, the algorithm was rediscovered about ten years later in 1963 by Floyd, despite the fact that it was at the time already well known in coding theory. == Idea of the algorithm == Consider the code { a ↦ 1 , b ↦ 011 , c ↦ 01110 , d ↦ 1110 , e ↦ 10011 } {\displaystyle \{\,{\texttt {a}}\mapsto {\texttt {1}},{\texttt {b}}\mapsto {\texttt {011}},{\texttt {c}}\mapsto {\texttt {01110}},{\texttt {d}}\mapsto {\texttt {1110}},{\texttt {e}}\mapsto {\texttt {10011}}\,\}} . This code, which is based on an example by Berstel, is an example of a code which is not uniquely decodable, since the string 011101110011 can be interpreted as the sequence of codewords 01110 – 1110 – 011, but also as the sequence of codewords 011 – 1 – 011 – 10011. Two possible decodings of this encoded string are thus given by cdb and babe. In general, a codeword can be found by the following idea: In the first round, we choose two codewords x 1 {\displaystyle x_{1}} and y 1 {\displaystyle y_{1}} such that x 1 {\displaystyle x_{1}} is a prefix of y 1 {\displaystyle y_{1}} , that is, x 1 w = y 1 {\displaystyle x_{1}w=y_{1}} for some "dangling suffix" w {\displaystyle w} . If one tries first x 1 = 011 {\displaystyle x_{1}={\texttt {011}}} and y 1 = 01110 {\displaystyle y_{1}={\texttt {01110}}} , the dangling suffix is w = 10 {\displaystyle {\texttt {w}}={\texttt {10}}} . If we manage to find two sequences x 2 , … , x p {\displaystyle x_{2},\ldots ,x_{p}} and y 2 , … , y q {\displaystyle y_{2},\ldots ,y_{q}} of codewords such that x 2 ⋯ x p = w y 2 ⋯ y q {\displaystyle x_{2}\cdots x_{p}=wy_{2}\cdots y_{q}} , then we are finished: For then the string x = x 1 x 2 ⋯ x p {\displaystyle x=x_{1}x_{2}\cdots x_{p}} can alternatively be decomposed as y 1 y 2 ⋯ y q {\displaystyle y_{1}y_{2}\cdots y_{q}} , and we have found the desired string having at least two different decompositions into codewords. In the second round, we try out two different approaches: the first trial is to look for a codeword that has w as prefix. Then we obtain a new dangling suffix w, with which we can continue our search. If we eventually encounter a dangling suffix that is itself a codeword (or the empty word), then the search will terminate, as we know there exists a string with two decompositions. The second trial is to seek for a codeword that is itself a prefix of w. In our example, we have w = 10 {\displaystyle w={\texttt {10}}} , and the sequence 1 is a codeword. We can thus also continue with w = 0 {\displaystyle w={\texttt {0}}} as the new dangling suffix. == Precise description of the algorithm == The algorithm is described most conveniently using quotients of formal languages. In general, for two sets of strings D and N, the (left) quotient N − 1 D {\displaystyle N^{-1}D} is defined as the residual words obtained from D by removing some prefix in N. Formally, N − 1 D = { y ∣ x y ∈ D and x ∈ N } {\displaystyle N^{-1}D=\{\,y\mid xy\in D~{\textrm {and}}~x\in N\,\}} . Now let C {\displaystyle C} denote the (finite) set of codewords in the given code. The algorithm proceeds in rounds, where we maintain in each round not only one dangling suffix as described above, but the (finite) set of all potential dangling suffixes. Starting with round i = 1 {\displaystyle i=1} , the set of potential dangling suffixes will be denoted by S i {\displaystyle S_{i}} . The sets S i {\displaystyle S_{i}} are defined inductively as follows: S 1 = C − 1 C ∖ { ε } {\displaystyle S_{1}=C^{-1}C\setminus \{\varepsilon \}} . Here, the symbol ε {\displaystyle \varepsilon } denotes the empty word. S i + 1 = C − 1 S i ∪ S i − 1 C {\displaystyle S_{i+1}=C^{-1}S_{i}\cup S_{i}^{-1}C} , for all i ≥ 1 {\displaystyle i\geq 1} . The algorithm computes the sets S i {\displaystyle S_{i}} in increasing order of i {\displaystyle i} . As soon as one of the S i {\displaystyle S_{i}} contains a word from C or the empty word, then the algorithm terminates and answers that the given code is not uniquely decodable. Otherwise, once a set S i {\displaystyle S_{i}} equals a previously encountered set S j {\displaystyle S_{j}} with j < i {\displaystyle j Read more →
Best arm identification

Best arm identification (BAI) is a sequential one-player game where the player has to find the best action (arm) among a list of actions (arms) by collecting information in the most efficient way. It is a multi-armed bandit game as a player only gets information about an arm by playing it. The most common objective in multi-armed bandit games is to minimize the regret (i.e., play the best action as much as possible), but in BAI, the goal is to find the best arm as efficiently as possible. This problem naturally arises in scenarios such as adaptive clinical trials where the number of patients is limited and the quantification of the confidence in a treatment is important. It also arises in hyperparameter optimization where the goal is to find the optimal choice of hyperparameters for an algorithm with the smallest possible number of experiments, as it can be costly in terms of time, energy, or money. == Stochastic multi-armed bandit == The stochastic multi-armed bandit (MAB) is a sequential game with one player and K {\displaystyle K} actions (arms). Each arm has an unknown probability distribution associated with it. At each turn, the player has to choose one action and receive an observation from the probability distribution associated with the arm. The more you play an arm, the more you get information on its probability distribution. === Best arm identification === In BAI the goal is to find the arm that has the probability distribution with the highest mean. BAI may be either fixed confidence or fixed horizon. In a fixed-confidence game, a confidence level δ {\displaystyle \delta } is fixed at the beginning of the game and the goal is to find the best arm with this confidence level in as few turns as possible. In a fixed horizon game, the number of turns T {\displaystyle T} is fixed, and the goal is to find the best arm with the highest possible confidence in T {\displaystyle T} turns. === Math formalisation === We have one player and K {\displaystyle K} actions (arms). Behind each arm k ∈ { 1 , … , K } {\displaystyle k\in \{1,\ldots ,K\}} lies an unknown distribution ν k {\displaystyle \nu _{k}} with mean μ k {\displaystyle \mu _{k}} . Each distribution ν k {\displaystyle \nu _{k}} belongs to a known family D {\displaystyle {\mathcal {D}}} (such as the set of Gaussian distributions or Bernoulli distributions). At each time step t {\displaystyle t} , the player selects an arm a t {\displaystyle a_{t}} and observes an independent sample X t ∼ ν a t {\displaystyle X_{t}\sim \nu _{a_{t}}} from the corresponding distribution. We will note μ ∗ := max μ a {\displaystyle \mu ^{}:=\max \mu _{a}} the highest mean. An arm a {\displaystyle a} that satisfies μ a = μ ∗ {\displaystyle \mu _{a}=\mu ^{}} is called an optimal arm; otherwise it is called suboptimal arm. In best arm identification (BAI) the objective is to identify an optimal arm. Two main settings for BAI appear in the literature: Fixed confidence: In this setting, one typically assumes that there exists a unique optimal arm. A confidence level δ ∈ ( 0 , 1 ) {\displaystyle \delta \in (0,1)} is specified at the beginning. The algorithm must stop at some finite stopping time τ δ < + ∞ {\displaystyle \tau _{\delta }<+\infty } and return an arm a ^ τ δ {\displaystyle {\hat {a}}_{\tau _{\delta }}} such that the probability of error is bounded: P ( a ^ τ δ ≠ a ∗ ) ≤ δ {\displaystyle \mathbb {P} ({\hat {a}}_{\tau _{\delta }}\neq a^{})\leq \delta } . The objective is to minimize the expected sample complexity E [ τ δ ] {\displaystyle \mathbb {E} [\tau _{\delta }]} . Such a setting appears, for example, when a constraint on the confidence is required (for example, if we require a confidence level of 95%, so δ = 1 − 0.95 = 0.05 {\displaystyle \delta =1-0.95=0.05} ). Fixed horizon: In this setting, the number of samples T {\displaystyle T} is fixed in advance. The goal is to design an algorithm that minimizes the probability of misidentifying the optimal arm: P ( a ^ T ≠ a ∗ ) {\displaystyle \mathbb {P} ({\hat {a}}_{T}\neq a^{})} . This setting appears when the number of experiments is limited (for drug tests, the number of patients can be fixed in advance). === Example of simple modelling === In the case where we have K {\displaystyle K} treatments and we want to be sure with a confidence level of 95% which treatment is the best to heal a specific disease. Each treatment heals or does not heal the disease with a probability μ k {\displaystyle \mu _{k}} , which means that each distribution is a Bernoulli distribution, so D {\displaystyle {\mathcal {D}}} is the set of Bernoulli distributions. We can use a BAI algorithm to minimize E [ τ 0.05 ] {\displaystyle \mathbb {E} [\tau _{0.05}]} , the number of patients required to find the best treatment with probability 95%. == Applications == Best arm identification naturally arises in several practical domains: Adaptive clinical trials: The objective is to identify the most effective treatment based on sequentially collected patient data. Each treatment can be modeled as having an underlying distribution of outcomes. The goal is to identify the treatment with the highest expected outcome with high confidence (fixed confidence setting δ {\displaystyle \delta } ) while minimizing the number of drug test patients (minimise E [ τ δ ] {\displaystyle \mathbb {E} [\tau _{\delta }]} ), as it costs to pay patients for this and we would like to use as little as possible less effective drugs. Hyperparameter tuning: Selecting the best configuration for machine learning models efficiently by treating each hyperparameter setting as an arm. The goal is to find the best hyperparameter with as few experiments possible as experiments are costly in time and in energy == Fixed confidence level == In the fixed-confidence setting, the goal is to design an algorithm that identifies the best arm with a prescribed confidence level δ {\displaystyle \delta } while minimizing the expected number of samples. Any such algorithm requires two key components: Stopping rule: A decision criterion that determines when to stop sampling. Formally, this defines a stopping time τ δ {\displaystyle \tau _{\delta }} and returns an arm a ^ τ δ {\displaystyle {\hat {a}}_{\tau _{\delta }}} such that P ( a ^ τ δ ≠ a ⋆ ) ≤ δ {\displaystyle \mathbb {P} ({\hat {a}}_{\tau _{\delta }}\neq a^{\star })\leq \delta } and P ( τ δ < + ∞ ) = 1 {\displaystyle \mathbb {P} (\tau _{\delta }<+\infty )=1} . Sampling rule: A policy π {\displaystyle \pi } that, at each round t {\displaystyle t} , selects the next arm to sample a t {\displaystyle a_{t}} based on all previous observations ( a s , X s ) s < t {\displaystyle (a_{s},X_{s})_{s Read more →
Non-personal data

Non-Personal Data (NPD) is electronic data that does not contain any information that can be used to identify a natural person. Thus, it can either be data that has no personal information to begin with (such as weather data, stock prices, data from anonymous IoT sensors); or it is data that had personal data that was subsequently pseudoanonymized (for example, identifiable strings substituted with random strings) or anonymized (such as by irreversibly removing all personal data). NPD is part of the overall Data Governance Strategy of a region or country. While personal data are covered by Data Protection Legislation such as GDPR, other kinds of data would fall under the scope of NPD Regulation. == Importance of non-personal data == It has been pointed out that the future is data-driven. What this means is that much of the present innovation taking place in domains such as Machine Learning and Artificial Intelligence is fueled by data, which is needed for calibrating the complex models (comprising neural network-based as well as other kinds). The larger the volume, diversity and quality of the data, the higher is the quality of the model, leading to better predictions and explanations. However, there is a flip-side to data availability. The newly-emerging awareness of privacy and the consequent need for powerful Data Protection Regulations (such as GDPR) makes it increasingly difficult or impossible to obtain data in the quantities required. This is a contradiction, and the only way out would be to remove all personal data from data sets (either by Data anonymization or Pseudonymization coupled with noise injection, at which point it becomes NPD. Therefore, many innovation-friendly countries are coming out with regulatory regimes that would ensure that personal data is protected, while, at the same time, non-personal data can be extracted from personal data so that innovation is fostered. In other words, NPD 'unlocks' value that was locked away in data sets that have personally-identifiable information. It is expected that multiple NPD data sets will begin to be available on free or commercial basis from different providers once the regulations are in place. == Emerging regulatory frameworks == Non-Personal Data has significant uses that may be economic, social, political or security-related. Several countries and regions are in the process of regulating the use of NPD. In May 2019, the European Union operationalized its Regulation of the Free Flow of NPD. India announced a nine-member expert committee to make recommendations on the regulation of NPD in 2019, which published its first report in mid-2020. The report was opened for public comments, after which it was revised and published in December 2020. == Proposed NPD regulatory framework in India == The following were the objectives of the proposed Indian regulation as per the revised report: Sovereignty: India has rights over the data of India, its people and organisations. Benefit India: Benefits of data must accrue to India and its people. Benefits the world: Innovation, new models and algorithms for the world. Privacy: Misuse, reidentification and harms must be prevented. Simplicity: The regulations should be simple, digital and unambiguous. Innovation and entrepreneurship: The data should be freely available for innovation and entrepreneurship in India. == Concerns == The major concern in the use of NPD is if there are techniques (statistical or AI-based) by which multiple data sets can be used to extract personally-identifiable data.
Read more →
List of ARM Cortex-M development tools

This is a list of development tools for 32-bit ARM Cortex-M-based microcontrollers, which consists of Cortex-M0, Cortex-M0+, Cortex-M1, Cortex-M3, Cortex-M4, Cortex-M7, Cortex-M23, Cortex-M33, Cortex-M35P, Cortex-M52, Cortex-M55, and Cortex-M85 cores. == Development toolchains == IDE, compiler, linker, debugger, flashing (in alphabetical order): Ac6 System Workbench for STM32 (based on Eclipse and the GNU GCC toolchain with direct support for all ST-provided evaluation boards, Eval, Discovery and Nucleo, debug with ST-LINK) ARM Development Studio 5 by ARM Ltd. Atmel Studio by Atmel (based on Visual Studio and GNU GCC Toolchain) Code Composer Studio by Texas Instruments CoIDE by CooCox (note - website dead since 2018) Crossware Development Suite for ARM by Crossware CrossWorks for ARM by Rowley Dave by Infineon. For XMC processors only. Includes project wizard, detailed register decoding and a code library still under development. DRT by SOMNIUM Technologies. Based on GCC toolchain and proprietary linker technology. Available as a plugin for Atmel Studio and an Eclipse-based IDE. EmBitz (formerly Em::Blocks) – free, fast (non-eclipse) IDE for ST-LINK (live data updates), OpenOCD, including GNU Tools for ARM and project wizards for ST, Atmel, EnergyMicro etc. Embeetle IDE - free, fast (non-eclipse) IDE. Works both on Linux and Windows. emIDE by emide – free Visual Studio Style IDE including GNU Tools for ARM GNU ARM Eclipse – A family of Eclipse CDT extensions and tools for GNU ARM development GNU Tools (aka GCC) for ARM Embedded Processors by ARM Ltd – free GCC for bare metal IAR Embedded Workbench for ARM by IAR Systems ICC by ImageCraft Keil MDK-ARM by Keil LPCXpresso by NXP (formerly Red Suite by Code Red Technologies) MikroC by mikroe – mikroC MULTI by Green Hills Software, for all Arm 7, 9, Cortex-M, Cortex-R, Cortex-A Ride and RKit for ARM by Raisonance SEGGER Embedded Studio for ARM by Segger. SEGGER Ozone by Segger. STM32CubeIDE by STMicroelectronics - Combines STCubeMX with TrueSTUDIO into a single Eclipse style package Sourcery CodeBench by Mentor Graphics TASKING VX-Toolset by Altium TrueSTUDIO by Atollic Visual Studio by Microsoft as IDE, with GNU Tools as compiler/linker – e.g. supported by VisualGDB VXM Design's Buildroot toolchain for Cortex. It integrates GNU toolchain, Nuttx, filesystem and debugger/flasher in one build. winIDEA/winIDEAOpen by iSYSTEM YAGARTO – free GCC (no longer supported) Code::Blocks (EPS edition) (debug with ST-LINK no GDB and no OpenOCD required) IDE for Arduino ARM boards Arduino – IDE for Atmel SAM3X (Arduino Due) Energia – Arduino IDE for Texas Instruments Tiva and CC3200 Notes: == Debugging tools == JTAG and/or SWD debug interface host adapters (in alphabetical order): Black Magic Probe by 1BitSquared. CMSIS-DAP by Mbed. Crossconnect by Rowley Associates. DSTREAM by ARM Holdings Green Hills Probe and SuperTrace Probe by Green Hills Software. iTAG by iSYSTEM. I-jet by IAR Systems. Jaguar by Crossware. J-Link by Segger Supports JTAG and SWD. Supports ARM7, ARM9, ARM11, Cortex-A, Cortex-M, Cortex-R, Renesas RX, Microchip PIC32. Eclipse plug-in available. Supports GDB, RDI, Ozone debuggers. J-Trace by Segger. Supports JTAG, SWD, and ETM trace on Cortex-M. JTAGjet by Signum. LPC-LINK by Embedded Artists (for NXP) This is only embedded on NXP LPCXpresso development boards. LPC-LINK 2 by NXP. This device can be reconfigured to support 3 different protocols: J-LINK by Segger, CMSIS-DAP by ARM, Redlink by Code Red. Multilink debug probes, Cyclone in-system programming/debugging interfaces, and a GDB Server plug-in for Eclipse-based ARM IDEs by PEmicro. OpenOCD open source GDB server supports a variety of JTAG probes OpenOCD Eclipse plug-in available in GNU ARM Eclipse Plug-ins. AK-OPENJTAG by Artekit (Open JTAG-compatible). AK-LINK by Artekit. PEEDI by RONETIX Debug Probe by Raspberry Pi. RLink by Raisonance. ST-LINK/V2 by STMicroelectronics The ST-LINK/V2 debugger embedded on STM32 Nucleo and Discovery development boards can be converted to SEGGER J-LINK protocol. TRACE32 Debugger and ETM/ITM Trace by Lauterbach. ULINK by Keil. Debugging tools and/or debugging plug-ins (in alphabetical order): Memfault Error Analysis for post mortem debugging Percepio Tracealyzer, RTOS trace visualizer (with Eclipse plugin). Segger SystemView, RTOS trace visualizer. == Real-time operating systems == Commonly referred to as RTOS: == C/C++ software libraries == The following are free C/C++ libraries: ARM Cortex libraries: Cortex Microcontroller Software Interface Standard (CMSIS) libopencm3 (formerly called libopenstm32) libmaple for STM32F1 chips LPCOpen for NXP LPC chips Alternate C standard libraries: Bionic libc, dietlibc, EGLIBC, glibc, klibc, musl, Newlib, uClibc FAT file system libraries: EFSL, FatFs, Petit FatFs Fixed-point math libraries: libfixmath, fixedptc, FPMLib Encryption libraries: Comparison of TLS implementations wolfSSL == Non-C/C++ computer languages and software libraries ==
Read more →
QuickPar

QuickPar is a computer program that creates parchives used as verification and recovery information for a file or group of files, and uses the recovery information, if available, to attempt to reconstruct the originals from the damaged files and the PAR volumes. Designed for the Microsoft Windows operating system, in the past it was often used to recover damaged or missing files that have been downloaded through Usenet. QuickPar may also be used under Linux via Wine. There are two main versions of PAR files: PAR and PAR2. The PAR2 file format lifts many of its previous restrictions. QuickPar is freeware but not open-source. It uses the Reed-Solomon error correction algorithm internally to create the error correcting information. == Replacement == Since QuickPar hasn't been updated in 21 years, it is considered abandonware. Currently, MultiPar is accepted as the software that replaces QuickPar. MultiPar is actively being developed by Yutaka Sawada. == 64-bit versions == At present the command line version of QuickPar for Linux command line is available as a 64-bit version. None of the GUI versions available presently offer a 64-bit version.
Read more →
Kleene's algorithm

In theoretical computer science, in particular in formal language theory, Kleene's algorithm transforms a given nondeterministic finite automaton (NFA) into a regular expression. Together with other conversion algorithms, it establishes the equivalence of several description formats for regular languages. Alternative presentations of the same method include the "elimination method" attributed to Brzozowski and McCluskey, the algorithm of McNaughton and Yamada, and the use of Arden's lemma. == Algorithm description == According to Gross and Yellen (2004), the algorithm can be traced back to Kleene (1956). A presentation of the algorithm in the case of deterministic finite automata (DFAs) is given in Hopcroft and Ullman (1979). The presentation of the algorithm for NFAs below follows Gross and Yellen (2004). Given a nondeterministic finite automaton M = (Q, Σ, δ, q0, F), with Q = { q0,...,qn } its set of states, the algorithm computes the sets Rkij of all strings that take M from state qi to qj without going through any state numbered higher than k. Here, "going through a state" means entering and leaving it, so both i and j may be higher than k, but no intermediate state may. Each set Rkij is represented by a regular expression; the algorithm computes them step by step for k = -1, 0, ..., n. Since there is no state numbered higher than n, the regular expression Rn0j represents the set of all strings that take M from its start state q0 to qj. If F = { q1,...,qf } is the set of accept states, the regular expression Rn01 | ... | Rn0f represents the language accepted by M. The initial regular expressions, for k = -1, are computed as follows for i≠j: R−1ij = a1 | ... | am where qj ∈ δ(qi,a1), ..., qj ∈ δ(qi,am) and as follows for i=j: R−1ii = a1 | ... | am | ε where qi ∈ δ(qi,a1), ..., qi ∈ δ(qi,am) In other words, R−1ij mentions all letters that label a transition from i to j, and we also include ε in the case where i=j. After that, in each step the expressions Rkij are computed from the previous ones by Rkij = Rk-1ik (Rk-1kk) Rk-1kj | Rk-1ij Another way to understand the operation of the algorithm is as an "elimination method", where the states from 0 to n are successively removed: when state k is removed, the regular expression Rk-1ij, which describes the words that label a path from state i>k to state j>k, is rewritten into Rkij so as to take into account the possibility of going via the "eliminated" state k. By induction on k, it can be shown that the length of each expression Rkij is at most ⁠1/3⁠(4k+1(6s+7) - 4) symbols, where s denotes the number of characters in Σ. Therefore, the length of the regular expression representing the language accepted by M is at most ⁠1/3⁠(4n+1(6s+7)f - f - 3) symbols, where f denotes the number of final states. This exponential blowup is inevitable, because there exist families of DFAs for which any equivalent regular expression must be of exponential size. In practice, the size of the regular expression obtained by running the algorithm can be very different depending on the order in which the states are considered by the procedure, i.e., the order in which they are numbered from 0 to n. == Example == The automaton shown in the picture can be described as M = (Q, Σ, δ, q0, F) with the set of states Q = { q0, q1, q2 }, the input alphabet Σ = { a, b }, the transition function δ with δ(q0,a)=q0, δ(q0,b)=q1, δ(q1,a)=q2, δ(q1,b)=q1, δ(q2,a)=q1, and δ(q2,b)=q1, the start state q0, and set of accept states F = { q1 }. Kleene's algorithm computes the initial regular expressions as After that, the Rkij are computed from the Rk-1ij step by step for k = 0, 1, 2. Kleene algebra equalities are used to simplify the regular expressions as much as possible. Step 0 Step 1 Step 2 Since q0 is the start state and q1 is the only accept state, the regular expression R201 denotes the set of all strings accepted by the automaton.
Read more →
SQL/PSM

SQL/PSM (SQL/Persistent Stored Modules) is an ISO standard mainly defining an extension of SQL with a procedural language for use in stored procedures. Initially published in 1996 as an extension of SQL-92 (ISO/IEC 9075-4:1996, a version sometimes called PSM-96 or even SQL-92/PSM), SQL/PSM was later incorporated into the multi-part SQL:1999 standard, and has been part 4 of that standard since then, most recently in SQL:2023. The SQL:1999 part 4 covered less than the original PSM-96 because the SQL statements for defining, managing, and invoking routines were actually incorporated into part 2 SQL/Foundation, leaving only the procedural language itself as SQL/PSM. The SQL/PSM facilities are still optional as far as the SQL standard is concerned; most of them are grouped in Features P001-P008. SQL/PSM standardizes syntax and semantics for control flow, exception handling (called "condition handling" in SQL/PSM), local variables, assignment of expressions to variables and parameters, and (procedural) use of cursors. It also defines an information schema (metadata) for stored procedures. SQL/PSM is one language in which methods for the SQL:1999 structured types can be defined. The other is Java, via SQL/JRT. SQL/PSM is derived, seemingly directly, from Oracle's PL/SQL. Oracle developed PL/SQL and released it in 1991, basing the language on the US Department of Defense's Ada programming language. However, Oracle has maintained a distance from the standard in its documentation. IBM's SQL PL (used in DB2) and Mimer SQL's PSM were the first two products officially implementing SQL/PSM. It is commonly thought that these two languages, and perhaps also MySQL/MariaDB's procedural language, are closest to the SQL/PSM standard. However, a PostgreSQL addon implements SQL/PSM (alongside its other procedural languages like the PL/SQL-derived plpgsql), although it is not part of the core product. RDF functionality in OpenLink Virtuoso was developed entirely through SQL/PSM, combined with custom datatypes (e.g., ANY for handling URI and Literal relation objects), sophisticated indexing, and flexible physical storage choices (column-wise or row-wise).
Read more →
Model compression

Model compression is a machine learning technique for reducing the size of trained models. Large models can achieve high accuracy, but often at the cost of significant resource requirements. Compression techniques aim to compress models without significant performance reduction. Smaller models require less storage space, and consume less memory and compute during inference. Compressed models enable deployment on resource-constrained devices such as smartphones, embedded systems, edge computing devices, and consumer electronics computers. Efficient inference is also valuable for large corporations that serve large model inference over an API, allowing them to reduce computational costs and improve response times for users. Model compression is not to be confused with knowledge distillation, in which a smaller "student" model is trained to imitate the input-output behavior of a larger "teacher" model (as opposed to using the "teacher"'s trained parameters or the "teacher"'s training targets). == Techniques == Several techniques are employed for model compression. === Pruning === Pruning sparsifies a large model by setting some parameters to exactly zero. This effectively reduces the number of parameters. This allows the use of sparse matrix operations, which are faster than dense matrix operations. Pruning criteria can be based on magnitudes of parameters, the statistical pattern of neural activations, Hessian values, etc. === Quantization === Quantization reduces the numerical precision of weights and activations. For example, instead of storing weights as 32-bit floating-point numbers, they can be represented using 8-bit integers. Low-precision parameters take up less space, and takes less compute to perform arithmetic with. It is also possible to quantize some parameters more aggressively than others, so for example, a less important parameter can have 8-bit precision while another, more important parameter, can have 16-bit precision. Inference with such models requires mixed-precision arithmetic. Quantized models can also be used during training (rather than after training). PyTorch implements automatic mixed-precision (AMP), which performs autocasting, gradient scaling, and loss scaling. === Low-rank factorization === Weight matrices can be approximated by low-rank matrices. Let W {\displaystyle W} be a weight matrix of shape m × n {\displaystyle m\times n} . A low-rank approximation is W ≈ U V T {\displaystyle W\approx UV^{T}} , where U {\displaystyle U} and V {\displaystyle V} are matrices of shapes m × k , n × k {\displaystyle m\times k,n\times k} . When k {\displaystyle k} is small, this both reduces the number of parameters needed to represent W {\displaystyle W} approximately, and accelerates matrix multiplication by W {\displaystyle W} . Low-rank approximations can be found by singular value decomposition (SVD). The choice of rank for each weight matrix is a hyperparameter, and jointly optimized as a mixed discrete-continuous optimization problem. The rank of weight matrices may also be pruned after training, taking into account the effect of activation functions like ReLU on the implicit rank of the weight matrices. == Training == Model compression may be decoupled from training, that is, a model is first trained without regard for how it might be compressed, then it is compressed. However, it may also be combined with training. The "train big, then compress" method trains a large model for a small number of training steps (less than it would be if it were trained to convergence), then heavily compress the model. It is found that at the same compute budget, this method results in a better model than lightly compressed, small models. In Deep Compression, the compression has three steps. First loop (pruning): prune all weights lower than a threshold, then finetune the network, then prune again, etc. Second loop (quantization): cluster weights, then enforce weight sharing among all weights in each cluster, then finetune the network, then cluster again, etc. Third step: Use Huffman coding to losslessly compress the model. The SqueezeNet paper reported that Deep Compression achieved a compression ratio of 35 on AlexNet, and a ratio of ~10 on SqueezeNets.
Read more →
Jump-and-Walk algorithm

Jump-and-Walk is an algorithm for point location in triangulations (though most of the theoretical analysis were performed in 2D and 3D random Delaunay triangulations). Surprisingly, the algorithm does not need any preprocessing or complex data structures except some simple representation of the triangulation itself. The predecessor of Jump-and-Walk was due to Lawson (1977) and Green and Sibson (1978), which picks a random starting point S and then walks from S toward the query point Q one triangle at a time. But no theoretical analysis was known for these predecessors until after mid-1990s. Jump-and-Walk picks a small group of sample points and starts the walk from the sample point which is the closest to Q until the simplex containing Q is found. The algorithm was a folklore in practice for some time, and the formal presentation of the algorithm and the analysis of its performance on 2D random Delaunay triangulation was done by Devroye, Mucke and Zhu in mid-1990s (the paper appeared in Algorithmica, 1998). The analysis on 3D random Delaunay triangulation was done by Mucke, Saias and Zhu (ACM Symposium of Computational Geometry, 1996). In both cases, a boundary condition was assumed, namely, Q must be slightly away from the boundary of the convex domain where the vertices of the random Delaunay triangulation are drawn. In 2004, Devroye, Lemaire and Moreau showed that in 2D the boundary condition can be withdrawn (the paper appeared in Computational Geometry: Theory and Applications, 2004). Jump-and-Walk has been used in many famous software packages, e.g., QHULL, Triangle and CGAL.
Read more →
Media aggregation platform

A Media Aggregation Platform or Media Aggregation Portal (MAP) is an over the top service for distributing web-based streaming media content from multiple sources to a large audience. MAPs consist of networks of sources who host their own content which viewers can choose and access directly from a larger variety of content to choose from than a single source can offer. The service is used by content providers, looking to extend the reach of their content. Unlike multichannel video programming distributor (MVPD) or multiple-system operators (MSO), MAPs rely on the Internet rather than cables or satellite. As more network television channels have moved online in the early 21st century, joining web-native channels like Netflix, MAPs aggregate content the way that MSOs and MVPDs have used cable, and to a lesser extent satellite and IPTV infrastructure. There are companies that offer a similar service for free, including Yidio and StreamingMoviesRight, while others charge a subscription fee like as FreeCast Inc's Rabbit TV Plus. When compared with MSOs and MVPDs, MAP networks have much lower costs due to lack of physical infrastructure. The majority of revenue from MAP services are retained by the content creators, and revenue is instead collected from advertisements, pay-per-view, and subscription-based content offerings instead of licensing and reselling content. MAP service consumers interact and purchase content directly from its source, without the markup added by a middleman.
Read more →