Linguatec

The Linguatec Sprachtechnologien GmbH is a language technology provider, specialized in the field of machine translation, speech synthesis and speech recognition. Linguatec was founded in Munich in 1996 and its headquarters are in Pasing. Linguatec has won the European Information Society Technologies Prize three times. On their website, they are now using the online service Voice Reader Web, so that the information can be read out in every language by means of a text-to-speech function. == Core areas == Machine translation The different versions of Personal Translator (seven language pairs) can be used "for home use" or for professional business use in the company network. In addition to this, specialist dictionaries are offered to broaden standard vocabulary. Speech synthesis The Voice Reader text-to-speech program reads in twelve languages: German, British English, American English, French, Quebec French, Spanish, Mexican Spanish, Italian, Dutch, Portuguese, Czech, Chinese. Speech recognition Voice Pro is based on ViaVoice technology from IBM. There are special software programs for doctors and lawyers. == Patents == 2005 pending patent application for a newly developed hybrid technology that uses the intelligence of neural networks for machine translation. == Awards == 2004 European IT Prize for Beyond Babel 2004 test winner Stiftung Warentest – best voice recognition 1998 European IT Prize – applied voice recognition 1996 European IT Prize – automated translation == Studies == 2005 University of Regensburg: Voice Reader user test 2002 Fraunhofer Institute for Industrial Engineering and Organization IAO: user study on the efficiency of machine translation

Transfer learning

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

Link encryption

Link encryption is an approach to communications security that encrypts and decrypts all network traffic at each network routing point (e.g. network switch, or node through which it passes) until arrival at its final destination. This repeated decryption and encryption is necessary to allow the routing information contained in each transmission to be read and employed further to direct the transmission toward its destination, before which it is re-encrypted. This contrasts with end-to-end encryption where internal information, but not the header/routing information, is encrypted by the sender at the point of origin and only decrypted by the intended recipient. Link encryption offers two main advantages: encryption is automatic so there is less opportunity for human error. if the communications link operates continuously and carries an unvarying level of traffic, link encryption defeats traffic analysis. On the other hand, end-to-end encryption ensures only the intended recipient has access to the plaintext. Link encryption can be used with end-to-end systems by superencrypting the messages. Bulk encryption refers to encrypting a large number of circuits at once, after they have been multiplexed.

Locally recoverable code

Locally recoverable codes are a family of error correction codes that were introduced first by D. S. Papailiopoulos and A. G. Dimakis and have been widely studied in information theory due to their applications related to distributive and cloud storage systems. An [ n , k , d , r ] q {\displaystyle [n,k,d,r]_{q}} LRC is an [ n , k , d ] q {\displaystyle [n,k,d]_{q}} linear code such that there is a function f i {\displaystyle f_{i}} that takes as input i {\displaystyle i} and a set of r {\displaystyle r} other coordinates of a codeword c = ( c 1 , … , c n ) ∈ C {\displaystyle c=(c_{1},\ldots ,c_{n})\in C} different from c i {\displaystyle c_{i}} , and outputs c i {\displaystyle c_{i}} . == Overview == Erasure-correcting codes, or simply erasure codes, for distributed and cloud storage systems, are becoming more and more popular as a result of the present spike in demand for cloud computing and storage services. This has inspired researchers in the fields of information and coding theory to investigate new facets of codes that are specifically suited for use with storage systems. It is well-known that LRC is a code that needs only a limited set of other symbols to be accessed in order to restore every symbol in a codeword. This idea is very important for distributed and cloud storage systems since the most common error case is when one storage node fails (erasure). The main objective is to recover as much data as possible from the fewest additional storage nodes in order to restore the node. Hence, Locally Recoverable Codes are crucial for such systems. The following definition of the LRC follows from the description above: an [ n , k , r ] {\displaystyle [n,k,r]} -Locally Recoverable Code (LRC) of length n {\displaystyle n} is a code that produces an n {\displaystyle n} -symbol codeword from k {\displaystyle k} information symbols, and for any symbol of the codeword, there exist at most r {\displaystyle r} other symbols such that the value of the symbol can be recovered from them. The locality parameter satisfies 1 ≤ r ≤ k {\displaystyle 1\leq r\leq k} because the entire codeword can be found by accessing k {\displaystyle k} symbols other than the erased symbol. Furthermore, Locally Recoverable Codes, having the minimum distance d {\displaystyle d} , can recover d − 1 {\displaystyle d-1} erasures. == Definition == Let C {\displaystyle C} be a [ n , k , d ] q {\displaystyle [n,k,d]_{q}} linear code. For i ∈ { 1 , … , n } {\displaystyle i\in \{1,\ldots ,n\}} , let us denote by r i {\displaystyle r_{i}} the minimum number of other coordinates we have to look at to recover an erasure in coordinate i {\displaystyle i} . The number r i {\displaystyle r_{i}} is said to be the locality of the i {\displaystyle i} -th coordinate of the code. The locality of the code is defined as An [ n , k , d , r ] q {\displaystyle [n,k,d,r]_{q}} locally recoverable code (LRC) is an [ n , k , d ] q {\displaystyle [n,k,d]_{q}} linear code C ∈ F q n {\displaystyle C\in \mathbb {F} _{q}^{n}} with locality r {\displaystyle r} . Let C {\displaystyle C} be an [ n , k , d ] q {\displaystyle [n,k,d]_{q}} -locally recoverable code. Then an erased component can be recovered linearly, i.e. for every i ∈ { 1 , … , n } {\displaystyle i\in \{1,\ldots ,n\}} , the space of linear equations of the code contains elements of the form x i = f ( x i 1 , … , x i r ) {\displaystyle x_{i}=f(x_{i_{1}},\ldots ,x_{i_{r}})} , where i j ≠ i {\displaystyle i_{j}\neq i} . == Optimal locally recoverable codes == Theorem Let n = ( r + 1 ) s {\displaystyle n=(r+1)s} and let C {\displaystyle C} be an [ n , k , d ] q {\displaystyle [n,k,d]_{q}} -locally recoverable code having s {\displaystyle s} disjoint locality sets of size r + 1 {\displaystyle r+1} . Then An [ n , k , d , r ] q {\displaystyle [n,k,d,r]_{q}} -LRC C {\displaystyle C} is said to be optimal if the minimum distance of C {\displaystyle C} satisfies == Tamo–Barg codes == Let f ∈ F q [ x ] {\displaystyle f\in \mathbb {F} _{q}[x]} be a polynomial and let ℓ {\displaystyle \ell } be a positive integer. Then f {\displaystyle f} is said to be ( r {\displaystyle r} , ℓ {\displaystyle \ell } )-good if • f {\displaystyle f} has degree r + 1 {\displaystyle r+1} , • there exist distinct subsets A 1 , … , A ℓ {\displaystyle A_{1},\ldots ,A_{\ell }} of F q {\displaystyle \mathbb {F} _{q}} such that – for any i ∈ { 1 , … , ℓ } {\displaystyle i\in \{1,\ldots ,\ell \}} , f ( A i ) = { t i } {\displaystyle f(A_{i})=\{t_{i}\}} for some t i ∈ F q {\displaystyle t_{i}\in \mathbb {F} _{q}} , i.e., f {\displaystyle f} is constant on A i {\displaystyle A_{i}} , – # A i = r + 1 {\displaystyle \#A_{i}=r+1} , – A i ∩ A j = ∅ {\displaystyle A_{i}\cap A_{j}=\varnothing } for any i ≠ j {\displaystyle i\neq j} . We say that { A 1 , … , A ℓ {\displaystyle A_{1},\ldots ,A_{\ell }} } is a splitting covering for f {\displaystyle f} . === Tamo–Barg construction === The Tamo–Barg construction utilizes good polynomials. • Suppose that a ( r , ℓ ) {\displaystyle (r,\ell )} -good polynomial f ( x ) {\displaystyle f(x)} over F q {\displaystyle \mathbb {F} _{q}} is given with splitting covering i ∈ { 1 , … , ℓ } {\displaystyle i\in \{1,\ldots ,\ell \}} . • Let s ≤ ℓ − 1 {\displaystyle s\leq \ell -1} be a positive integer. • Consider the following F q {\displaystyle \mathbb {F} _{q}} -vector space of polynomials V = { ∑ i = 0 s g i ( x ) f ( x ) i : deg ⁡ ( g i ( x ) ) ≤ deg ⁡ ( f ( x ) ) − 2 } . {\displaystyle V=\left\{\sum _{i=0}^{s}g_{i}(x)f(x)^{i}:\deg(g_{i}(x))\leq \deg(f(x))-2\right\}.} • Let T = ⋃ i = 1 ℓ A i {\textstyle T=\bigcup _{i=1}^{\ell }A_{i}} . • The code { ev T ⁡ ( g ) : g ∈ V } {\displaystyle \{\operatorname {ev} _{T}(g):g\in V\}} is an ( ( r + 1 ) ℓ , ( s + 1 ) r , d , r ) {\displaystyle ((r+1)\ell ,(s+1)r,d,r)} -optimal locally coverable code, where ev T {\displaystyle \operatorname {ev} _{T}} denotes evaluation of g {\displaystyle g} at all points in the set T {\displaystyle T} . === Parameters of Tamo–Barg codes === • Length. The length is the number of evaluation points. Because the sets A i {\displaystyle A_{i}} are disjoint for i ∈ { 1 , … , ℓ } {\displaystyle i\in \{1,\ldots ,\ell \}} , the length of the code is | T | = ( r + 1 ) ℓ {\displaystyle |T|=(r+1)\ell } . • Dimension. The dimension of the code is ( s + 1 ) r {\displaystyle (s+1)r} , for s {\displaystyle s} ≤ ℓ − 1 {\displaystyle \ell -1} , as each g i {\displaystyle g_{i}} has degree at most deg ⁡ ( f ( x ) ) − 2 {\displaystyle \deg(f(x))-2} , covering a vector space of dimension deg ⁡ ( f ( x ) ) − 1 = r {\displaystyle \deg(f(x))-1=r} , and by the construction of V {\displaystyle V} , there are s + 1 {\displaystyle s+1} distinct g i {\displaystyle g_{i}} . • Distance. The distance is given by the fact that V ⊆ F q [ x ] ≤ k {\displaystyle V\subseteq \mathbb {F} _{q}[x]_{\leq k}} , where k = r + 1 − 2 + s ( r + 1 ) {\displaystyle k=r+1-2+s(r+1)} , and the obtained code is the Reed-Solomon code of degree at most k {\displaystyle k} , so the minimum distance equals ( r + 1 ) ℓ − ( ( r + 1 ) − 2 + s ( r + 1 ) ) {\displaystyle (r+1)\ell -((r+1)-2+s(r+1))} . • Locality. After the erasure of the single component, the evaluation at a i ∈ A i {\displaystyle a_{i}\in A_{i}} , where | A i | = r + 1 {\displaystyle |A_{i}|=r+1} , is unknown, but the evaluations for all other a ∈ A i {\displaystyle a\in A_{i}} are known, so at most r {\displaystyle r} evaluations are needed to uniquely determine the erased component, which gives us the locality of r {\displaystyle r} . To see this, g {\displaystyle g} restricted to A j {\displaystyle A_{j}} can be described by a polynomial h {\displaystyle h} of degree at most deg ⁡ ( f ( x ) ) − 2 = r + 1 − 2 = r − 1 {\displaystyle \deg(f(x))-2=r+1-2=r-1} thanks to the form of the elements in V {\displaystyle V} (i.e., thanks to the fact that f {\displaystyle f} is constant on A j {\displaystyle A_{j}} , and the g i {\displaystyle g_{i}} 's have degree at most deg ⁡ ( f ( x ) ) − 2 {\displaystyle \deg(f(x))-2} ). On the other hand | A j ∖ { a j } | = r {\displaystyle |A_{j}\backslash \{a_{j}\}|=r} , and r {\displaystyle r} evaluations uniquely determine a polynomial of degree r − 1 {\displaystyle r-1} . Therefore h {\displaystyle h} can be constructed and evaluated at a j {\displaystyle a_{j}} to recover g ( a j ) {\displaystyle g(a_{j})} . === Example of Tamo–Barg construction === We will use x 5 ∈ F 41 [ x ] {\displaystyle x^{5}\in \mathbb {F} _{41}[x]} to construct [ 15 , 8 , 6 , 4 ] {\displaystyle [15,8,6,4]} -LRC. Notice that the degree of this polynomial is 5, and it is constant on A i {\displaystyle A_{i}} for i ∈ { 1 , … , 8 } {\displaystyle i\in \{1,\ldots ,8\}} , where A 1 = { 1 , 10 , 16 , 18 , 37 } {\displaystyle A_{1}=\{1,10,16,18,37\}} , A 2 = 2 A 1 {\displaystyle A_{2}=2A_{1}} , A 3 = 3 A 1 {\displaystyle A_{3}=3A_{1}} , A 4 = 4 A 1 {\displaystyle A_{4}=4A_{1}} , A 5 = 5 A 1 {\displaystyle A_{5}=5A_{1}} , A 6 = 6 A 1 {\displaystyle A_{6}=6A_{1}}

Shorty Awards

The Shorty Awards (also known as "The Shortys") are awards for outstanding and innovative work in digital and social media content by brands, advertising agencies, and creators. The awards, which generally focus on short-term content, honor achievements in content creation on Twitter, Facebook, YouTube, Instagram, TikTok, Twitch, and other social networking sites. The Shorty Awards began in 2008 and initially recognized achievements by independent creators on Twitter, with the first formal awards ceremony occurring in February 2009. Since then, the awards, which are now awarded each spring, have shifted their focus to recognize content across numerous platforms. Entrant work is judged on the merits of excellence in creativity, strategy, and engagement by the Real Time Academy, a group of industry professionals selected by the Shorty Awards on the basis of their professional reputations, industry knowledge, and personal achievements (which may include previous Shorty wins). An additional public voting component, known as Audience Honor Voting, is also used to select Shorty Awards contenders. Notable Shorty Award winners include Malala Yousafzai, Trevor Noah, Michelle Obama, Conan O’Brien, Lady Gaga, Bill Nye, Jacob Reed, and Lizzo. Brands and organizations such as Chipotle, Duolingo, Marvel Studios, HBO, Red Bull, Airbnb, Nestle, BMW, UNICEF and the Human Rights Campaign have also been awarded. The Shorty Awards also produces an annual award program called The Shorty Impact Awards, a competition dedicated to showcasing digital and social media-based projects by brands, agencies, and organizations that seek to make the world a better place. == List of ceremonies == == 1st Shorty Awards == The awards were created in 2008 by tech entrepreneurs Greg Galant, Adam Varga, and Lee Semel of Sawhorse Media. They invited Twitter account holders to nominate the best Twitter users in general categories such as humor, news, food, and design. Winners were chosen by more than 30,000 Twitter users during the voting period. The founders of Twitter first heard about the awards after the contest had gotten underway and expressed support for it. The first Shorty Awards ceremony was held on February 11, 2009, at the Galapagos Art Space in Brooklyn, New York. Approximately 300 people attended the event. The event was hosted by CNN anchor Rick Sanchez and featured appearances by prominent Twitter users MC Hammer and Gary Vaynerchuk and a video appearance by Shaquille O'Neal. The awards, in 26 categories, were voted on by Twitter users. == 2nd Shorty Awards == Voting for the second Shorty Awards opened in January 2010 in 26 official categories. A Real-Time Photo of the Year category was added to the list of official categories for the first time, recognizing the best photo posted to services such as Twitpic, Yfrog, or Facebook. The second Shorty Awards competition introduced a panel of judges called the Real-Time Academy of Short Form Arts & Sciences whose members were Craig Newmark, David Pogue, Kurt Andersen, Caterina Fake, Joi Ito, Frank Moss, Alberto Ibargüen, Sreenath Sreenivasan, MC Hammer, Alyssa Milano and Jimmy Wales. After public nominations determined the finalists, the academy decided on the winners. Winners were announced at a ceremony held in the Times Center in The New York Times building in Manhattan that was also streamed online. The ceremony was hosted by CNN anchor Rick Sanchez, who presented awards in the official categories as well as the newly added Real-Time Photo of the Year and a special humanitarian award. == 3rd Shorty Awards == The nomination period for the third annual Shorty Awards opened in January 2011 and ran through February 11, 2011, except for new categories that had extended nomination deadlines. There were 30 official categories and five special categories. In addition to Real-Time Photo of the Year, for the first time the awards accepted nominations for Foursquare Mayor of the Year, Foursquare Location of the Year, Microblog of the Year on Tumblr, and a Connecting People award. The awards also introduced new Shorty Industry Awards to recognize the best uses of social media by brands and agencies. Winners were announced at a ceremony on March 28, 2011, hosted by Aasif Mandvi in the Times Center. Other Shorty Awards presenters were scheduled to include Kiefer Sutherland, Jerry Stiller, Anne Meara, Stephen Wallem, Miss USA Rima Fakih, and Miss Teen USA Kamie Crawford. == 4th Shorty Awards == The 4th Annual Shorty Awards featured Ricky Gervais and Tiffani Thiessen. 1.6 million tweeted nominations were made across all the categories to honor the top users on Twitter, Facebook, Tumblr, Foursquare, YouTube and other internet platforms. == 5th Shorty Awards == The 5th Annual Shorty Awards ceremony featured Felicia Day, James Urbaniak, Kristian Nairn, Hannibal Buress, Carrie Keagan, Chris Hardwick, David Karp and Coco Rocha. 2.4 million tweeted nominations were made across all the categories to honor the top users on Twitter, Facebook, Tumblr, Foursquare, YouTube and other internet sites. == 6th Shorty Awards == The ceremony took place on April 7, 2014, at the New York TimesCenter and was hosted by Comedian Natasha Leggero. The show included appearances by Patton Oswalt, Jamie Oliver, Kristen Bell, Jerry Seinfeld, Moshe Kasher, Julie Klausner, Erin Brady, Guy Kawasaki, Matt Walsh, Retta, Us the Duo, Big Boi, Gilbert Gottfried, Thomas Middleditch, Billie Jean King and Leandra Medine. Winners included Jerry Seinfeld and Will Ferrell. == 7th Shorty Awards == The Seventh Annual Shorty Awards was hosted by comedian Rachel Dratch and took place on April 20, 2015, at The Times Center in NYC. The Real-Time Academy, the judging body of the Shortys, tripled in size for the 7th annual Awards and included Alton Brown, Mamrie Hart, Nikki Glaser, OK Go, The Fine Bros, Debbie Sterling, Dan Savage, Deena Varshavskaya and Palmer Luckey. Panic! at the Disco was the musical guest at the ceremony. On-stage presenters included Kevin Jonas, Bill Nye, Bella Thorne, Wyclef Jean, Emily Kinney and Tyler Oakley. == 8th Shorty Awards == The Eighth Annual Shorty Awards were held in NYC at the TimesCenter on April 11, 2016. They were hosted by YouTuber, Writer and Comedian Mamrie Hart with musical performances from Nico & Vinz. Winners of the night included Bill Wurtz, DJ Khaled, Misty Copeland, Casey Neistat, Dwayne Johnson, Hannah Hart, Troye Sivan, Baddie Winkle, Kevin Hart, Taraji P. Henson, King Bach, and Zach King. == 9th Shorty Awards == The Ninth Annual Shorty Awards were held in NYC at the PlayStation Theater on April 23, 2017. They were hosted by two-time Emmy Award winner Tony Hale with a musical performance by Lizzo. Winners of the night included Bill Nye, Shay Mitchell, Doug the Pug, Gigi Gorgeous, Simone Biles, Mara Wilson, Gaten Matarazzo and Chrissy Teigen. == 10th Shorty Awards == The 10th Annual Shorty Awards, took place on April 15, 2018, at the PlayStation Theater, New York City. The ceremony was hosted by actress, singer, and songwriter Keke Palmer with a musical performance by Betty Who. == 11th Shorty Awards == The 11th Annual Shorty Awards were held on May 5, 2019, in New York City at the PlayStation Theater. The ceremony was hosted by American actress and comedian Kathy Griffin, with a musical performance by Tank and the Bangas. == 12th Shorty Awards == The 12th Annual Shorty Awards were held on May 3, 2020. Due to the COVID-19 pandemic, the ceremony took place online for the first time, with presenters and award winners filming from their own homes. The ceremony was hosted by actor J.B. Smoove and featured a remixed performance of Trap Queen by Fetty Wap. Award winners included Jack Stauber, Supercar Blondie, Rose and Rosie, and Greta Thunberg. == 13th Shorty Awards == The 13th Annual Shorty Awards took place from April 26 to May 14, 2021. The ceremony was hosted on different social media platforms, such as Instagram and Clubhouse, to create a more tailored experience. Winners were announced from May 11 to May 14, with 10 winners being revealed each hour from 1 to 4 p.m. EST on the Shorty Awards Instagram account. == 14th Shorty Awards == The 14th Annual Shorty Awards were held virtually on May 15, 2022, honoring the best in social media and digital content. Hosted by Jay Shetty, the event recognized influencers, brands, and organizations across various categories, celebrating excellence in digital storytelling and innovative online campaigns. Notable winners included Tabitha Brown for her food content and the D'Amelio Family for their contributions to family and parenting content. The event highlighted the role of digital media in connecting and inspiring audiences during challenging times. == 15th Shorty Awards == The 15th Annual Shorty Awards celebrated the best in social media and digital content on May 24, 2023, at Tribeca 360° in New York City. Hosted by Jay Pharoah, the event honored creators, brands, and organizations ac

Word error rate

Word error rate (WER) is a common metric of the performance of a speech recognition or machine translation system. The WER metric typically ranges from 0 to 1, where 0 indicates that the compared pieces of text are exactly identical, and 1 (or larger) indicates that they are completely different with no similarity. This way, a WER of 0.8 means that there is an 80% error rate for compared sentences. The general difficulty of measuring performance lies in the fact that the recognized word sequence can have a different length from the reference word sequence (supposedly the correct one). The WER is derived from the Levenshtein distance, working at the word level instead of the phoneme level. The WER is a valuable tool for comparing different systems as well as for evaluating improvements within one system. This kind of measurement, however, provides no details on the nature of translation errors and further work is therefore required to identify the main source(s) of error and to focus any research effort. This problem is solved by first aligning the recognized word sequence with the reference (spoken) word sequence using dynamic string alignment. Examination of this issue is seen through a theory called the power law that states the correlation between perplexity and word error rate. Word error rate can then be computed as: W E R = S + D + I N = S + D + I S + D + C {\displaystyle {\mathit {WER}}={\frac {S+D+I}{N}}={\frac {S+D+I}{S+D+C}}} where S is the number of substitutions, D is the number of deletions, I is the number of insertions, C is the number of correct words, N is the number of words in the reference (N=S+D+C) The intuition behind 'deletion' and 'insertion' is how to get from the reference to the hypothesis. So if we have the reference "This is wikipedia" and hypothesis "This _ wikipedia", we call it a deletion. Note that since N is the number of words in the reference, the word error rate can be larger than 1.0, namely if the number of insertions I is larger than the number of correct words C. When reporting the performance of a speech recognition system, sometimes word accuracy (WAcc) is used instead: W A c c = 1 − W E R = N − S − D − I N = C − I N {\displaystyle {\mathit {WAcc}}=1-{\mathit {WER}}={\frac {N-S-D-I}{N}}={\frac {C-I}{N}}} Since the WER can be larger than 1.0, the word accuracy can be smaller than 0.0. == Experiments == It is commonly believed that a lower word error rate shows superior accuracy in recognition of speech, compared with a higher word error rate. However, at least one study has shown that this may not be true. In a Microsoft Research experiment, it was shown that, if people were trained under "that matches the optimization objective for understanding", (Wang, Acero and Chelba, 2003) they would show a higher accuracy in understanding of language than other people who demonstrated a lower word error rate, showing that true understanding of spoken language relies on more than just high word recognition accuracy. == Other metrics == One problem with using a generic formula such as the one above, however, is that no account is taken of the effect that different types of error may have on the likelihood of successful outcome, e.g. some errors may be more disruptive than others and some may be corrected more easily than others. These factors are likely to be specific to the syntax being tested. A further problem is that, even with the best alignment, the formula cannot distinguish a substitution error from a combined deletion plus insertion error. Hunt (1990) has proposed the use of a weighted measure of performance accuracy where errors of substitution are weighted at unity but errors of deletion and insertion are both weighted only at 0.5, thus: W E R = S + 0.5 D + 0.5 I N {\displaystyle {\mathit {WER}}={\frac {S+0.5D+0.5I}{N}}} There is some debate, however, as to whether Hunt's formula may properly be used to assess the performance of a single system, as it was developed as a means of comparing more fairly competing candidate systems. A further complication is added by whether a given syntax allows for error correction and, if it does, how easy that process is for the user. There is thus some merit to the argument that performance metrics should be developed to suit the particular system being measured. Whichever metric is used, however, one major theoretical problem in assessing the performance of a system is deciding whether a word has been “mis-pronounced,” i.e. does the fault lie with the user or with the recogniser. This may be particularly relevant in a system which is designed to cope with non-native speakers of a given language or with strong regional accents. The pace at which words should be spoken during the measurement process is also a source of variability between subjects, as is the need for subjects to rest or take a breath. All such factors may need to be controlled in some way. For text dictation it is generally agreed that performance accuracy at a rate below 95% is not acceptable, but this again may be syntax and/or domain specific, e.g. whether there is time pressure on users to complete the task, whether there are alternative methods of completion, and so on. The term "Single Word Error Rate" is sometimes referred to as the percentage of incorrect recognitions for each different word in the system vocabulary. == Edit distance == The word error rate may also be referred to as the length normalized edit distance. The normalized edit distance between X and Y, d( X, Y ) is defined as the minimum of W( P ) / L ( P ), where P is an editing path between X and Y, W ( P ) is the sum of the weights of the elementary edit operations of P, and L(P) is the number of these operations (length of P).

Knapsack problem

The knapsack problem is the following problem in combinatorial optimization: Given a set of items, each with a weight and a value, determine which items to include in the collection so that the total weight is less than or equal to a given limit and the total value is as large as possible. It derives its name from the problem faced by someone who is constrained by a fixed-size knapsack and must fill it with the most valuable items. The problem often arises in resource allocation where the decision-makers have to choose from a set of non-divisible projects or tasks under a fixed budget or time constraint, respectively. The knapsack problem has been studied for more than a century, with early works dating back to 1897. The subset sum problem is a special case of the decision and 0-1 problems where for each kind of item, the weight equals the value: w i = v i {\displaystyle w_{i}=v_{i}} . In the field of cryptography, the term knapsack problem is often used to refer specifically to the subset sum problem. The subset sum problem is one of Karp's 21 NP-complete problems. == Applications == Knapsack problems appear in real-world decision-making processes in a wide variety of fields, such as finding the least wasteful way to cut raw materials, selection of investments and portfolios, selection of assets for asset-backed securitization, and generating keys for the Merkle–Hellman and other knapsack cryptosystems. One early application of knapsack algorithms was in the construction and scoring of tests in which the test-takers have a choice as to which questions they answer. For small examples, it is a fairly simple process to provide the test-takers with such a choice. For example, if an exam contains 12 questions each worth 10 points, the test-taker need only answer 10 questions to achieve a maximum possible score of 100 points. However, on tests with a heterogeneous distribution of point values, it is more difficult to provide choices. Feuerman and Weiss proposed a system in which students are given a heterogeneous test with a total of 125 possible points. The students are asked to answer all of the questions to the best of their abilities. Of the possible subsets of problems whose total point values add up to 100, a knapsack algorithm would determine which subset gives each student the highest possible score. A 1999 study of the Stony Brook University Algorithm Repository showed that, out of 75 algorithmic problems related to the field of combinatorial algorithms and algorithm engineering, the knapsack problem was the 19th most popular and the third most needed after suffix trees and the bin packing problem. == Definition == The most common problem being solved is the 0-1 knapsack problem, which restricts the number x i {\displaystyle x_{i}} of copies of each kind of item to zero or one. Given a set of n {\displaystyle n} items numbered from 1 up to n {\displaystyle n} , each with a weight w i {\displaystyle w_{i}} and a value v i {\displaystyle v_{i}} , along with a maximum weight capacity W {\displaystyle W} , maximize ∑ i = 1 n v i x i {\displaystyle \sum _{i=1}^{n}v_{i}x_{i}} subject to ∑ i = 1 n w i x i ≤ W {\displaystyle \sum _{i=1}^{n}w_{i}x_{i}\leq W} and x i ∈ { 0 , 1 } {\displaystyle x_{i}\in \{0,1\}} . Here x i {\displaystyle x_{i}} represents the number of instances of item i {\displaystyle i} to include in the knapsack. Informally, the problem is to maximize the sum of the values of the items in the knapsack so that the sum of the weights is less than or equal to the knapsack's capacity. The bounded knapsack problem (BKP) removes the restriction that there is only one of each item, but restricts the number x i {\displaystyle x_{i}} of copies of each kind of item to a maximum non-negative integer value c {\displaystyle c} : maximize ∑ i = 1 n v i x i {\displaystyle \sum _{i=1}^{n}v_{i}x_{i}} subject to ∑ i = 1 n w i x i ≤ W {\displaystyle \sum _{i=1}^{n}w_{i}x_{i}\leq W} and x i ∈ { 0 , 1 , 2 , … , c } . {\displaystyle x_{i}\in \{0,1,2,\dots ,c\}.} The unbounded knapsack problem (UKP) places no upper bound on the number of copies of each kind of item and can be formulated as above except that the only restriction on x i {\displaystyle x_{i}} is that it is a non-negative integer. maximize ∑ i = 1 n v i x i {\displaystyle \sum _{i=1}^{n}v_{i}x_{i}} subject to ∑ i = 1 n w i x i ≤ W {\displaystyle \sum _{i=1}^{n}w_{i}x_{i}\leq W} and x i ∈ N . {\displaystyle x_{i}\in \mathbb {N} .} One example of the unbounded knapsack problem is given using the figure shown at the beginning of this article and the text "if any number of each book is available" in the caption of that figure. == Computational complexity == The knapsack problem is interesting from the perspective of computer science for many reasons: The decision problem form of the knapsack problem (Can a value of at least V be achieved without exceeding the weight W?) is NP-complete, thus there is no known algorithm that is both correct and fast (polynomial-time) in all cases. There is no known polynomial algorithm which can tell, given a solution, whether it is optimal (which would mean that there is no solution with a larger V). This problem is co-NP-complete. There is a pseudo-polynomial time algorithm using dynamic programming. There is a fully polynomial-time approximation scheme, which uses the pseudo-polynomial time algorithm as a subroutine, described below. Many cases that arise in practice, and "random instances" from some distributions, can nonetheless be solved exactly. There is a link between the "decision" and "optimization" problems in that if there exists a polynomial algorithm that solves the "decision" problem, then one can find the maximum value for the optimization problem in polynomial time by applying this algorithm iteratively while increasing the value of k. On the other hand, if an algorithm finds the optimal value of the optimization problem in polynomial time, then the decision problem can be solved in polynomial time by comparing the value of the solution output by this algorithm with the value of k. Thus, both versions of the problem are of similar difficulty. One theme in research literature is to identify what the "hard" instances of the knapsack problem look like, or viewed another way, to identify what properties of instances in practice might make them more amenable than their worst-case NP-complete behaviour suggests. The goal in finding these "hard" instances is for their use in public-key cryptography systems, such as the Merkle–Hellman knapsack cryptosystem. More generally, better understanding of the structure of the space of instances of an optimization problem helps to advance the study of the particular problem and can improve algorithm selection. Furthermore, notable is the fact that the hardness of the knapsack problem depends on the form of the input. If the weights and profits are given as integers, it is weakly NP-complete, while it is strongly NP-complete if the weights and profits are given as rational numbers. However, in the case of rational weights and profits it still admits a fully polynomial-time approximation scheme. === Unit-cost models === The NP-hardness of the Knapsack problem relates to computational models in which the size of integers matters (such as the Turing machine). In contrast, decision trees count each decision as a single step. Dobkin and Lipton show an 1 2 n 2 {\displaystyle {1 \over 2}n^{2}} lower bound on linear decision trees for the knapsack problem, that is, trees where decision nodes test the sign of affine functions. This was generalized to algebraic decision trees by Steele and Yao. If the elements in the problem are real numbers or rationals, the decision-tree lower bound extends to the real random-access machine model with an instruction set that includes addition, subtraction and multiplication of real numbers, as well as comparison and either division or remaindering ("floor"). This model covers more algorithms than the algebraic decision-tree model, as it encompasses algorithms that use indexing into tables. However, in this model all program steps are counted, not just decisions. An upper bound for a decision-tree model was given by Meyer auf der Heide who showed that for every n there exists an O(n4)-deep linear decision tree that solves the subset-sum problem with n items. Note that this does not imply any upper bound for an algorithm that should solve the problem for any given n. == Solving == Several algorithms are available to solve knapsack problems, based on the dynamic programming approach, the branch and bound approach or hybridizations of both approaches. === Dynamic programming in-advance algorithm === The unbounded knapsack problem (UKP) places no restriction on the number of copies of each kind of item. Besides, here we assume that x i > 0 {\displaystyle x_{i}>0} m [ w ′ ] = max ( ∑ i = 1 n v i x i ) {\displaystyle m[w']=\max \left(\sum _{i=1}^{n}v_{i}x_{i}\right)} subject to ∑