AI Coding Scaffold

AI Coding Scaffold — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  Roadie (app)

  Roadie Inc. is an American package delivery company for business and private same-day, urgent and scheduled delivery in the United States. The company was founded in 2014 and launched its web and mobile apps in January 2015. As of September 2021, it reported having over 200,000 drivers covering more than 20,000 zip codes. Roadie states it matches gig drivers with deliveries that are directed along the routes they plan to travel. Major customers include The Home Depot, Walmart, Tractor Supply Company, Best Buy and Delta Air Lines. In September 2021, UPS entered into an agreement to acquire Roadie for an undisclosed amount with the transaction expected to be closed in the fourth quarter. == History == Roadie was founded by Marc Gorlin, a co-founder of Kabbage and founder of VerticalOne and Pretty Good Privacy, as a same-day and urgent delivery company in 2014. In January 2015, Roadie launched the first consumer to consumer (C2C) version of its app with a Series A funding round of $10 million. In February, Roadie announced a partnership with Waffle House to designate its restaurants "Roadie Roadhouses", offering a neutral meeting place for drivers and senders. Drivers receive free food and drink through the partnership. In May, late-night host Jimmy Kimmel discussed the Roadie-Waffle House relationship in an opening monologue on Jimmy Kimmel Live!. Roadie's driver network expanded significantly as a result. Roadie closed a Series B round of funding in June, raising $15 million, and its first business to business (B2B) app version launched that November. In 2015, Delta Air Lines signed an agreement with Roadie to deliver mishandled luggage, becoming Roadie’s first enterprise customer. Roadie launched a pilot program with Delta at Daytona Beach International Airport. Since then, the relationship has expanded to include over 70 airports around the United States and a first mile/last mile line haul relationship with Delta Cargo. In 2017, the company signed a deal with The Home Depot, also based in Atlanta, and in February 2019, closed a Series C round of funding. In October 2019, Roadie and Delta Cargo announced a partnership to create a same-day cross-country delivery offering, DASH Door-to-Door, the first of its kind from a U.S. passenger airline. Tractor Supply Company became the first general merchandise retailer to offer same-day delivery from every store in April 2020 through Roadie. In September 2021, UPS entered an agreement to acquire Roadie for an undisclosed amount. The transaction was expected to close in the fourth quarter of 2021. Roadies, which at the time reported having 200,000 operators serving over 20,000 ZIP Codes, was expected to continue operations under its name as a separate company with no transfer of packages between the UPS and Roadies networks. The relationship between the companies goes back several years with UPS being an early investor. Earlier in 2021, UPS had begun a pilot program testing same-day deliveries via Roadies. == Operations == === On-the-way model === Roadie’s app works by connecting drivers with senders, businesses or consumers who have items that need to be delivered. Deliveries within the app are referred to as "Gigs", which Gorlin said was inspired by live music road crews, also known as roadies. A sender creates a Gig on Roadie's web app or via its API. Drivers then review deliveries in their area on their mobile app and may choose to offer to take on individual or groups of deliveries along the same route. Gigs are then assigned to drivers by Roadie's algorithm. According to the company, this model encourages drivers to choose Gigs that align with their planned schedules and routes. Roadie calls this its "on-the-way" delivery model. The go-to-market approach taken by Roadie also differs from its competitors. Rather than launching in major cities and sequentially adding new markets city-by-city, Roadie launched nationwide from its inception. The company relies on retail and airline partners to drive volume of deliveries in individual markets, which in turn builds up a network of drivers in those areas, making it easier for small businesses and consumers to send deliveries as well. This strategy allows Roadie to reach smaller cities and towns in rural or exurban communities, traditionally difficult markets for delivery providers to serve. === Service lines === Roadie’s platform is most popular for same-day, on-demand or scheduled first mile/last mile delivery, especially delivery from stores and warehouses. Some retailers also use it for returns and reverse logistics, moving inventory, and hot shot shipping. Roadie operates 1-hour grocery delivery for Walmart, and delivers perishable food items for others including small, independent retailers. The on-the-way model complements the grocery industry’s just in time model, making last-mile deliveries that do not break the cold chain. === Cross-country same-day delivery === In October 2019, Roadie and Delta Cargo launched DASH Door-to-Door, a 24/7 door-to-door pick-up and delivery service. Roadie handles the first and last mile and Delta manages the line haul via passenger flights. The service launched originally from Atlanta to 55 cities and is an industry-first for a US commercial airline. === Promotion, awards and corporate citizenship === In September 2015, Roadie announced a partnership with Atlanta-based musician Ludacris, to promote the app. Following the devastation caused by flooding in Baton Rouge in 2016, Roadie offered free pickup and delivery for all deliveries traveling to and from the Baton Rouge area. In December 2020, Walmart named Roadie its top delivery partner for "Highest Driver Customer Satisfaction" and "Highest Net Promoter Score", after expanding into general merchandise deliveries as well as grocery that same year.

  Read more →
• 

  Ranking SVM

  In machine learning, a ranking SVM is a variant of the support vector machine algorithm, which is used to solve certain ranking problems (via learning to rank). The ranking SVM algorithm was published by Thorsten Joachims in 2002. The original purpose of the algorithm was to improve the performance of an internet search engine. However, it was found that ranking SVM also can be used to solve other problems such as Rank SIFT. == Description == The ranking SVM algorithm is a learning retrieval function that employs pairwise ranking methods to adaptively sort results based on how 'relevant' they are for a specific query. The ranking SVM function uses a mapping function to describe the match between a search query and the features of each of the possible results. This mapping function projects each data pair (such as a search query and clicked web-page, for example) onto a feature space. These features are combined with the corresponding click-through data (which can act as a proxy for how relevant a page is for a specific query) and can then be used as the training data for the ranking SVM algorithm. Generally, ranking SVM includes three steps in the training period: It maps the similarities between queries and the clicked pages onto a certain feature space. It calculates the distances between any two of the vectors obtained in step 1. It forms an optimization problem which is similar to a standard SVM classification and solves this problem with the regular SVM solver. == Background == === Ranking method === Suppose C {\displaystyle \mathbb {C} } is a data set containing N {\displaystyle N} elements c i {\displaystyle c_{i}} . r {\displaystyle r} is a ranking method applied to C {\displaystyle \mathbb {C} } . Then the r {\displaystyle r} in C {\displaystyle \mathbb {C} } can be represented as a N × N {\displaystyle N\times N} binary matrix. If the rank of c i {\displaystyle c_{i}} is higher than the rank of c j {\displaystyle c_{j}} , i.e. r c i < r c j {\displaystyle r\ c_{i} Read more →

• 

  State complexity

  State complexity is an area of theoretical computer science dealing with the size of abstract automata, such as different kinds of finite automata. The classical result in the area is that simulating an n {\displaystyle n} -state nondeterministic finite automaton by a deterministic finite automaton requires exactly 2 n {\displaystyle 2^{n}} states in the worst case. == Transformation between variants of finite automata == Finite automata can be deterministic and nondeterministic, one-way (DFA, NFA) and two-way (2DFA, 2NFA). Other related classes are unambiguous (UFA), self-verifying (SVFA) and alternating (AFA) finite automata. These automata can also be two-way (2UFA, 2SVFA, 2AFA). All these machines can accept exactly the regular languages. However, the size of different types of automata necessary to accept the same language (measured in the number of their states) may be different. For any two types of finite automata, the state complexity tradeoff between them is an integer function f {\displaystyle f} where f ( n ) {\displaystyle f(n)} is the least number of states in automata of the second type sufficient to recognize every language recognized by an n {\displaystyle n} -state automaton of the first type. The following results are known. NFA to DFA: 2 n {\displaystyle 2^{n}} states. This is the subset construction by Rabin and Scott, proved optimal by Lupanov. UFA to DFA: 2 n {\displaystyle 2^{n}} states, see Leung, An earlier lower bound by Schmidt was smaller. NFA to UFA: 2 n − 1 {\displaystyle 2^{n}-1} states, see Leung. There was an earlier smaller lower bound by Schmidt. SVFA to DFA: Θ ( 3 n / 3 ) {\displaystyle \Theta (3^{n/3})} states, see Jirásková and Pighizzini 2DFA to DFA: n ( n n − ( n − 1 ) n ) {\displaystyle n(n^{n}-(n-1)^{n})} states, see Kapoutsis. Earlier construction by Shepherdson used more states, and an earlier lower bound by Moore was smaller. 2DFA to NFA: ( 2 n n + 1 ) = O ( 4 n n ) {\displaystyle {\binom {2n}{n+1}}=O({\frac {4^{n}}{\sqrt {n}}})} , see Kapoutsis. Earlier construction by Birget used more states. 2NFA to NFA: ( 2 n n + 1 ) {\displaystyle {\binom {2n}{n+1}}} , see Kapoutsis. 2NFA to NFA accepting the complement: O ( 4 n ) {\displaystyle O(4^{n})} states, see Vardi. AFA to DFA: 2 2 n {\displaystyle 2^{2^{n}}} states, see Chandra, Kozen and Stockmeyer. AFA to NFA: 2 n {\displaystyle 2^{n}} states, see Fellah, Jürgensen and Yu. 2AFA to DFA: 2 n 2 n {\displaystyle 2^{n2^{n}}} , see Ladner, Lipton and Stockmeyer. 2AFA to NFA: 2 Θ ( n log ⁡ n ) {\displaystyle 2^{\Theta (n\log n)}} , see Geffert and Okhotin. === The 2DFA vs. 2NFA problem and logarithmic space === It is an open problem whether all 2NFAs can be converted to 2DFAs with polynomially many states, i.e. whether there is a polynomial p ( n ) {\displaystyle p(n)} such that for every n {\displaystyle n} -state 2NFA there exists a p ( n ) {\displaystyle p(n)} -state 2DFA. The problem was raised by Sakoda and Sipser, who compared it to the P vs. NP problem in the computational complexity theory. Berman and Lingas discovered a formal relation between this problem and the L vs. NL open problem. This relation was further elaborated by Kapoutsis. == State complexity of operations for finite automata == Given a binary regularity-preserving operation on languages ∘ {\displaystyle \circ } and a family of automata X (DFA, NFA, etc.), the state complexity of ∘ {\displaystyle \circ } is an integer function f ( m , n ) {\displaystyle f(m,n)} such that for each m-state X-automaton A and n-state X-automaton B there is an f ( m , n ) {\displaystyle f(m,n)} -state X-automaton for L ( A ) ∘ L ( B ) {\displaystyle L(A)\circ L(B)} , and for all integers m, n there is an m-state X-automaton A and an n-state X-automaton B such that every X-automaton for L ( A ) ∘ L ( B ) {\displaystyle L(A)\circ L(B)} must have at least f ( m , n ) {\displaystyle f(m,n)} states. Analogous definition applies for operations with any number of arguments. The first results on state complexity of operations for DFAs were published by Maslov and by Yu, Zhuang and Salomaa. Holzer and Kutrib pioneered the state complexity of operations on NFA. The known results for basic operations are listed below. === Union === If language L 1 {\displaystyle L_{1}} requires m states and language L 2 {\displaystyle L_{2}} requires n states, how many states does L 1 ∪ L 2 {\displaystyle L_{1}\cup L_{2}} require? DFA: m n {\displaystyle mn} states, see Maslov and Yu, Zhuang and Salomaa. NFA: m + n + 1 {\displaystyle m+n+1} states, see Holzer and Kutrib. UFA: at least min ( n , m ) Ω ( log ⁡ ( min ( n , m ) ) ) {\displaystyle \min(n,m)^{\Omega (\log(\min(n,m)))}} ; between m n + m + n {\displaystyle mn+m+n} and m + n m 2 0.79 m {\displaystyle m+nm2^{0.79m}} states, see Jirásek, Jirásková and Šebej. SVFA: m n {\displaystyle mn} states, see Jirásek, Jirásková and Szabari. 2DFA: between m + n {\displaystyle m+n} and 4 m + n + 4 {\displaystyle 4m+n+4} states, see Kunc and Okhotin. 2NFA: m + n {\displaystyle m+n} states, see Kunc and Okhotin. === Intersection === How many states does L 1 ∩ L 2 {\displaystyle L_{1}\cap L_{2}} require? DFA: m n {\displaystyle mn} states, see Maslov and Yu, Zhuang and Salomaa. NFA: m n {\displaystyle mn} states, see Holzer and Kutrib. UFA: m n {\displaystyle mn} states, see Jirásek, Jirásková and Šebej. SVFA: m n {\displaystyle mn} states, see Jirásek, Jirásková and Szabari. 2DFA: between m + n {\displaystyle m+n} and m + n + 1 {\displaystyle m+n+1} states, see Kunc and Okhotin. 2NFA: between m + n {\displaystyle m+n} and m + n + 1 {\displaystyle m+n+1} states, see Kunc and Okhotin. === Complementation === If language L requires n states then how many states does its complement require? DFA: n {\displaystyle n} states, by exchanging accepting and rejecting states. NFA: 2 n {\displaystyle 2^{n}} states, see Birget. or Jirásková UFA: at least n Ω ~ ( log ⁡ n ) {\displaystyle n^{{\tilde {\Omega }}(\log n)}} states, see Göös, Kiefer and Yuan, (this follows an earlier bound by Raskin); and at most n + 1 ⋅ 2 0.5 n {\displaystyle {\sqrt {n+1}}\cdot 2^{0.5n}} states, see Indzhev and Kiefer. SVFA: n {\displaystyle n} states, by exchanging accepting and rejecting states. 2DFA: at least n {\displaystyle n} and at most 4 n {\displaystyle 4n} states, see Geffert, Mereghetti and Pighizzini. === Concatenation === How many states does L 1 L 2 = { w 1 w 2 ∣ w 1 ∈ L 1 , w 2 ∈ L 2 } {\displaystyle L_{1}L_{2}=\{w_{1}w_{2}\mid w_{1}\in L_{1},w_{2}\in L_{2}\}} require? DFA: m ⋅ 2 n − 2 n − 1 {\displaystyle m\cdot 2^{n}-2^{n-1}} states, see Maslov and Yu, Zhuang and Salomaa. NFA: m + n {\displaystyle m+n} states, see Holzer and Kutrib. UFA: 3 4 2 m + n − 1 {\displaystyle {\frac {3}{4}}2^{m+n}-1} states, see Jirásek, Jirásková and Šebej. SVFA: Θ ( 3 n / 3 2 m ) {\displaystyle \Theta (3^{n/3}2^{m})} states, see Jirásek, Jirásková and Szabari. 2DFA: at least 2 Ω ( n ) log ⁡ m {\displaystyle {\frac {2^{\Omega (n)}}{\log m}}} and at most 2 m m + 1 ⋅ 2 n n + 1 {\displaystyle 2m^{m+1}\cdot 2^{n^{n+1}}} states, see Jirásková and Okhotin. === Kleene star === DFA: 3 4 2 n {\displaystyle {\frac {3}{4}}2^{n}} states, see Maslov and Yu, Zhuang and Salomaa. NFA: n + 1 {\displaystyle n+1} states, see Holzer and Kutrib. UFA: 3 4 2 n {\displaystyle {\frac {3}{4}}2^{n}} states, see Jirásek, Jirásková and Šebej. SVFA: 3 4 2 n {\displaystyle {\frac {3}{4}}2^{n}} states, see Jirásek, Jirásková and Szabari. 2DFA: at least 1 n 2 n 2 − 1 {\displaystyle {\frac {1}{n}}2^{{\frac {n}{2}}-1}} and at most 2 O ( n n + 1 ) {\displaystyle 2^{O(n^{n+1})}} states, see Jirásková and Okhotin. === Reversal === DFA: 2 n {\displaystyle 2^{n}} states, see Mirkin, Leiss, and Yu, Zhuang and Salomaa. NFA: n + 1 {\displaystyle n+1} states, see Holzer and Kutrib. UFA: n {\displaystyle n} states. SVFA: 2 n + 1 {\displaystyle 2n+1} states, see Jirásek, Jirásková and Szabari. 2DFA: between n + 1 {\displaystyle n+1} and n + 2 {\displaystyle n+2} states, see Jirásková and Okhotin. == Finite automata over a unary alphabet == State complexity of finite automata with a one-letter (unary) alphabet, pioneered by Chrobak, is different from the multi-letter case. Let g ( n ) = e Θ ( n ln ⁡ n ) {\displaystyle g(n)=e^{\Theta ({\sqrt {n\ln n}})}} be Landau's function. === Transformation between models === For a one-letter alphabet, transformations between different types of finite automata are sometimes more efficient than in the general case. NFA to DFA: g ( n ) + O ( n 2 ) {\displaystyle g(n)+O(n^{2})} states, see Chrobak. 2DFA to DFA: g ( n ) + O ( n ) {\displaystyle g(n)+O(n)} states, see Chrobak and Kunc and Okhotin. 2NFA to DFA: O ( g ( n ) ) {\displaystyle O(g(n))} states, see Mereghetti and Pighizzini. and Geffert, Mereghetti and Pighizzini. NFA to 2DFA: at most O ( n 2 ) {\displaystyle O(n^{2})} states, see Chrobak. 2NFA to 2DFA: at most n O ( log ⁡ n ) {\displaystyle n^{O(\log n)}} states, proved by implementing the method of Savitch's theorem, see

  Read more →
• 

  How to Choose an AI Content Generator

  Curious about the best AI content generator? An AI content generator is software that uses machine learning to help you get more done — it combines speed, accuracy, and an interface that just works. Hands-on testing shows real-world results vary, so a short free trial is the smartest way to decide. Whether you are a beginner or a pro, the right AI content generator slots into your workflow and pays for itself fast. Read on for hands-on impressions, pricing tiers, and the standout features that matter.

  Read more →
• 

  Deep Learning Super Sampling

  Deep Learning Super Sampling (DLSS) is a suite of real-time deep learning image enhancement and upscaling technologies developed by Nvidia that are available in a number of video games. The goal of these technologies is to allow the majority of the graphics pipeline to run at a lower resolution for increased performance, and then infer a higher resolution image from this that approximates the same level of detail as if the image had been rendered at this higher resolution. This allows for higher graphical settings or frame rates for a given output resolution, depending on user preference. All generations of DLSS are available on all RTX-branded cards from Nvidia in supported titles. However, the Frame Generation feature is only supported on RTX 40 series GPUs or newer and Multi Frame Generation is only available on 50 series GPUs. == History == Nvidia advertised DLSS as a key feature of GeForce RTX 20 series GPUs when they launched in September 2018. At that time, the results were limited to a few video games, namely Battlefield V, or Metro Exodus, because the algorithm had to be trained specifically on each game on which it was applied and the results were usually not as good as simple resolution upscaling. In 2019, Control shipped with ray tracing and an image processing algorithm that approximated DLSS, which did not use the Tensor Cores. In April 2020, Nvidia advertised and shipped an improved version of DLSS named DLSS 2 with driver version 445.75. DLSS 2.0 was available for a few existing games including Control and Wolfenstein: Youngblood, and would later be added to many newly released games and game engines such as Unreal Engine and Unity. This time Nvidia said that it used the Tensor Cores again, and that the AI did not need to be trained specifically on each game. Despite sharing the DLSS branding, the two iterations of DLSS differ significantly and are not backwards-compatible. In January 2025, Nvidia stated that there are over 540 games and apps supporting DLSS, and that over 80% of Nvidia RTX users activate DLSS. In March 2025, there were more than 100 games that support DLSS 4, according to Nvidia. By May 2025, over 125 games supported DLSS 4. The first video game console to use DLSS, the Nintendo Switch 2, was released on June 5, 2025. Nvidia announced DLSS 4.5 at CES 2026. In January 2026, Nvidia stated that over 250 games and applications support Multi Frame Generation. On March 16, 2026, at GTC 2026, Nvidia CEO Jensen Huang presented DLSS 5, a real-time AI model based on neural rendering that realistically enhances lighting and material surfaces at up to 4K resolution while retaining the developer's intended art style. It is planned to release in fall of 2026. In a blog post on its website, Nvidia has announced that DLSS 5 will be available in such games as Assassin's Creed Shadows, Delta Force, Hogwarts Legacy, Naraka: Bladepoint, Phantom Blade Zero, Resident Evil Requiem, Starfield, The Elder Scrolls IV: Oblivion Remastered, and more. On May 31, 2026, Nvidia announced an updated version of Ray Reconstruction for DLSS 4.5 in a blog post, scheduled for release on all RTX GPUs in August of the same year. They said it is designed to better embed spatial awareness into scenes and analyze engine data on movements and lighting conditions, resulting in a sharper, more stable, and less noisy image. === Release timeline === == Technology == === DLSS 1 === The first iteration of DLSS is a predominantly spatial image upscaler with two stages, both relying on convolutional auto-encoder neural networks. The first step is an image enhancement network which uses the current frame and motion vectors to perform edge enhancement, and spatial anti-aliasing. The second stage is an image upscaling step which uses the single raw, low-resolution frame to upscale the image to the desired output resolution. Using just a single frame for upscaling means the neural network itself must generate a large amount of new information to produce the high-resolution output, which can result in slight hallucinations such as leaves that differ in style to the source content. The neural networks are trained on a per-game basis by generating a "perfect frame" using traditional supersampling to 64 samples per pixel, as well as the motion vectors for each frame. The data collected must be as comprehensive as possible, including as many levels, times of day, graphical settings, resolutions, etc. as possible. This data is also augmented using common augmentations such as rotations, colour changes, and random noise to help generalize the test data. Training is performed on Nvidia's Saturn V supercomputer. This first iteration received a mixed response, with many criticizing the often soft appearance and artifacts along with glitches in certain situations; likely a side effect of the limited data from only using a single frame input to the neural networks which could not be trained to perform optimally in all scenarios and edge-cases. Nvidia also demonstrated the ability for the auto-encoder networks to learn the ability to recreate depth-of-field and motion blur, although this functionality has never been included in a publicly released product. === DLSS 2 === DLSS 2 is a temporal anti-aliasing upsampling (TAAU) implementation, using data from previous frames extensively through sub-pixel jittering to resolve fine detail and reduce aliasing. The data DLSS 2 collects includes: the raw low-resolution input, motion vectors, depth buffers, and exposure / brightness information. It can also be used as a simpler TAA implementation where the image is rendered at 100% resolution, rather than being upsampled by DLSS, Nvidia brands this as DLAA (Deep Learning Anti-Aliasing). TAA(U) is used in many modern video games and game engines; however, all previous implementations have used some form of manually written heuristics to prevent temporal artifacts such as ghosting and flickering. One example of this is neighborhood clamping which forcefully prevents samples collected in previous frames from deviating too much compared to nearby pixels in newer frames. This helps to identify and fix many temporal artifacts, but deliberately removing fine details in this way is analogous to applying a blur filter, and thus the final image can appear blurry when using this method. DLSS 2 uses a convolutional auto-encoder neural network trained to identify and fix temporal artifacts, instead of manually programmed heuristics as mentioned above. Because of this, DLSS 2 can generally resolve detail better than other TAA and TAAU implementations, while also removing most temporal artifacts. This is why DLSS 2 can sometimes produce a sharper image than rendering at higher, or even native resolutions using traditional TAA. However, no temporal solution is perfect, and artifacts (ghosting in particular) are still visible in some scenarios when using DLSS 2. Because temporal artifacts occur in most art styles and environments in broadly the same way, the neural network that powers DLSS 2 does not need to be retrained when being used in different games. Despite this, Nvidia does frequently ship new minor revisions of DLSS 2 with new titles, so this could suggest some minor training optimizations may be performed as games are released, although Nvidia does not provide changelogs for these minor revisions to confirm this. The main advancements compared to DLSS 1 include: Significantly improved detail retention, a generalized neural network that does not need to be re-trained per-game, and ~2x less overhead (~1–2 ms vs ~2–4 ms). It should also be noted that forms of TAAU such as DLSS 2 are not upscalers in the same sense as techniques such as ESRGAN or DLSS 1, which attempt to create new information from a low-resolution source; instead, TAAU works to recover data from previous frames, rather than creating new data. In practice, this means low resolution textures in games will still appear low-resolution when using current TAAU techniques. This is why Nvidia recommends game developers use higher resolution textures than they would normally for a given rendering resolution by applying a mip-map bias when DLSS 2 is enabled. === DLSS 3 === Augments DLSS 2 with improved image quality and the introduction of a new motion interpolation feature, called Frame Generation. The DLSS Frame Generation algorithm takes two rendered frames from the rendering pipeline and generates a new frame that smoothly transitions between them. For every frame rendered, one additional frame is generated. DLSS 3.0 makes use of a new generation Optical Flow Accelerator (OFA) included in the Ada Lovelace architecture of GeForce RTX 40 series GPUs and with that is exclusive to them. The new OFA is said to be faster and more accurate than the one already available in previous Turing and Ampere RTX GPUs. === DLSS 3.5 === DLSS 3.5 adds Ray Reconstruction, replacing multiple denoising algorithms with a single AI model trained o

  Read more →
• 

  The Best Free AI Analytics Tool for Beginners

  Trying to pick the best AI analytics tool? An AI analytics tool is software that uses machine learning to help you get more done — it scales effortlessly from a single task to thousands. The best picks balance beginner-friendly simplicity with the depth power users need, and they ship updates often. Whether you are a beginner or a pro, the right AI analytics tool slots into your workflow and pays for itself fast. This guide breaks down the top picks, their pros and cons, and who each one is best for.

  Read more →
• 

  AI Customer-support Bots Reviews: What Actually Works in 2026

  Looking for the best AI customer-support bot? An AI customer-support bot is software that uses machine learning to help you get more done — it can save you hours every week by automating repetitive work. Most options offer a generous free tier, with paid plans unlocking higher limits, faster processing, and team features. Whether you are a beginner or a pro, the right AI customer-support bot slots into your workflow and pays for itself fast. Read on for hands-on impressions, pricing tiers, and the standout features that matter.

  Read more →
• 

  RAMnets

  RAMnets is one of the oldest practical neurally inspired classification algorithms. The RAMnets is also known as a type of "n-tuple recognition method" or "weightless neural network". == Algorithm == Consider (let us say N) sets of n distinct bit locations are selected randomly. These are the n-tuples. The restriction of a pattern to an n-tuple can be regarded as an n-bit number which, together with the identity of the n-tuple, constitutes a `feature' of the pattern. The standard n-tuple recognizer operates simply as follows: A pattern is classified as belonging to the class for which it has the most features in common with at least one training pattern of that class. This is the Θ {\displaystyle \Theta } = 0 case of a more general rule whereby the class assigned to unclassified pattern u is a c r g m a x ( ∑ i = 1 N Θ ( ∑ v ∈ D c δ ( α i ( u ) , α i ( v ) ) ) ) {\displaystyle {\begin{aligned}{\underset {c}{a}}rgmax(\sum _{i=1}^{N}\Theta (\sum _{v\in D_{c}}\delta (\alpha _{i}(u),\alpha _{i}(v))))\end{aligned}}} where Dc is the set of training patterns in class c, Θ ( x ) {\displaystyle \Theta (x)} = x for 0 ≤ x ≤ θ {\displaystyle 0\leq x\leq \theta } , Θ ( x ) = θ {\displaystyle \Theta (x)=\theta } for x ≥ θ {\displaystyle x\geq \theta } , δ i , j {\displaystyle \delta _{i,j}} is the Kronecker delta( δ i , j {\displaystyle \delta _{i,j}} =1 if i=j and 0 otherwise.)and ( α i ( u ) ) {\displaystyle (\alpha _{i}(u))} is the ith feature of the pattern u: ∑ j = 0 n − 1 u η i ( j ) 2 j {\displaystyle \sum _{j=0}^{n-1}u_{\eta }i(j)2^{j}} Here uk is the kth bit of u and u η i ( j ) {\displaystyle u_{\eta }i(j)} is the jth bit location of the ith n-tuple. With C classes to distinguish, the system can be implemented as a network of NC nodes, each of which is a random access memory (RAM); hence the term RAMnet. The memory content m c i α {\displaystyle m_{ci\alpha }} at address α {\displaystyle \alpha } of the ith node allocated to class c is set to m c i α {\displaystyle m_{ci\alpha }} = Θ ( ∑ v ∈ D c δ ( α , α i ( v ) ) ) {\displaystyle \Theta (\sum _{v\in D_{c}}\delta (\alpha ,\alpha _{i}(v)))} In the usual θ {\displaystyle \theta } = 1 case, the 1-bit content of m c i α {\displaystyle m_{ci\alpha }} is set if any pattern of Dc has feature α {\displaystyle \alpha } and unset otherwise. Recognition is accomplished by summing the contents of the nodes of each class at the addresses given by the features of the unclassified pattern. That is, pattern u is assigned to class a c r g m a x ( ∑ i = 1 N m c i α ( u ) ) {\displaystyle {\begin{aligned}{\underset {c}{a}}rgmax(\sum _{i=1}^{N}m_{ci\alpha }(u))\end{aligned}}} == RAM-discriminators and WiSARD == The RAMnets formed the basis of a commercial product known as WiSARD (Wilkie, Stonham and Aleksander Recognition Device) was the first artificial neural network machine to be patented. A RAM-discriminator consists of a set of X one-bit word RAMs with n inputs and a summing device (Σ). Any such RAM-discriminator can receive a binary pattern of X⋅n bits as input. The RAM input lines are connected to the input pattern by means of a biunivocal pseudo-random mapping. The summing device enables this network of RAMs to exhibit – just like other ANN models based on synaptic weights – generalization and noise tolerance. In order to train the discriminator one has to set all RAM memory locations to 0 and choose a training set formed by binary patterns of X⋅n bits. For each training pattern, a 1 is stored in the memory location of each RAM addressed by this input pattern. Once the training of patterns is completed, RAM memory contents will be set to a certain number of 0's and 1's. The information stored by the RAM during the training phase is used to deal with previous unseen patterns. When one of these is given as input, the RAM memory contents addressed by the input pattern are read and summed by Σ. The number r thus obtained, which is called the discriminator response, is equal to the number of RAMs that output 1. r reaches the maximum X if the input belongs to the training set. r is equal to 0 if no n-bit component of the input pattern appears in the training set (not a single RAM outputs 1). Intermediate values of r express a kind of “similarity measure” of the input pattern with respect to the patterns in the training set. A system formed by various RAM-discriminators is called WiSARD. Each RAM-discriminator is trained on a particular class of patterns, and classification by the multi-discriminator system is performed in the following way. When a pattern is given as input, each RAM-discriminator gives a response to that input. The various responses are evaluated by an algorithm which compares them and computes the relative confidence c of the highest response (e.g., the difference d between the highest response and the second highest response, divided by the highest response). A schematic representation of a RAM-discriminator and a 10 RAM-discriminator WiSARD is shown in Figure 1.

  Read more →
• 

  Scientific Working Group – Imaging Technology

  The Scientific Working Group on Imaging Technology was convened by the Federal Bureau of Investigation in 1997 to provide guidance to law enforcement agencies and others in the criminal justice system regarding the best practices for photography, videography, and video and image analysis. This group was terminated in 2015. == History == As technology has advanced through the years, law enforcement has needed to stay abreast of emerging technological advances and use these in the investigation of crime. A factor that is considered when new technology is used in these investigations is the determination of whether the use of that new technology will be admissible in court. The judicial system in the United States currently has two standards used in the determination of admissibility of testimony regarding scientific evidence; the Daubert Standard and the Frye Standard. These standards guide the courts in the admissibility of testimony derived from the use of new technologies and scientific techniques. The Federal Bureau of Investigation (FBI), seeking to address possible admissibility issues with such testimony, established Scientific Working Groups starting with the Scientific Working Group on DNA Analysis and Methods (SWGDAM) in 1988. The goal of these groups is to open lines of communication between law enforcement agencies and forensic laboratories around the world while providing guidance on the use of new and innovative technologies and techniques. This guidance can lead to admissibility of evidence and/or testimony, provided proper methods in the collection of evidence and its analysis are employed. In 2009, the National Academy of Sciences released a report entitled, "Strengthening Forensic Science in the United States: A Path Forward." This report addresses many topics including challenges and disparities facing the forensic science community, standardization, certification of practitioners and accreditation of their respective entities, problems related to the interpretation of forensic evidence, the need for research, and the admission of forensic science evidence in litigation. This report mentions the Scientific Working Groups and their role in forensic science. The history of imaging technology (photography) can be said to extend back to the times of Chinese philosopher Mo-Ti (470-390 B.C.) who described the principles behind the precursor to the camera obscura. Since that time, advances in imaging technology include the discovery of chemical photographic processes in the 19th century and the use of electronic imaging technology that includes analog video cameras and digital video and still cameras. By the mid 1990s, it was apparent that technologically advanced camera systems such as these were being adopted for use in the criminal justice system. This led the FBI to convene a meeting of individuals working in the field of forensic imaging from federal, state, local, and foreign law enforcement, and the U.S. military, during the summer of 1997. As a result of this meeting, the Technical Working Group on Imaging Technology was formed from a core group of the meeting’s participants. This group later became the Scientific Working Group on Imaging Technology (SWGIT). Prior to the inception of SWGIT, some law enforcement agencies began adopting digital imaging technology. Due to the lack of guidelines or standards, some of these agencies attempted to replace all their film cameras with substandard digital cameras, only to find that the equipment they had purchased was not capable of accomplishing the mission for which they were intended. At that time only low resolution digital cameras were deemed affordable by some law enforcement agencies. Some of these agencies were forced to rethink their photography procedures and reverted to the use of film cameras or replaced their low-resolution digital cameras with higher quality, more expensive equipment. Also lacking at this early stage was guidance on how to store and archive digital image files. When SWGIT was formed, it was tasked with providing guidance to law enforcement and others in the criminal justice system by releasing documents that describe the best practices and guidelines for the use of imaging technology, to include these concerns and many others. This group was terminated in 2015. == SWGIT Function == During its existence, SWGIT provided information on the appropriate use of various imaging technologies including both established and new. This was accomplished through the release of documents such as the SWGIT Best Practices documents. As changes in technology occurred, these documents were updated. Over the course of its existence, SWGIT collaborated with other Scientific Working Groups to address imaging concerns within their respective disciplines. SWGIT published over 20 documents that dealt specifically with imaging technology. SWGIT also co-published documents with the Scientific Working Group on Digital Evidence (SWGDE) that had a component or components dealing with imaging technology. SWGIT also provided imaging technology guidance and input for documents from the Scientific Working Group on Friction Ridge Analysis, Study and Technology (SWGFAST), the Scientific Working Group for Forensic Document Examination (SWGDOC), and the Scientific Working Group on Shoeprint and Tire Tread Evidence (SWGTREAD). SWGIT assisted the American Society of Crime Lab Directors/Laboratory Accreditation Board (ASCLD/LAB) in the writing of definitions and standards for the accreditation of Digital and Multimedia Evidence sections of crime laboratories. In addition to releasing documents, SWGIT members disseminated best practices for law enforcement professionals where imaging technology was concerned. This was carried out by attending and lecturing at meetings and conferences of various forensic organizations that included: The American Academy of Forensic Sciences (AAFS) The International Association for Identification (IAI) The Law Enforcement and Emergency Services Video Association (LEVA) The American Society of Crime Lab Directors (ASCLD) The SWGIT membership consisted of approximately fifty scientists, photographers, instructors, and managers from more than two dozen federal, state, and local law enforcement agencies, as well as from the academic and research communities. The membership elected its officers from within. SWGIT was composed of the Executive Committee, four standing subcommittees, and ad hoc subcommittees appointed on an as-needed basis. The standing subcommittees were: Image Analysis, Forensic Photography, Video, and Outreach. This group was terminated in 2015. == Legal Proceedings == The following court cases have conducted Daubert v. Merrell Dow Pharm., Inc., 509 U.S. 579 (1993) hearings in which SWGIT best practice documents have been cited as accepted protocol, methodology, and as generally accepted techniques in the forensic community: U. S. v. Rudy Frabizio, U.S. District Court, Boston, MA, 2008 (Image Authentication) U.S. v. Nobumochi Furukawa, U.S. District Court, Minnesota, 2007 (Video Authentication) U.S. v. John Stroman, U.S. District Court, South Carolina, 2007 (Facial Comparison Analysis) State of Texas v. Daniel Day, Tarrant County Texas, 2005 (Camera Identification to Images) U.S. v. Marc Watzman, U.S. District Court, Northern Illinois, 2004 (Video Authentication) U.S. v. McKreith, U.S. District Court, Fort Lauderdale, FL, 2002 (Photo comparison of shirt) == Termination == This group was unfunded by the FBI in 2015.

  Read more →
• 

  Deep Learning Studio

  Deep Learning Studio is a software tool that aims to simplify the creation of deep learning models used in artificial intelligence. It is compatible with a number of open-source programming frameworks popularly used in artificial neural networks, including MXNet and Google's TensorFlow. Prior to the release of Deep Learning Studio in January 2017, proficiency in Python, among other programming languages, was essential in developing effective deep learning models. Deep Learning Studio sought to simplify the model creation process through a visual, drag-and-drop interface and the application of pre-trained learning models on available data. Irving, Texas–based Deep Cognition Inc. is the developer behind Deep Learning Studio. In 2017, the software allowed Deep Cognition to become a finalist for Best Innovation in Deep Learning in the Alconics Awards, which are given annually to the best artificial intelligence software. Deep Cognition launched version 2.0 of Deep Learning Studio at NVIDIA's GTC 2018 Conference in San Jose, California. Fremont, California–based computing products supplier Exxact Corp provides desktop computers specifically built to handle Deep Learning Studio workloads. == Features == Source: Deep Learning Studio is available in two versions: Desktop and Cloud, both of which are free software. The Desktop version is available on Windows and Ubuntu. The Cloud version is available in single-user and multi-user configurations. A Deep Cognition account is needed to access the Cloud version. Account registration is free. Deep Learning Studio can import existing Keras models; it also takes a data set as an input. Deep Learning Studio's AutoML feature allows automatic generation of deep learning models. More advanced users may choose to generate their own models using various types of layers and neural networks. Deep Learning Studio also has a library of loss functions and optimizers for use in hyperparameter tuning, a traditionally complicated area in neural network programming. Generated models can be trained using either CPUs or GPUs. Trained models can then be used for predictive analytics.

  Read more →
• 

  Evaluation of machine translation

  Various methods for the evaluation for machine translation have been employed. This article focuses on the evaluation of the output of machine translation, rather than on performance or usability evaluation. == Round-trip translation == A typical way for lay people to assess machine translation quality is to translate from a source language to a target language and back to the source language with the same engine. Though intuitively this may seem like a good method of evaluation, it has been shown that round-trip translation is a "poor predictor of quality". The reason why it is such a poor predictor of quality is reasonably intuitive. A round-trip translation is not testing one system, but two systems: the language pair of the engine for translating into the target language, and the language pair translating back from the target language. Consider the following examples of round-trip translation performed from English to Italian and Portuguese from Somers (2005): In the first example, where the text is translated into Italian then back into English—the English text is significantly garbled, but the Italian is a serviceable translation. In the second example, the text translated back into English is perfect, but the Portuguese translation is meaningless; the program thought "tit" was a reference to a tit (bird), which was intended for a "tat", a word it did not understand. While round-trip translation may be useful to generate a "surplus of fun," the methodology is deficient for serious study of machine translation quality. == Human evaluation == This section covers two of the large scale evaluation studies that have had significant impact on the field—the ALPAC 1966 study and the ARPA study. === Automatic Language Processing Advisory Committee (ALPAC) === One of the constituent parts of the ALPAC report was a study comparing different levels of human translation with machine translation output, using human subjects as judges. The human judges were specially trained for the purpose. The evaluation study compared an MT system translating from Russian into English with human translators, on two variables. The variables studied were "intelligibility" and "fidelity". Intelligibility was a measure of how "understandable" the sentence was, and was measured on a scale of 1–9. Fidelity was a measure of how much information the translated sentence retained compared to the original, and was measured on a scale of 0–9. Each point on the scale was associated with a textual description. For example, 3 on the intelligibility scale was described as "Generally unintelligible; it tends to read like nonsense but, with a considerable amount of reflection and study, one can at least hypothesize the idea intended by the sentence". Intelligibility was measured without reference to the original, while fidelity was measured indirectly. The translated sentence was presented, and after reading it and absorbing the content, the original sentence was presented. The judges were asked to rate the original sentence on informativeness. So, the more informative the original sentence, the lower the quality of the translation. The study showed that the variables were highly correlated when the human judgment was averaged per sentence. The variation among raters was small, but the researchers recommended that at the very least, three or four raters should be used. The evaluation methodology managed to separate translations by humans from translations by machines with ease. The study concluded that, "highly reliable assessments can be made of the quality of human and machine translations". === Advanced Research Projects Agency (ARPA) === As part of the Human Language Technologies Program, the Advanced Research Projects Agency (ARPA) created a methodology to evaluate machine translation systems, and continues to perform evaluations based on this methodology. The evaluation programme was instigated in 1991, and continues to this day. Details of the programme can be found in White et al. (1994) and White (1995). The evaluation programme involved testing several systems based on different theoretical approaches; statistical, rule-based and human-assisted. A number of methods for the evaluation of the output from these systems were tested in 1992 and the most recent suitable methods were selected for inclusion in the programmes for subsequent years. The methods were; comprehension evaluation, quality panel evaluation, and evaluation based on adequacy and fluency. Comprehension evaluation aimed to directly compare systems based on the results from multiple choice comprehension tests, as in Church et al. (1993). The texts chosen were a set of articles in English on the subject of financial news. These articles were translated by professional translators into a series of language pairs, and then translated back into English using the machine translation systems. It was decided that this was not adequate for a standalone method of comparing systems and as such abandoned due to issues with the modification of meaning in the process of translating from English. The idea of quality panel evaluation was to submit translations to a panel of expert native English speakers who were professional translators and get them to evaluate them. The evaluations were done on the basis of a metric, modelled on a standard US government metric used to rate human translations. This was good from the point of view that the metric was "externally motivated", since it was not specifically developed for machine translation. However, the quality panel evaluation was very difficult to set up logistically, as it necessitated having a number of experts together in one place for a week or more, and furthermore for them to reach consensus. This method was also abandoned. Along with a modified form of the comprehension evaluation (re-styled as informativeness evaluation), the most popular method was to obtain ratings from monolingual judges for segments of a document. The judges were presented with a segment, and asked to rate it for two variables, adequacy and fluency. Adequacy is a rating of how much information is transferred between the original and the translation, and fluency is a rating of how good the English is. This technique was found to cover the relevant parts of the quality panel evaluation, while at the same time being easier to deploy, as it didn't require expert judgment. Measuring systems based on adequacy and fluency, along with informativeness is now the standard methodology for the ARPA evaluation program. == Automatic evaluation == In the context of this article, a metric is a measurement. A metric that evaluates machine translation output represents the quality of the output. The quality of a translation is inherently subjective, there is no objective or quantifiable "good." Therefore, any metric must assign quality scores so they correlate with the human judgment of quality. That is, a metric should score highly translations that humans score highly, and give low scores to those humans give low scores. Human judgment is the benchmark for assessing automatic metrics, as humans are the end-users of any translation output. The measure of evaluation for metrics is correlation with human judgment. This is generally done at two levels, at the sentence level, where scores are calculated by the metric for a set of translated sentences, and then correlated against human judgment for the same sentences. And at the corpus level, where scores over the sentences are aggregated for both human judgments and metric judgments, and these aggregate scores are then correlated. Figures for correlation at the sentence level are rarely reported, although Banerjee et al. (2005) do give correlation figures that show that, at least for their metric, sentence-level correlation is substantially worse than corpus level correlation. While not widely reported, it has been noted that the genre, or domain, of a text has an effect on the correlation obtained when using metrics. Coughlin (2003) reports that comparing the candidate text against a single reference translation does not adversely affect the correlation of metrics when working in a restricted domain text. Even if a metric correlates well with human judgment in one study on one corpus, this successful correlation may not carry over to another corpus. Good metric performance, across text types or domains, is important for the reusability of the metric. A metric that only works for text in a specific domain is useful, but less useful than one that works across many domains—because creating a new metric for every new evaluation or domain is undesirable. Another important factor in the usefulness of an evaluation metric is to have a good correlation, even when working with small amounts of data, that is candidate sentences and reference translations. Turian et al. (2003) point out that, "Any MT evaluation measure is less reliable on shorter translations", and

  Read more →
• 

  The Best Free AI Background Remover for Beginners

  In search of the best AI background remover? An AI background remover is software that uses machine learning to help you get more done — it turns a rough idea into a polished result in seconds. When choosing one, weigh output quality, pricing, export formats, and how well it fits the tools you already use. Whether you are a beginner or a pro, the right AI background remover slots into your workflow and pays for itself fast. We tested the leading options and ranked them by quality, value, and ease of use.

  Read more →
• 

  Dr. Sbaitso

  Dr. Sbaitso ( SPAYT-soh) is an artificial intelligence speech synthesis program released late in 1991 by Creative Labs in Singapore for MS-DOS-based personal computers. The name is an acronym for "SoundBlaster Acting Intelligent Text-to-Speech Operator." == History == Dr. Sbaitso was distributed with various sound cards manufactured by Creative Technology in the early 1990s. The text-to-speech engine used is a version of Monologue, which was developed by First Byte Software. Monologue is a later release of First Byte's "SmoothTalker" software from 1984. The program "conversed" with the user as if it were a psychologist, though most of its responses were along the lines of "WHY DO YOU FEEL THAT WAY?" rather than any sort of complicated interaction. When confronted with a phrase it could not understand, it would often reply with something such as "THAT'S NOT MY PROBLEM." Dr. Sbaitso repeated text out loud that was typed after the word "SAY." Repeated swearing or abusive behavior on the part of the user caused Dr. Sbaitso to "break down" in a "PARITY ERROR" before resetting itself. The same would happen, if the user types "SAY PARITY." The program introduced itself with the following lines: HELLO [UserName], MY NAME IS DOCTOR SBAITSO. I AM HERE TO HELP YOU. SAY WHATEVER IS IN YOUR MIND FREELY, OUR CONVERSATION WILL BE KEPT IN STRICT CONFIDENCE. MEMORY CONTENTS WILL BE WIPED OFF AFTER YOU LEAVE, SO, TELL ME ABOUT YOUR PROBLEMS. The program was designed to showcase the digitized voices the cards were able to produce, though the quality was far from lifelike. Additionally, there was a version of this program for Microsoft Windows through the use of a program called Prody Parrot; this version of the software featured a more detailed graphical user interface. The text-to-speech was also used as the voice of 1st Prize from the Baldi's Basics series, albeit slowed down. == Commands == If the user submits "HELP", a list of commands will appear. If the user then submits "M", more commands will appear. There are three pages of commands in total, with guidance on how to use each of the features.

  Read more →
• 

  Chris Callison-Burch

  Chris Callison-Burch is an American computer scientist and professor of computer and information science at the University of Pennsylvania (Penn), specializing in natural language processing (NLP), artificial intelligence (AI), and crowdsourcing. He is recognised for his contributions to machine translation, paraphrase generation, and the application of large language models (LLMs) to AI challenges, with over 200 publications cited more than 33,000 times. Callison-Burch has influenced public policy on AI and copyright, testifying before the U.S. Congress in 2023 on generative AI’s implications. He serves as the faculty director for Penn’s Online Master of Science in Engineering in AI program. == Education == Callison-Burch earned his PhD in Computer Science from the University of Edinburgh in 2008, focusing on machine translation and paraphrasing techniques. His doctoral research developed statistical methods for generating paraphrases in machine translation systems, laying the foundation for his later NLP work. Prior to his PhD, he studied at Stanford University, where he developed an interest in computational linguistics. == Career == After his PhD, Callison-Burch joined the Centre for Language and Speech Processing at Johns Hopkins University as a research faculty member from 2008 to 2013, working on NLP projects, including machine translation and crowdsourcing for creating training data. In 2013, he joined the University of Pennsylvania as an assistant professor in the Department of Computer and Information Science and was promoted to associate professor in 2017, and to full professor in 2024. At Penn, Callison-Burch teaches courses on AI and NLP, including CIS 5300 (Natural Language Processing) and CIS 5210 (Artificial Intelligence), which attract over 500 students annually. He directs Penn’s Online Master of Science in Engineering in AI program, launched in 2025. He teaches AI and NLP courses on Coursera, reaching thousands of global learners. Callison-Burch was a part-time visiting researcher at Google in 2019 and 2020, where he collaborated on applying Google's LLM to Dungeons & Dragons dialogues. In 2023, he took a sabbatical at the Allen Institute for AI (AI2), where he contributed to vision-language models. == Research == Callison-Burch’s research focuses on NLP, AI, and crowdsourcing, with significant contributions to machine translation, paraphrase generation, and LLMs for tasks like text simplification and bias detection. His early work developed crowdsourcing methods for machine translation, leveraging non-expert annotators for paraphrase-based evaluation, influencing platforms like Amazon Mechanical Turk. Recent projects have included several notable works. Molmo and PixMo (2025) are open-weight vision-language models developed with AI2, achieving state-of-the-art multimodal performance and earning a Best Paper Honourable Mention at CVPR 2025. Also in 2025, his work on Calibrating Large Language Models with Sample Consistency improves LLM reliability via sample-based calibration, presented at NAACL 2025. The Media Bias Detector (2025) is a real-time tool analysing selection and framing bias in news, using LLMs to detect persuasive language differences (e.g., Russian vs. English Wikipedia). Holodeck (2024) is a language-guided system for generating 3D embodied AI environments, presented at CVPR 2024. BORDIRLINES (2024) is a dataset for cross-lingual retrieval-augmented generation, focusing on culturally sensitive tasks. He has co-authored over 200 publications, featured at conferences like ACL, EMNLP, and CVPR. == Awards and recognition == Callison-Burch has received numerous awards: Best Paper Honourable Mention at CVPR 2025 for "Molmo and PixMo". Best Paper Award at the Workshop on Cognitive Modelling and Computational Linguistics (CMCL) 2024 for "Evaluating Vision-Language Models on Bistable Images". Best Paper Award at STARSEM 2016 for "So-Called Non-Subsective Adjectives". Best Paper Award at the Workshop on Sense, Concept and Entity Representations 2017 for "Word Sense Filtering Improves Embedding-Based Lexical Substitution". Honourable Mention Award at CHI 2018 for "A Data-Driven Analysis of Workers’ Earnings on Amazon Mechanical Turk". Google Faculty Research Award (2013) for crowdsourcing in NLP. Sloan Research Fellowship (2014). He has received research funding from Google, Microsoft, Amazon, Facebook, Roblox, DARPA, IARPA, and NSF. His h-index is 72, with over 33,000 citations. He served as General Chair of ACL 2017 and as the Program Co-Chair EMNLP 2015. == Public policy and testimony == On May 17, 2023, Callison-Burch testified before the U.S. House Subcommittee on Courts, Intellectual Property, and the Internet on AI and copyright law. His testimony emphasised generative AI’s role in creative industries and the need for balanced copyright frameworks. He has appeared on Fox News to discuss AI’s societal impact, and discussed its impact with other print news sources. He contributes to AI ethics discussions, including workshops on AI’s effects on writing and creative professions.

  Read more →
• 

  Tagged Deterministic Finite Automaton

  In the automata theory, a tagged deterministic finite automaton (TDFA) is an extension of deterministic finite automaton (DFA). In addition to solving the recognition problem for regular languages, TDFA is also capable of submatch extraction and parsing. While canonical DFA can find out if a string belongs to the language defined by a regular expression, TDFA can also extract substrings that match specific subexpressions. More generally, TDFA can identify positions in the input string that match tagged positions in a regular expression (tags are meta-symbols similar to capturing parentheses, but without the pairing requirement). == History == TDFA were first described by Ville Laurikari in 2000. Prior to that it was unknown whether it is possible to perform submatch extraction in one pass on a deterministic finite-state automaton, so this paper was an important advancement. Laurikari described TDFA construction and gave a proof that the determinization process terminates, however the algorithm did not handle disambiguation correctly. In 2007 Chris Kuklewicz implemented TDFA in a Haskell library Regex-TDFA with POSIX longest-match semantics. Kuklewicz gave an informal description of the algorithm and answered the principal question whether TDFA are capable of POSIX longest-match disambiguation, which was doubted by other researchers. In 2017 Ulya Trafimovich described TDFA with one-symbol lookahead. The use of a lookahead symbol reduces the number of registers and register operations in a TDFA, which makes it faster and often smaller than Laurikari TDFA. Trafimovich called TDFA variants with and without lookahead TDFA(1) and TDFA(0) by analogy with LR parsers LR(1) and LR(0). The algorithm was implemented in the open-source lexer generator RE2C. Trafimovich formalized Kuklewicz disambiguation algorithm. In 2018 Angelo Borsotti worked on an experimental Java implementation of TDFA; it was published later in 2021. In 2019 Borsotti and Trafimovich adapted POSIX disambiguation algorithm by Okui and Suzuki to TDFA. They gave a formal proof of correctness of the new algorithm and showed that it is faster than Kuklewicz algorithm in practice. In 2020 Trafimovich published an article about TDFA implementation in RE2C. In 2022 Borsotti and Trafimovich published a paper with a detailed description of TDFA construction. The paper incorporated their past research and presented multi-pass TDFA that are better suited to just-in-time determinization. They also compared TDFA against other algorithms and provided benchmarks. == Formal definition == TDFA have the same basic structure as ordinary DFA: a finite set of states linked by transitions. In addition to that, TDFA have a fixed set of registers that hold tag values, and register operations on transitions that set or copy register values. The values may be scalar offsets, or offset lists for tags that match repeatedly (the latter can be represented efficiently using a trie structure). There is no one-to-one mapping between tags in a regular expression and registers in a TDFA: a single tag may need many registers, and the same register may hold values of different tags. The following definition is according to Trafimovich and Borsotti. The original definition by Laurikari is slightly different. A tagged deterministic finite automaton F {\displaystyle F} is a tuple ( Σ , T , S , S f , s 0 , R , R f , δ , φ ) {\displaystyle (\Sigma ,T,S,S_{f},s_{0},R,R_{f},\delta ,\varphi )} , where: Σ {\displaystyle \Sigma } is a finite set of symbols (alphabet) T {\displaystyle T} is a finite set of tags S {\displaystyle S} is a finite set of states with initial state s 0 {\displaystyle s_{0}} and a subset of final states S f ⊆ S {\displaystyle S_{f}\subseteq S} R {\displaystyle R} is a finite set of registers with a subset of final registers R f {\displaystyle R_{f}} (one per tag) δ : S × Σ → S × O ∗ {\displaystyle \delta :S\times \Sigma \rightarrow S\times O^{}} is a transition function φ : S f → O ∗ {\displaystyle \varphi :S_{f}\rightarrow O^{}} is a final function, where O {\displaystyle O} is a set of register operations of the following types: set register i {\displaystyle i} to nil or to the current position: i ← v {\displaystyle i\leftarrow v} , where v ∈ { n , p } {\displaystyle v\in \{\mathbf {n} ,\mathbf {p} \}} copy register j {\displaystyle j} to register i {\displaystyle i} : i ← j {\displaystyle i\leftarrow j} copy register j {\displaystyle j} to register i {\displaystyle i} and append history: i ← j ⋅ h {\displaystyle i\leftarrow j\cdot h} , where h {\displaystyle h} is a string over { n , p } {\displaystyle \{\mathbf {n} ,\mathbf {p} \}} === Example === Figure 0 shows an example TDFA for regular expression ( 1 a 2 ) ∗ 3 ( a | 4 b ) 5 b ∗ {\displaystyle (1a2)^{}3(a|4b)5b^{}} with alphabet Σ = { a , b } {\displaystyle \Sigma =\{a,b\}} and a set of tags T = { 1 , 2 , 3 , 4 , 5 } {\displaystyle T=\{1,2,3,4,5\}} that matches strings of the form a … a b … b {\displaystyle a\dots ab\dots b} with at least one symbol. TDFA has four states S = { 0 , 1 , 2 , 3 } {\displaystyle S=\{0,1,2,3\}} three of which are final S f = { 1 , 2 , 3 } {\displaystyle S_{f}=\{1,2,3\}} . The set of registers is R = { r 1 , r 2 , r 3 , r 4 , r 5 } {\displaystyle R=\{r_{1},r_{2},r_{3},r_{4},r_{5}\}} with a subset of final registers R f = { r 1 , r 2 , r 3 , r 4 , r 5 } {\displaystyle R_{f}=\{r_{1},r_{2},r_{3},r_{4},r_{5}\}} where register r i {\displaystyle r_{i}} corresponds to i {\displaystyle i} -th tag. Transitions have operations defined by the δ {\displaystyle \delta } function, and final states have operations defined by the φ {\displaystyle \varphi } function (marked with wide-tipped arrow). For example, to match string a a b {\displaystyle aab} , one starts in state 0, matches the first a {\displaystyle a} and moves to state 1 (setting registers r 1 , r 2 {\displaystyle r_{1},r_{2}} to undefined and r 3 {\displaystyle r_{3}} to the current position 0), matches the second a {\displaystyle a} and loops to state 1 (register values are now r 1 = 0 , r 2 = r 3 = 1 {\displaystyle r_{1}=0,r_{2}=r_{3}=1} ), matches b {\displaystyle b} and moves to state 2 (register values are now r 1 = 1 , r 2 = r 3 = r 4 = 2 {\displaystyle r_{1}=1,r_{2}=r_{3}=r_{4}=2} ), executes the final operations in state 2 (register values are now r 1 = 1 , r 2 = r 3 = r 4 = 2 , r 5 = 3 {\displaystyle r_{1}=1,r_{2}=r_{3}=r_{4}=2,r_{5}=3} ) and finally exits TDFA. == Complexity == Canonical DFA solve the recognition problem in linear time. The same holds for TDFA, since the number of registers and register operations is fixed and depends only on the regular expression, but not on the length of input. The overhead on submatch extraction depends on tag density in a regular expression and nondeterminism degree of each tag (the maximum number of registers needed to track all possible values of the tag in a single TDFA state). On one extreme, if there are no tags, a TDFA is identical to a canonical DFA. On the other extreme, if every subexpression is tagged, a TDFA effectively performs full parsing and has many operations on every transition. In practice for real-world regular expressions with a few submatch groups the overhead is negligible compared to matching with canonical DFA. == TDFA construction == TDFA construction is performed in a few steps. First, a regular expression is converted to a tagged nondeterministic finite automaton (TNFA). Second, a TNFA is converted to a TDFA using a determinization procedure; this step also includes disambiguation that resolves conflicts between ambiguous TNFA paths. After that, a TDFA can optionally go through a number of optimizations that reduce the number of registers and operations, including minimization that reduces the number of states. Algorithms for all steps of TDFA construction with pseudocode are given in the paper by Borsotti and Trafimovich. This section explains TDFA construction on the example of a regular expression a ∗ t b ∗ | a b {\displaystyle a^{}tb^{}|ab} , where t {\displaystyle t} is a tag and { a , b } {\displaystyle \{a,b\}} are alphabet symbols. === Tagged NFA === TNFA is a nondeterministic finite automaton with tagged ε-transitions. It was first described by Laurikari, although similar constructions were known much earlier as Mealy machines and nondeterministic finite-state transducers. TNFA construction is very similar to Thompson's construction: it mirrors the structure of a regular expression. Importantly, TNFA preserves ambiguity in a regular expression: if it is possible to match a string in two different ways, then TNFA for this regular expression has two different accepting paths for this string. TNFA definition by Borsotti and Trafimovich differs from the original one by Laurikari in that TNFA can have negative tags on transitions: they are needed to make the absence of match explicit in cases when there is a bypass for a tagged transition. Figure 1 shows TNFA for the example regu

  Read more →