In computational linguistics, second-order co-occurrence pointwise mutual information (SOC-PMI) is a method used to measure semantic similarity, or how close in meaning two words are. The method does not require the two words to appear together in a text. Instead, it works by analyzing the "neighbor" words that typically appear alongside each of the two target words in a large body of text (corpus). If the two target words frequently share the same neighbors, they are considered semantically similar. For example, the words "cemetery" and "graveyard" may not appear in the same sentence often, but they both frequently appear near words like "buried," "dead," and "funeral." SOC-PMI uses this shared context to determine that they have a similar meaning. The method is called "second-order" because it doesn't look at the direct co-occurrence of the target words (which would be first-order), but at the co-occurrence of their neighbors (a second level of association). The strength of these associations is quantified using pointwise mutual information (PMI). == History == The method builds on earlier work like the PMI-IR algorithm, which used the AltaVista search engine to calculate word association probabilities. The key advantage of a second-order approach like SOC-PMI is its ability to measure similarity between words that do not co-occur often, or at all. The British National Corpus (BNC) has been used as a source for word frequencies and contexts for this method. == Methodology == The SOC-PMI algorithm measures the similarity between two words, w 1 {\displaystyle w_{1}} and w 2 {\displaystyle w_{2}} , in several steps. === Step 1: Score neighboring words with PMI === First, for each target word ( w 1 {\displaystyle w_{1}} and w 2 {\displaystyle w_{2}} ), the algorithm identifies its "neighbor" words within a certain text window (e.g., within 5 words to the left or right) across a large corpus. The strength of the association between a target word t i {\displaystyle t_{i}} and its neighbor w {\displaystyle w} is calculated using pointwise mutual information (PMI). A higher PMI value means the two words appear together more often than would be expected by chance. The PMI between a target word t i {\displaystyle t_{i}} and a neighbor word w {\displaystyle w} is calculated as: f pmi ( t i , w ) = log 2 f b ( t i , w ) × m f t ( t i ) f t ( w ) {\displaystyle f^{\text{pmi}}(t_{i},w)=\log _{2}{\frac {f^{b}(t_{i},w)\times m}{f^{t}(t_{i})f^{t}(w)}}} where: f b ( t i , w ) {\displaystyle f^{b}(t_{i},w)} is the number of times t i {\displaystyle t_{i}} and w {\displaystyle w} appear together in the context window. f t ( t i ) {\displaystyle f^{t}(t_{i})} is the total number of times t i {\displaystyle t_{i}} appears in the corpus. f t ( w ) {\displaystyle f^{t}(w)} is the total number of times w {\displaystyle w} appears in the corpus. m {\displaystyle m} is the total number of tokens (words) in the corpus. === Step 2: Create a semantic 'signature' for each word === For each target word ( w 1 {\displaystyle w_{1}} and w 2 {\displaystyle w_{2}} ), the algorithm creates a list of its most significant neighbors. This is done by taking the top β {\displaystyle \beta } neighbor words, sorted in descending order by their PMI score with the target word. This list of top neighbors, X w {\displaystyle X^{w}} , acts as a semantic "signature" for the word w {\displaystyle w} . X w = { X i w } {\displaystyle X^{w}=\{X_{i}^{w}\}} , for i = 1 , 2 , … , β {\displaystyle i=1,2,\ldots ,\beta } The size of this list, β {\displaystyle \beta } , is a parameter of the method. === Step 3: Compare the signatures === The algorithm then compares the signatures of w 1 {\displaystyle w_{1}} and w 2 {\displaystyle w_{2}} . It looks for words that are present in both signatures. The similarity of w 1 {\displaystyle w_{1}} to w 2 {\displaystyle w_{2}} is calculated by summing the PMI scores of w 2 {\displaystyle w_{2}} with every word in w 1 {\displaystyle w_{1}} 's signature list. The β {\displaystyle \beta } -PMI summation function defines this score. The score for w 1 {\displaystyle w_{1}} with respect to w 2 {\displaystyle w_{2}} is: f ( w 1 , w 2 , β ) = ∑ i = 1 β ( f pmi ( X i w 1 , w 2 ) ) γ {\displaystyle f(w_{1},w_{2},\beta )=\sum _{i=1}^{\beta }(f^{\text{pmi}}(X_{i}^{w_{1}},w_{2}))^{\gamma }} This sum only includes terms where the PMI value is positive. The exponent γ {\displaystyle \gamma } (with a value > 1) is used to give more weight to neighbors that are more strongly associated with w 2 {\displaystyle w_{2}} . This calculation is done in both directions: The similarity of w 1 {\displaystyle w_{1}} with respect to w 2 {\displaystyle w_{2}} : f ( w 1 , w 2 , β 1 ) = ∑ i = 1 β 1 ( f pmi ( X i w 1 , w 2 ) ) γ {\displaystyle f(w_{1},w_{2},\beta _{1})=\sum _{i=1}^{\beta _{1}}(f^{\text{pmi}}(X_{i}^{w_{1}},w_{2}))^{\gamma }} The similarity of w 2 {\displaystyle w_{2}} with respect to w 1 {\displaystyle w_{1}} : f ( w 2 , w 1 , β 2 ) = ∑ i = 1 β 2 ( f pmi ( X i w 2 , w 1 ) ) γ {\displaystyle f(w_{2},w_{1},\beta _{2})=\sum _{i=1}^{\beta _{2}}(f^{\text{pmi}}(X_{i}^{w_{2}},w_{1}))^{\gamma }} === Step 4: Calculate final similarity score === Finally, the total semantic similarity is the average of the two scores from the previous step. S i m ( w 1 , w 2 ) = f ( w 1 , w 2 , β 1 ) β 1 + f ( w 2 , w 1 , β 2 ) β 2 {\displaystyle \mathrm {Sim} (w_{1},w_{2})={\frac {f(w_{1},w_{2},\beta _{1})}{\beta _{1}}}+{\frac {f(w_{2},w_{1},\beta _{2})}{\beta _{2}}}} This score can be normalized to fall between 0 and 1. For example, using this method, the words cemetery and graveyard achieve a high similarity score of 0.986 (with specific parameter settings).
Vigloo
Vigloo (Korean: 비글루) is a South Korean microdrama, also known as short-form drama, series streaming platform owned by SpoonLabs, with headquarters in Seoul. It provides content produced in South Korea, Japan, and the United States. Vigloo produced the first AI-created short-form drama in South Korea. == History == Vigloo launched in July 2024. After receiving an equity investment of $86 million (₩120 billion) by South Korean video game company Krafton in September 2024, Vigloo expanded to the U.S. In January 2025, Vigloo unveiled its first in-house produced drama, Xs Who Want to Kill: Adultery Investigation Unit. Vigloo had been testing the use of AI in post-production and visual effects, and in October 2025 released two original dramas produced entirely with AI. It adapted its live action Japanese short-form drama Boyfriend Search Project – Kissing 5 Men into the first short-form animation series made with AI technology in South Korea. Of the top free entertainment iOS apps in South Korea, Vigloo ranks Number 3 as of January 2026. == Service == === Content === Vigloo offers both original and licensed content. It partnered with Passionflix to repackage the latter's original series The Secret Life of Amy Bensen into 35 vertical "bite-sized episodes". The most popular genre is romance, such as romantasy. === Business Model === Vigloo is available around the world, providing subtitles in nine languages, including Korean, English, and Japanese. Fifty percent of Vigloo's revenue comes from the U.S. Vigloo operates on a freemium model, where viewers can try several episodes and then can choose to continue by subscription or in-app purchases. As of September 2025, 70% of Vigloo viewers were over 35 years old. === Microdramas === Emerging during the early COVID period in China, microdramas have grown into a 7-billion-dollar market with dozens of dedicated platforms now operating. Although the format first expanded across Asia, short-form scripted content optimized for mobile viewing is increasingly being produced and watched in markets worldwide. == Series == A Vampire in the Alpha's Den Fight for Love Matrimoney Signed, Sealed, Deceived by My Billionaire Mailboy Spring Break Bucket List Stake to the Heart
Social computing
Social computing is an area of computer science that is concerned with the intersection of social behavior and computational systems. It is based on creating or fostering existing social conventions and social contexts through the use of software and technology. Blogs, email, instant messaging, social network services, wikis, social bookmarking and other instances of what is often called social software illustrate ideas from social computing. The rise in social computing is attributed to the prevalence of personal devices and increased overall computing power. This enables a growing number of users to participate in sharing content and interact with another. == Definitions == Humans—and human behavior—are profoundly social. Humans tend to orient to one another and develop abilities to interact with each other and other species. This ranges from expression and gesture through spoken, written, and body language. Humans are influenced by the behavior of those around them and can rely on social context and cues to make decisions. An example of a behavior relying on social contexts is applauding at the end of the play. This is based on the context that the show ended, and other audience members are applauding. Social information provides a basis for inferences, planning, and coordinating activity. == Examples == Common tools include blogs, email, instant messaging, social networking sites, wikis, and social bookmarking platforms. These technologies enable users to generate content, share knowledge, and interact in real time. == Applications == The rise of social computing has highlighted opportunities for businesses. Businesses are interacting on social computing platforms and investing in facilities to support and research social computing.Business models can leverage the massive customer bases that accumulate through social computing channels. Some organizations have started their own blogs and networks (McAfee, 2006, Joe, 2005). Organizations from diverse industry sectors such as Google, Cisco, and Fox, have sought to acquire or invest in successful social computing enterprises. A business blog can serve as a source of information and promotion for the company. This allows the company to share content about the company and their initiatives. Businesses have also interacted with social computing to market themselves and interact with customers. A notable example is Wendy's with their X (formerly Twitter) account. The account was primarily used to promote business promotions and interact with users in a playful or meaningful way. E-commerce web sites have allowed users to leave reviews and feedback on purchases which has improved online shopping experience for sellers and consumers.As another example of social computing’s business applications, many e-commerce Web sites have adopted online product/vendor feedback/reputation systems. Such systems provide an asynchronous platform for the consumer community to share experiences collectively and influence their purchasing behavior. They also provide a vehicle for eliciting feedback information valuable to the vendors and e-commerce site operators.Consumers can use the feedback systems to make a more educated choice on a purchase by comparing reviews between products or vendors. Sellers can track consumer behaviors and trends regarding a product and adjust their supply according to the demand. == Challenges and criticism == Social computing raises several concerns related to privacy, data security, and algorithmic bias. The widespread collection and analysis of user-generated data can lead to ethical dilemmas, especially when users are unaware of how their information is used. Critics also highlight issues of digital labor, surveillance, and the spread of misinformation, which can influence public opinion and social dynamics. === Term appearance === The term appeared in the mid 1990s after technology advancements and development of the web. In 1994, the concept of social computing was first proposed by Schuler. He thought, "Social computing is a computing application, with software as the medium or focus of social relationships." === Premise === The premise of social computing is that it is possible to design digital systems that support useful functionality by making socially produced information available to their users. This information may be provided directly, as when systems show the number of users who have rated a review as helpful or not. Or the information may be provided after being filtered and aggregated, as is done when systems recommend a product based on what else people with similar purchase history have purchased. Alternatively, the information may be provided indirectly, as is the case with Google's page rank algorithms which orders search results based on the number of pages that (recursively) point to them. In all of these cases, information that is produced by a group of people is used to provide or enhance the functioning of a system. Social computing is concerned with systems of this sort and the mechanisms and principles that underlie them. Social computing can be defined as follows: "Social Computing" refers to systems that support the gathering, representation, processing, use, and dissemination of information that is distributed across social collectivities such as teams, communities, organizations, and markets. Moreover, the information is not "anonymous" but is significantly precise because it is linked to people, who are in turn linked to other people. More recent definitions, however, have foregone the restrictions regarding anonymity of information, acknowledging the continued spread and increasing pervasiveness of social computing. As an example, Hemmatazad, N. (2014) defined social computing as "the use of computational devices to facilitate or augment the social interactions of their users, or to evaluate those interactions in an effort to obtain new information." Social computing has to do with supporting "computations" that are carried out by groups of people, an idea that has been popularized in James Surowiecki's book, The Wisdom of Crowds. Examples of social computing in this sense include collaborative filtering, online auctions, reputation systems, computational social choice, tagging, and verification games. The social information processing page focuses on this sense of social computing. == History == === Technology infrastructure === Users were able to interact more with websites after the development of Web 2.0. This was an advancement from Web 1.0. Comode G. and Krishnamurthy B. (2008) note that "content creators were few in Web 1.0 with the vast majority of users simply acting as consumers of content." Web 2.0 provided functionalities that allowed for low-cost web-hosting services and introduced features with browser windows that used basic information structure and expanded it to as many devices as possible using HTTP, or Hypertext Transfer Protocol. Sometimes referred to as "Enterprise 2.0", a term derived from Web 2.0, social software for enterprise generally refers to the use of social computing in corporate intranets and in other medium- and large-scale business environments. It consisted of a class of tools that allowed for networking and social changes to businesses at the time. It was a layering of the business tools on Web 2.0 and brought forth several applications and collaborative software with specific uses. FinanceElectronic negotiation, which first came up in 1969 and was adapted over time to suit financial markets networking needs, represents an important and desirable coordination mechanism for electronic markets. Negotiation between agents (software agents as well as humans) allows cooperative and competitive sharing of information to determine a proper price. Recent research and practice has also shown that electronic negotiation is beneficial for the coordination of complex interactions among organizations. Electronic negotiation has recently emerged as a very dynamic, interdisciplinary research area covering aspects from disciplines such as Economics, Information Systems, Computer Science, Communication Theory, Sociology and Psychology.Social computing has become more widely known because of its relationship to a number of recent trends. These include the growing popularity of social software and Web 3.0, increased academic interest in social network analysis, the rise of open source as a viable method of production, and a growing conviction that all of this can have a profound impact on daily life. A February 13, 2006 paper by market research company Forrester Research suggested that: === Developments === PLATO was one of the earliest examples of social computing in a live production environment with initially hundreds and soon thousands of users. The PLATO computer system was developed by the University of Illinois at Urbana Champaign in 1960s. In the 70s, the system supported social software applications for multi-us
Media engagement framework
The media engagement framework is a planning framework used by marketing professionals to understand the behavior of social media marketing-based audiences. The construct was introduced in the book, ROI of Social Media. Powell’s background in marketing ROI and Groves' experience and understanding of the applications of social media in business led to a collaboration. Dimos joined as a brand strategist for Litmus Group, a global management consulting firm. The media engagement framework consists of the definitions of personas (Individuals, Consumers and Influencers), referenced by the competitive set or constraint that applies to that persona and the measurement framework that might be applied to those personas. It is referenced at the center of the marketing process diagram, surrounded by the marketing functions of strategy, tactics, metrics and ROI. The marketing process diagram describes how the media engagement framework can apply to any strategic marketing activity but was developed to establish a completely integrated framework describing how both traditional and social media marketing activities can be planned, executed, measured and improved. == Application == The media engagement framework provides a strategic planning construct in which measurements and metrics play a crucial role. Applying the media engagement framework aids in the development and management of an effective online marketing presence leveraging social media to engage a market or audience. By first personifying the audience, the marketer is able to identify the limiting aspect of the engagements possible with that audience segment and then, understand the type of engagement metrics to apply. Each persona makes decisions differently about how he/she acts in the social media universe. A framework metric can be applied for each of these personas: Endorsement funnel for influencers Community engagement funnel for individuals Purchase funnel for consumers Individuals, influencers and consumers make decisions based on alternatives available to them and constraints put on them. To engage with an individual brands must realize they are competing against the time an individual spends on line. If they find something else more engaging, they will engage with that activity. Brands compete against other brands for the purchases of consumers acting in the category. Lastly, influencers have only so many endorsements they can make and therefore brands compete with other endorsers for the endorsement of an influencer. Creating engaging content by keeping target audience in mind like create content that audience find it funny, interesting, and relatable will encourage audience to share it on social networks. Which will be beneficial for you brand, getting more people to know about your business and brand. Contact Digilord to create engaging content for your brand. Use of listening tools (Google Alerts, Twitter Search, SocialMention.com, Veooz.com, Alterian SM2, Radian6, Sysomos, Buzzient etc.) can be employed within the model to help identify the members of the audience segment and to support the formation of other social engagement planning and management tools.
Professional network service
A professional network service (or, in an Internet context, simply a professional network) is a type of social network service that focuses on interactions and relationships for business opportunities and career growth, with less emphasis on activities in personal life. A professional network service is used by working individuals, job-seekers, and businesses to establish and maintain professional contacts, to find work or hire employees, share professional achievements, sell or promote services, and stay up-to-date with industry news and trends. According to LinkedIn managing director Clifford Rosenberg in an interview with AAP in 2010, "[t]his is a call to action for professionals to re-address their use of social networks and begin to reap as many rewards from networking professionally as they do personally." Businesses mostly depend on resources and information outside the company and to get what they need, they need to reach out and professionally network with others, such as employees or clients as well as potential opportunities. "Nardi, Whittaker, and Schwarz (2002) point out three main tasks that they believe networkers need to attend to keep a successful professional (intentional) network: building a network, maintaining the network, and activating selected contacts. They stress that networkers need to continue to add new contacts to their network to access as many resources as possible and to maintain their network by staying in touch with their contacts. This is so that the contacts are easy to activate when the networker has work that needs to be done." By using a professional network service, businesses can keep all of their networks up-to-date, and in order, and helps figure out the best way to efficiently get in touch with each of them. A service that can do all that helps relieve some of the stress when trying to get things done. Not all professional network services are online sites that help promote a business. Some services connect the user to other services that help promote the business other than online sites, such as phone/Internet companies that provide services and companies that specifically are designed to do all of the promoting, online and in person, for a business. == History == In 1997, professional network services started up throughout the world and continue to grow. The first recognizable site to combine all features, such as creating profiles, adding friends, and searching for friends, was SixDegrees.com. According to Boyd and Ellison's article, "Social Network Sites: Definition, History, and Scholarship", from 1997 to 2001, several community tools began supporting various combinations of profiles and publicly articulated Friends. Boyd and Ellison go on to say that the next wave began with Ryze.com in 2001. It was introduced as a new way "to help people leverage their business networks". == Inside the works == Quite a lot of work is put into a professional network service, such as the number of hours that go into them and the type of people they work for, as well as the business model of it all, such as the professional interaction and the multiple services they deal with. === Types of services === Some professional network services not only help promote the business but can also help in connecting to other people. Those services may include a specific phone and/or Internet company or a company that helps to connect with other businesses. According to the Society for New Communications Research (SNCR), there are at least nine online professional networks that are being used. === Professional interaction === Kaplan and Haenlein elaborate on five key considerations for companies when utilizing media. These include the importance of careful selection, the option to choose existing applications or develop custom ones, ensuring alignment with organizational activities, integrating a comprehensive media plan, and providing accessibility to all stakeholders. ==== Choose carefully ==== "Choosing the right medium for any given purpose depends on the target group to be reached and the message to be communicated. On one hand, each Social Media application usually attracts a certain group of people, and firms should be active wherever their customers are present. On the other hand, there may be situations whereby certain features are necessary to ensure effective communication, and these features are only offered by one specific application." ==== Ensure activity alignment ==== "Sometimes you may decide to rely on various Social Media, or a set of different applications within the same group, to have the largest possible reach." "Using different contact channels can be a worthwhile and profitable strategy." According to the Society for New Communications Research at Harvard University, "the average professional belongs to 3–5 online networks for business use, and LinkedIn, Facebook, and Twitter are among the top used." ==== Integrate a media plan ==== Social media and traditional media are "both part of the same: your corporate image" in the customers' eyes. ==== Allow access to all ==== "...once the firm has decided to utilize Social Media applications, it is worth checking that all employees may access them." According to the SNCR, "the convergence of Internet, mobile, and social media has taken significant shape as professionals rely on anywhere access to information, relationships, and networks." ==== Online usage ==== "Half of the respondents report participating in 3 to 5 online professional networks. Another three in ten participate in 6 or more professional networks." "Popular social networks are now being used frequently as Professional Communities. More than nine in ten respondents indicated that they use LinkedIn and half reported using Facebook. Twitter and blogs were frequently listed as 'professional networks'." === Business model === According to Michael Rappa's article, Business models on the Web", "a business model is the method of doing business by which a company can sustain itself – that is, generate revenue. The business model spells out how a company makes money by specifying where it is positioned in the value chain." Rappa mentions that there are at least nine basic categories from which a business model can be separated. Those categories are a brokerage, advertising, infomediary, merchant, manufacturer, affiliate, community, subscription, and utility. "...a firm may combine several different models as part of its overall Internet business strategy." At first, Flickr started as a way to mainstream public relations. == Social impact == When it comes to the social impact that professional network services have on today's society, it has proved to increase activity. According to the SNCR, "[t]hree quarters of respondents rely on professional networks to support business decisions. Reliance has increased for essentially all respondents over the past three years. Younger (20–35) and older professionals (55+) are more active users of social tools than middle-aged professionals. More people are collaborating outside their company wall than within their organizational intranet." == Limitations == Since the internet and social media are a part of this "world where consumers can speak so freely with each other and businesses have increasingly less control over the information available about them in cyberspace", most firms and businesses are uncomfortable with all the freedom. According to Kaplan and Haenlein's article, "Users of the world, unite! The challenges and opportunities of Social Media", businesses are pushed aside and are only able to sit back and watch as their customers publicly post comments, which may or may not be well-written.
Multi-task learning
Multi-task learning (MTL) is a subfield of machine learning in which multiple learning tasks are solved at the same time, while exploiting commonalities and differences across tasks. This can result in improved learning efficiency and prediction accuracy for the task-specific models, when compared to training the models separately. Inherently, Multi-task learning is a multi-objective optimization problem having trade-offs between different tasks. Early versions of MTL were called "hints". In a widely cited 1997 paper, Rich Caruana gave the following characterization:Multitask Learning is an approach to inductive transfer that improves generalization by using the domain information contained in the training signals of related tasks as an inductive bias. It does this by learning tasks in parallel while using a shared representation; what is learned for each task can help other tasks be learned better. In the classification context, MTL aims to improve the performance of multiple classification tasks by learning them jointly. One example is a spam-filter, which can be treated as distinct but related classification tasks across different users. To make this more concrete, consider that different people have different distributions of features which distinguish spam emails from legitimate ones, for example an English speaker may find that all emails in Russian are spam, not so for Russian speakers. Yet there is a definite commonality in this classification task across users, for example one common feature might be text related to money transfer. Solving each user's spam classification problem jointly via MTL can let the solutions inform each other and improve performance. Further examples of settings for MTL include multiclass classification and multi-label classification. Multi-task learning works because regularization induced by requiring an algorithm to perform well on a related task can be superior to regularization that prevents overfitting by penalizing all complexity uniformly. One situation where MTL may be particularly helpful is if the tasks share significant commonalities and are generally slightly under sampled. However, as discussed below, MTL has also been shown to be beneficial for learning unrelated tasks. == Methods == The key challenge in multi-task learning, is how to combine learning signals from multiple tasks into a single model. This may strongly depend on how well different task agree with each other, or contradict each other. There are several ways to address this challenge: === Task grouping and overlap === Within the MTL paradigm, information can be shared across some or all of the tasks. Depending on the structure of task relatedness, one may want to share information selectively across the tasks. For example, tasks may be grouped or exist in a hierarchy, or be related according to some general metric. Suppose, as developed more formally below, that the parameter vector modeling each task is a linear combination of some underlying basis. Similarity in terms of this basis can indicate the relatedness of the tasks. For example, with sparsity, overlap of nonzero coefficients across tasks indicates commonality. A task grouping then corresponds to those tasks lying in a subspace generated by some subset of basis elements, where tasks in different groups may be disjoint or overlap arbitrarily in terms of their bases. Task relatedness can be imposed a priori or learned from the data. Hierarchical task relatedness can also be exploited implicitly without assuming a priori knowledge or learning relations explicitly. For example, the explicit learning of sample relevance across tasks can be done to guarantee the effectiveness of joint learning across multiple domains. === Exploiting unrelated tasks: Auxiliary learning === In auxiliary learning, one attempts learning a group of principal tasks using a group of auxiliary tasks, unrelated to the principal ones. With the right unrelated tasks, joint learning of unrelated tasks which use the same input data have been shown to be beneficial, and provide significant improvement over standard MTL. The reason is that prior knowledge about task relatedness can lead to sparser and more informative representations for each task grouping, essentially by screening out idiosyncrasies of the data distribution. It has been proposed to build on a prior multitask methodology by favoring a shared low-dimensional representation within each task grouping, and imposing a penalty on tasks from different groups which encourages the two representations to be orthogonal. Learning with auxiliary unrelated tasks poses two major challenges: Finding useful auxiliary tasks and combining losses of all tasks in a useful way. Some methods can learn these from data together with the training process, and combine tasks efficiently. === Transfer of knowledge === Related to multi-task learning is the concept of knowledge transfer. Whereas traditional multi-task learning implies that a shared representation is developed concurrently across tasks, transfer of knowledge implies a sequentially shared representation. Large scale machine learning projects such as the deep convolutional neural network GoogLeNet, an image-based object classifier, can develop robust representations which may be useful to further algorithms learning related tasks. For example, the pre-trained model can be used as a feature extractor to perform pre-processing for another learning algorithm. Or the pre-trained model can be used to initialize a model with similar architecture which is then fine-tuned to learn a different classification task. === Multiple non-stationary tasks === Traditionally Multi-task learning and transfer of knowledge are applied to stationary learning settings. Their extension to non-stationary environments is termed Group online adaptive learning (GOAL). Sharing information could be particularly useful if learners operate in continuously changing environments, because a learner could benefit from previous experience of another learner to quickly adapt to their new environment. Such group-adaptive learning has numerous applications, from predicting financial time-series, through content recommendation systems, to visual understanding for adaptive autonomous agents. === Multi-task optimization === Multi-task optimization focuses on solving optimizing the whole process. The paradigm has been inspired by the well-established concepts of transfer learning and multi-task learning in predictive analytics. The key motivation behind multi-task optimization is that if optimization tasks are related to each other in terms of their optimal solutions or the general characteristics of their function landscapes, the search progress can be transferred to substantially accelerate the search on the other. The success of the paradigm is not necessarily limited to one-way knowledge transfers from simpler to more complex tasks. In practice an attempt is to intentionally solve a more difficult task that may unintentionally solve several smaller problems. There is a direct relationship between multitask optimization and multi-objective optimization. In some cases, the simultaneous training of seemingly related tasks may hinder performance compared to single-task models. Commonly, MTL models employ task-specific modules on top of a joint feature representation obtained using a shared module. Since this joint representation must capture useful features across all tasks, MTL may hinder individual task performance if the different tasks seek conflicting representation, i.e., the gradients of different tasks point to opposing directions or differ significantly in magnitude. This phenomenon is commonly referred to as negative transfer. To mitigate this issue, various MTL optimization methods have been proposed. It has been reported that meta-knowledge transfer could help avoid negative transfer.Besides, the per-task gradients are combined into a joint update direction through various aggregation algorithms or heuristics. There are several common approaches for multi-task optimization: Bayesian optimization, evolutionary computation, and approaches based on Game theory. ==== Multi-task Bayesian optimization ==== Multi-task Bayesian optimization is a modern model-based approach that leverages the concept of knowledge transfer to speed up the automatic hyperparameter optimization process of machine learning algorithms. The method builds a multi-task Gaussian process model on the data originating from different searches progressing in tandem. The captured inter-task dependencies are thereafter utilized to better inform the subsequent sampling of candidate solutions in respective search spaces. ==== Evolutionary multi-tasking ==== Evolutionary multi-tasking has been explored as a means of exploiting the implicit parallelism of population-based search algorithms to simultaneously progress multiple distinct optimization tasks. By mapping all task
Airborne Networking
An Airborne Network (AN) is the infrastructure owned by the United States Air Force that provides communication transport services through at least one node that is on a platform capable of flight. == Background == === Definition === The intent of the US Air Force's Airborne Network is to expand the Global Information Grid (GIG) to connect the three major domains of warfare: Air, Space, and Terrestrial. The Transformational Satellite Communications System network currently provides connectivity for all communication through space assets. The Combat Information Transport System and Theater Deployable Communications provide terrestrial connectivity for theatre based operations. The Airborne Network is engineered to utilize all airborne assets to connect with space and surface networks building a seamless communications platform across all domains. === Capabilities === The capabilities identified by this type of system are vastly beyond that of our current military. This system will enable the Air Force to provide a transportable network, flexible enough to communicate with any air, space, or ground asset in the area. The network will provide a beyond line-of-sight (LoS) communications infrastructure that can be packed up and moved in and out of the designated battlespace, enabling the military to have a reliable and secure communications network that extends globally. The network is designed to be flexible enough to provide the right communication and network packages for a specific region, mission, or technology. Operationally, The AN is designed to be self-forming, self-organizing, and self-generating, with nodes joining and leaving the network as they enter and exit a specific region. The network consists of dedicated tactical links, wideband air-to-air links, and ad hoc networks constructed by the Joint Tactical Radio System (JTRS) networking services. JTRS is a software-defined radio that will work with many existing military and civilian radios. It includes integrated encryption and Wideband Networking Software to create mobile ad hoc networks. It also provides system performance analysis and fault diagnostics automatically, reducing the demand for human intervention and network maintenance. === Intended Use === The AN was designed as the cornerstone for the new military doctrine known as Network Centric Warfare. This doctrine was developed to use information superiority to equip warfighters with more precise information enabling commanders and shooters to make smarter decisions faster. The AN contributes to Network Centric Warfare by enabling commanders to provide real-time information to warfighters in the air and on the ground. Warfighters can then utilize more information and make more educated decisions about how to act in a particular situation. Once the act has been carried out commanders will have immediate information about the result and can make judgments on how to continue. All-in-all the AN was designed to reduce the time necessary to identify a target, make clear and educated decisions to pull or not to pull the trigger, and assess battle == Topologies == There are four main network topologies that will be deployed and vary based on the placement of backbone and subnet class networks. === Space, Air, Ground Tether === Establishing a direct connection to another aircraft or ground node, via a point-to-point link for nodes within LOS or via a Satellite Communications (SATCOM) link for nodes that are beyond line-of-sight is known as tethering. SATCOM links provide connectivity to a network ground entry point. Strike aircraft that accompany C2 aircraft such as an AWACS are tethered via point-to-point links. Finally, C2 or intelligence, surveillance, and reconnaissnce (ISR) aircraft may connect via a LOS link directly to a network ground entry point. Each of these tethered alternatives works exactly like a hub or switch that has an entry point to a larger network and allows their connected users access to that network. === Flat Ad Hoc === A flat ad hoc topology refers to establishing nonpersistent network connections as needed among AN nodes that are present at a given time. With this network the nodes dynamically “discover” other nodes to which they can interconnect and form the network. The specific interconnections between the nodes are not planned in advance, but are made as opportunities arise. The nodes join and leave the network at will, continually changing connections to neighbor nodes based upon their location and mobility characteristics. === Tiered Ad Hoc === Ad hoc networks can be flat in the sense that all nodes are peers of each other in a single network, as discussed above, or they can dynamically organize themselves into hierarchical tiers such that higher tiers are used to move data between more localized subnets. This network topology can be compared to any conventional deployed network that utilizes routers, switches, and hubs to temporarily connect users. === Persistent Backbone === A network topology characterized by a persistent backbone is established using relatively persistent wideband connections among high-value platforms flying relatively stable orbits. It provides the connectivity between the tactical subnets which are considered edge networks relative to the backbone. This provides concentration points for connectivity to the space backbone as well as to terrestrial networks. This type of network topology is comparable to a conventional permanent network with established data trunks, routers, switches, and hubs to connect users. == Architecture == === Network Management === The platform management system enables operators to manage all on-board network elements. It interfaces and interoperates with the Airborne Network management system to enable operators to manage remote network elements in the airborne network. The network management system monitors the health of the network by passively testing the network for faults and latency. The system will also actively troubleshoot faults with probes to identify and isolate faulty connections, and enables operators to apply network parameters and security changes to all systems based on the status of the network. === Routing/Switching === Routing and switching enables data to be dynamically transmitted over the network to other nodes. Routing protocols must be able to identify nodes transmitted within their own platform and data to be sent to other platforms regardless of the current topology. The routing protocol must also provide seamless roaming by ensuring that no routed packets are lost when a node changes its point of attachment to the network. Maintaining scalability is important in routing as the network is constantly changing. The network must be able to function with numerous levels of platforms, varying numbers of fast moving platforms, and varying amounts of traffic per platform. Routers and switches will use metrics to determine the best paths to take when routing data. The routing protocol utilized for the AN will be an Adaptive Quality of Service routing protocol. === Gateways/Proxies === Gateways and proxies enable the connection numerous technology types regardless of age to communicate across the IP-based network. Gateways and proxies are essential in the operation of this network because so many different technologies are used to communicate in each domain. These systems will facilitate the transition of the legacy on-board infrastructure, transmission systems, tactical data link systems, and user applications to the objective airborne network systems. Therefore, they are only temporary until all platforms use a standardized IP radio for transmission. === Performance Enhancing Proxies === Performance Enhancing Proxies improve the performance of user applications running across the Airborne Network by countering wireless network impairments, such as limited bandwidth, long delays, high loss rates, and disruptions in network connections. Proxy systems are implemented between the user application and the network and can be used to improve performance at the application and transport functional layers of the OSI model. Some techniques that can be employed include: Compression: Data compression or header compression can be used to minimize the number of bits sent over the network. Data bundling: Smaller data packets can be combined (bundled) into a single large packet for transmission over the network. Caching: A local cache can be used to save and provide data objects that are requested multiple times, reducing transmissions over the network (and improving response times). Store and forward: Message queuing can be used to ensure message delivery to users who become disconnected from the network or are unable to connect to the network for a period of time. Once the platform connects, the stored messages are sent. Pipelining: Rather than opening several separate network connections pipelining can be used to share a single networ