AppBlock is a software tool for managing screen time that limits access to selected mobile applications and websites. Developed by the Czech studio MobileSoft, it is distributed for Android and iOS devices as well as through browser extensions for Google Chrome, Microsoft Edge and Brave, and as desktop solutions. The application is used primarily to restrict time spent on social media and similar distracting services while working and studying. By 2025, the application reported 700,000 monthly active users, with the domestic Czech market accounting for less than one percent of its total user base and revenue. == History == === Origins === AppBlock was created by the Czech software studio MobileSoft, based in Hradec Králové. The studio was founded in 2012 by Miroslav Novosvětský, who remains the sole owner. The idea for the application arose from the use of browser-based website blockers on desktop computers. AppBlock was conceived as a way to reduce the time spent on mobile devices. === Early releases === In its early phase, AppBlock was available only for phones running on Android. Early versions allowed users to limit access to selected applications and websites during specified periods. From the outset, the application was distributed internationally rather than only within the Czech market, and early coverage reported a multi-million number of downloads worldwide. === Expansion of functionality === Over time, AppBlock has expanded beyond basic application blocking to include additional functions related to limiting procrastination and managing attention. The development of AppBlock accelerated during the COVID-19 pandemic. Following a reduction in external client orders, the studio reallocated resources from contract development to the application. Increased digital content consumption during lockdowns contributed to a rise in the application's usage and revenue. As the application developed, it became the company's product with the largest user base. Novosvětský described an increase in downloads over a twelve-month period, which he linked in part to the company's activities abroad, including participation in events focused on mobile marketing in the United States. These activities were an important factor in the further development of AppBlock. === Internationalization and market expansion === Within roughly the first eight years of the company's existence, MobileSoft became active both in the domestic Czech market and in the United States, supported among other things by participation in the CzechAccelerator program, which is intended to help Czech firms enter foreign markets. In mid-August 2021 the developers launched a version for iOS, which soon began to attract paying users. The expansion to iOS was accompanied by plans for cooperation with the Procrastination.com platform, intended to complement the blocking functions with educational content related to digital media use, sleep and work habits. By 2025, AppBlock was localised into 15 languages, with the largest share of users in the United States, the United Kingdom, Germany, and France, with recent growth in Brazil, and usage extending across several continents. AppBlock has reached more than 10 million installations. In the same period its creators announced plans to refine existing functions and to expand support beyond mobile phones to desktop use, including through support for additional web browsers. == Features == === Supported platforms === AppBlock is distributed as a mobile application for Android and iOS users through Google Play and the Apple App Store. Browser extensions for desktop systems are available for Google Chrome, Microsoft Edge and Brave. === Functionality === AppBlock's core function is to restrict access to selected applications and websites. The mobile application shows a list of installed apps and lets the user select which ones to block. It also includes tools to block specific websites and, on iOS, to block certain phrases entered in the Safari browser. AppBlock can mute notifications from selected applications, so alerts from those apps do not appear while blocking is active. In addition to choosing which apps or content to block, the software also offers an allowlist mode, where only selected applications remain accessible and all others are blocked. Blocking rules are organized into configurable schedules, called profiles. Users can create profiles that define time periods when selected apps and websites are unavailable. Newer versions also allow profiles to be activated automatically based on the time of day, days of the week, the device's location, or connection to specific Wi-Fi networks. The iOS version lets users set limits on how often or how long certain apps can be used before they are blocked, and it can track and restrict screen time for individual apps. In addition to these recurring rules, AppBlock includes a Quick Block feature that temporarily blocks selected apps and websites with a single action, without requiring a separate long-term schedule. Strict Mode is an optional setting that limits the ability to change blocking once it is active. For a specified period, it prevents editing AppBlock's rules and can be configured to stop the app from being uninstalled during that time. While Strict Mode is enabled, users cannot modify or disable the restrictions they have set. Deactivation requires specific verification steps, such as connecting the device to a charger or obtaining approval from a designated contact person. The mobile application also includes statistical and reporting features. In addition to blocking, AppBlock lets users view statistics and data about their use of applications and websites, including screen-time summaries and focus sessions that silence notifications and enforce blocking during defined work or study periods. Browser extensions for desktop environments apply AppBlock's website-blocking functions on Windows and macOS systems through supported web browsers. == Business model == AppBlock uses a freemium revenue model. The basic version of the application is available free of charge and allows blocking of up to three applications at the same time. The premium version removes this limit and adds further configuration options. In 2020, the application shifted from a one-time payment structure to a subscription model. By 2021, AppBlock had more than seven thousand paying users and annual revenue of about four million Czech crowns. By 2025, annual revenue reached approximately 4 million US dollars (80 million CZK) before taxes and platform fees, with roughly 20 percent of active users subscribing to the paid version. == Usage == AppBlock limits access to selected applications and websites in order to reduce smartphone overuse and digital distraction. It is used to block social media, games and other services considered addictive, with the aim of reducing frequent checking of mobile devices and creating time intervals in which these services are unavailable. Reported use cases of AppBlock cover work, students, parents, ADHD, mental health, well-being and business. The application is used both by individual users and within workplace initiatives in which employees install it to reduce digital distractions during working hours.
Art Recognition
Art Recognition is a Swiss technology company headquartered in Adliswil, within the Zurich metropolitan area, Switzerland. Art Recognition specializes in the application of artificial intelligence (AI) for art authentication and the detection of art forgeries. == Overview == Art Recognition was established in 2019 by Dr. Carina Popovici and Christiane Hoppe-Oehl. Art Recognition employs a combination of machine learning techniques, computer vision algorithms, and deep neural networks to assess the authenticity of artworks. The company's technology undergoes a process of data collection, dataset preparation, and training. === Academic partnerships and grants === Art Recognition has established a relationship with Innosuisse, a Swiss innovation agency, to expand its research and development initiatives. It has also formed a strategic collaboration with Nils Büttner, an art historian and professor at the State Academy of Fine Arts Stuttgart (ABK Stuttgart). === Notable developments === In May 2024, Art Recognition played a key role in identifying counterfeit artworks, including alleged Monets and Renoirs, being sold on eBay. Germann Auction in November 2024 became the first auction house to successfully conduct a sale of artwork authenticated entirely by artificial intelligence. As of January 2025, Art Recognition has appointed art crime expert and Pulitzer Prize finalist Noah Charney as an advisor. === Recognition and debates === The company was featured on the front page of The Wall Street Journal for its involvement in the authentication case of the Flaget Madonna, believed to have been partly painted by Raphael. A broadcast by the Swiss public television SRF covered how the algorithm can be used to detect art forgeries with high accuracy. The technology developed by Art Recognition has been recognized for its role in providing a technology-based art authentication solution, compared to traditional methods. == Controversial cases == Art Recognition's AI algorithm has been applied to several high-profile and controversial artworks, sparking significant interest and debate in the art world. Samson and Delilah at the National Gallery in London: The National Gallery's "Samson and Delilah", traditionally attributed to the artist Rubens, has also been examined using Art Recognition's AI, which has assessed the painting as non-authentic. De Brecy Tondo Madonna. A research team from Bradford University and the University of Nottingham initially attributed the painting to Raphael, employing an AI face recognition software, while the AI developed at Art Recognition returned a negative result. The Bradford group's AI was trained on 49 images, whereas Art Recognition employed a larger dataset of over 100 images. Lucian Freud Painting Controversy: Featured in The New Yorker, a painting attributed to Lucian Freud became a subject of dispute. Art Recognition's AI analysis played a big role in examining the painting's authenticity. Titian at Kunsthaus Zürich: A painting attributed to Titian, housed at Kunsthaus Zürich, has been a topic of debate among art experts. The application of Art Recognition's technology offered a new perspective. Following this debate, Kunsthaus Zürich has announced plans to initiate a comprehensive project aimed at resolving the authenticity questions surrounding the painting. Art Recognition has contributed to the authentication debate surrounding The Polish Rider, a painting traditionally attributed to Rembrandt but subject to scholarly debate.
Social media use in the financial services sector
Social media in the financial services sector refers to the use of social media by the financial services sector to promote and distribute financial services. Social media is used in various aspects of the financial industry including customer service, marketing, and product development. It has enabled financial institutions to extend their reach through direct and real-time communication with customers, fostering more personal connections. It also allows individuals to talk to other individuals creating lending and trading via social groups as well as developing new financial services by fintech startup companies. In terms of marketing, social media is utilized by both traditional financial companies as well as disruptive fintech companies such as peer-to-peer lending (P2P) companies. The financial industry has used information technology since its inception in the 1960s and social media fits in with this ongoing development. Larger, traditional financial firms have integrated social media into their marketing strategies. Companies in the financial sector are subject to strict regulations that include how they use social media. In the United States, the Financial Industry Regulatory Authority (FINRA) is a key regulator that sets rules how financial firms can interact with consumers. This includes ensuring that social media posts follow financial advertising rules, such as being fair and balanced and not providing misleading information, and that financial advice is not provided by unqualified personnel, such as influencers. == History == In 2003, at the beginning of social media development, MySpace was founded as a "social networking service." It allowed people to create a profile, connect with other people, and post videos, pictures, and songs. As MySpace grew in popularity, it attracted interest from companies wishing to promote their brands on the social platform. They were joined by Facebook and in 2010 by Instagram. Financial service firms were initially slow to adapt to promotion via social media but soon joined other big firms after they saw the success other industries had in engaging with younger people. == Uses == === Branding === While companies are able to connect with more people remotely through providing online financial services, their branding strategy has shifted from customized to standardized. Prior to the outbreak of technology, most banks used customized branding where they targeted only customers in their regions. Businesses can now use technology to operate beyond their geographic location and maintain a consistent image across multiple countries with standardized branding. By being able to extend a consistent brand reputation across a wider geographic location, financial services companies can take advantage of economies of scale in advertising cost, lower administrative complexity, lower entry into new markets, and improved cross-border learning within the company. === Customer engagement === Online banking reduced face-to-face interaction between customers and their banks. Most banking transactions can now be conducted online or through mobile devices, rather than at a local branch with a teller. Social media provides a channel for firms to maintain personal contact with customers, replicating some of the interaction that was previously available at local branches. For example, a bank's Facebook page may feature an employee profile describing their job duties, which serves to present a more human face for larger institutions. === Lending === Social media is a core marketing channel for online peer-to-peer lending as well as small business lenders. Since these companies operate exclusively online, it makes sense for them to market online through social media channels. They are able to grow and find new lenders and buyers by utilizing social networks. === Trading === Social trading is an alternative way of analyzing financial data by looking at what other traders are doing and comparing, copying and discussing their techniques and strategies. Prior to the advent of social trading, investors and traders were relying on fundamental or technical analysis to form their investment decisions. Using social trading investors and traders could integrate into their investment decision-process social indicators from trading data-feeds of other traders. Investors also use platform like Reddit, Signal messaging or WeChat to create social communities to discuss investments and finance. In some cases they use this to join together using meme stocks to move financial markets, such as the 2021 GameStop short squeeze incident. They can also use social groups to launch and promote new products such as cryptocurrencies. Investing application like WeBull incorporate a forum style messaging system on each stock that is available for trading. Financial brokers such as Fidelity Investments, Interactive Brokers, and E-Trade have moved to incorporate community features in their investment apps. == Regulations == The use of social media by investors and financial services professionals for business purposes is subject to regulatory oversight, in the United States this is done primarily by the Financial Industry Regulatory Authority (FINRA). FINRA's rules, designed to protect investors from misleading information in all communications and this also applies to social media communications. This includes ensuring that social media posts follow financial advertising rules, such as being fair and balanced and not providing misleading information, and that advice is not provided by unqualified personnel, such as influencers and bank staff acting in a personal capacity. Financial firms have to maintain books and records of all interaction with customers and this includes social media. == New products and services == Social media has created entirely new products for the financial services sector, revolutionizing products and developing new industries through the merging of social technology and financial services. Fintech startups use social media to promote products to get them established. Several developing nations have used social media to leapfrog traditional financial technology; for example, WeChat Pay, which developed from the Chinese WeChat social media platform, became a major payment system in China within a few years. In 2015, according to consulting firm Accenture, 390 million people in China had registered to use mobile banking. This figure is more than the population of the United States. In the United States, the fintech company Venmo combines technology and financial services on a social platform. Other financial technology companies that have used social media to develop or promote financial products include: Lending Club – One of the first peer-to-peer lending businesses OnDeck Capital – A US online-only lending business Funding Circle – A UK-based online lending company Wise – A global online money transfers company Kabbage – A US online unsecured loan company later acquired by American Express Avant – A US online unsecured loan company Zopa – A UK online neobank providing peer-to-peer lending == Risks == === Reputational damage === Due to the real-time nature of social media, financial services companies can be impacted by potential reputational issues. Any negative experience by customers can easily be shared online and could become a viral phenomenon, those comments could likely have a detrimental effect on the company’s stock price and reputation. On the other hand, any positive experience a customer has can also be shared online. However, positive experiences are much less likely to become viral. === Scams === The nature of social media makes it easy to target individuals without being seen by the wider community, this allows scammers to target individuals. Example include romance scams such as the pig butchering scam where an individual is tricked to transfer funds or assets to the scammer over social media making it hard for law enforcement to track them or recover funds. === Customer privacy === Customer privacy is important for the financial services industry. It is critical that customer information such as a bank account numbers and other personal information is kept private. However, this information can be leaked if for example, a customer is unhappy with a bank’s service, they may tweet at the bank expressing their frustrations and include their name and account number.
Data product
In data management and product management, a data product is a reusable, active, and standardized data asset designed to deliver measurable value to its users, whether internal or external, by applying the rigorous principles of product thinking and management. It comprises one or more data artifacts (e.g., datasets, models, pipelines) and is enriched with metadata, including governance policies, data quality rules, data contracts, and, where applicable, a software bill of materials (SBOM) to document its dependencies and components. Ownership of a data product is aligned to a specific domain or use case, ensuring accountability, stewardship, and its continuous evolution throughout its lifecycle. Adhering to the FAIR principles – findable, accessible, interoperable, and reusable – a data product is designed to be discoverable, scalable, reusable, and aligned with both business and regulatory standards, driving innovation and efficiency in modern data ecosystems. == History == In 2012, DJ Patil proposed the first documented definition: a data product is a product that facilitates an end goal through the use of data. In 2019, Zhamak Dehghani introduced Data Mesh, with a strong focus on domain-oriented data products. Later, in 2020, she solidifies Data Mesh around four principles, one being Data as a Product, in which she defines Data Product as the node on the mesh that encapsulates three structural components required for its function, providing access to the domain's analytical data as a product. In 2024, Andrea Gioia published one of the first books specifically on data products post Data Mesh announcement. In his book, Gioia defines the concept of pure data product. In 2025, during the Data Day Texas conference, Jean-Georges Perrin and a collective of product managers and data engineers got together to craft the current definition and make it available to the public domain. In July 2025, Bitol, a project of The Linux Foundation, released and early version of the Open Data Product Standard (ODPS) aiming at normalizing data products
Intranet
An intranet is a computer network for sharing information, easier communication, collaboration tools, operational systems, and other computing services within an organization, usually to the exclusion of access by outsiders. The term is used in contrast to public networks, such as the Internet, but uses the same technology based on the Internet protocol suite. An organization-wide intranet can constitute a focal point of internal communication and collaboration, and provide a single starting point to access internal and external resources. In its simplest form, an intranet is established with the technologies for local area networks (LANs) and wide area networks (WANs). Many modern intranets have search engines, user profiles, blogs, mobile apps with notifications, and events planning within their infrastructure. An intranet is sometimes contrasted to an extranet. While an intranet is generally restricted to employees of the organization, extranets may also be accessed by customers, suppliers, or other approved parties. Extranets extend a private network onto the Internet with special provisions for authentication, authorization and accounting (AAA protocol). == Uses == Intranets are increasingly being used to deliver tools, such as for collaboration (to facilitate working in groups and teleconferencing) or corporate directories, sales and customer relationship management, or project management. Intranets are also used as corporate culture-change platforms. For example, a large number of employees using an intranet forum application to host a discussion about key issues could come up with new ideas related to management, productivity, quality, and other corporate issues. In large intranets, website traffic is often similar to public website traffic and can be better understood by using web metrics software to track overall activity. User surveys also improve intranet website effectiveness. Larger businesses allow users within their intranet to access public internet through firewall servers. They have the ability to screen incoming and outgoing messages, keeping security intact. When part of an intranet is made accessible to customers and others outside the business, it becomes part of an extranet. Businesses can send private messages through the public network using special encryption/decryption and other security safeguards to connect one part of their intranet to another. Intranet user-experience, editorial, and technology teams work together to produce in-house sites. Most commonly, intranets are managed by the communications, HR or CIO departments of large organizations, or some combination of these. Because of the scope and variety of content and the number of system interfaces, the intranets of many organizations are much more complex than their respective public websites. Intranets and the use of intranets are growing rapidly. According to the Intranet Design Annual 2007 from Nielsen Norman Group, the number of pages on participants' intranets averaged 200,000 over the years 2001 to 2003 and has grown to an average of 6 million pages over 2005–2007. == Benefits == Intranets can help users locate and view information faster and use applications relevant to their roles and responsibilities. With a web browser interface, users can access data held in any database the organization wants to make available at any time and — subject to security provisions — from anywhere within company workstations, increasing employees' ability to perform their jobs faster, more accurately, and with confidence that they have the right information. It also helps improve services provided to users. Using hypermedia and Web technology, Web publishing allows for the maintenance of and easy access to cumbersome corporate knowledge, such as employee manuals, benefits documents, company policies, business standards, news feeds, and even training, all of which can be accessed throughout a company using common Internet standards (Acrobat files, Flash files, CGI applications). Because each business unit can update the online copy of a document, the most recent version is usually available to employees using the intranet. Intranets are also used as a platform for developing and deploying applications to support business operations and decisions across the internetworked enterprise. Information is easily accessible to all authorised users, enabling collaboration. Being able to communicate in real-time through integrated third-party tools, such as an instant messenger, promotes the sharing of ideas and removes blockages to communication to help boost a business's productivity. Intranets can serve as powerful tools for communicating (such as through chat, email and/or blogs) within a given organization about vertically strategic initiatives that have a global reach throughout said organization. The type of information that can easily be conveyed is the purpose of the initiative and what it is aiming to achieve, who is driving it, results achieved to date, and whom to speak to for more information. By providing this information on the intranet, staff can keep up-to-date with the strategic focus of their organization. For example, when Nestlé had a number of food processing plants in Scandinavia, their central support system had to deal with a number of queries every day. When Nestlé decided to invest in an intranet, they quickly realized the savings. Gerry McGovern says that the savings from the reduction in query calls was substantially greater than the investment in the intranet. Users can view information and data via a web browser rather than maintaining physical documents such as procedure manuals, internal phone list and requisition forms. This can potentially save the business money on printing, duplicating documents, and the environment, as well as document maintenance overhead. For example, the HRM company PeopleSoft "derived significant cost savings by shifting HR processes to the intranet". McGovern goes on to say the manual cost of enrolling in benefits was found to be US$109.48 per enrollment. "Shifting this process to the intranet reduced the cost per enrollment to $21.79; a saving of 80 percent". Another company that saved money on expense reports was Cisco. "In 1996, Cisco processed 54,000 reports and the amount of dollars processed was USD19 million". Many companies dictate computer specifications which, in turn, may allow Intranet developers to write applications that only have to work on one browser such that there are no cross-browser compatibility issues. Being able to specifically address one's "viewer" is a great advantage. Since intranets are user-specific (requiring database/network authentication prior to access), users know exactly who they are interfacing with and can personalize their intranet based on role (job title, department) or individual ("Congratulations Jane, on your 3rd year with our company!"). Since "involvement in decision making" is one of the main drivers of employee engagement, offering tools (like forums or surveys) that foster peer-to-peer collaboration and employee participation can make employees feel more valued and involved. == Planning and creation == Most organizations devote considerable resources into the planning and implementation of their intranet as it is of strategic importance to the organization's success. Some of the planning would include topics such as determining the purpose and goals of the intranet, identifying persons or departments responsible for implementation and management and devising functional plans, page layouts and designs. The appropriate staff would also ensure that implementation schedules and phase-out of existing systems were organized, while defining and implementing security of the intranet and ensuring it lies within legal boundaries and other constraints. In order to produce a high-value end product, systems planners should determine the level of interactivity (e.g. wikis, on-line forms) desired. Planners may also consider whether the input of new data and updating of existing data is to be centrally controlled or devolve. These decisions sit alongside to the hardware and software considerations (like content management systems), participation issues (like good taste, harassment, confidentiality), and features to be supported. Intranets are often static sites; they are a shared drive, serving up centrally stored documents alongside internal articles or communications (often one-way communication). By leveraging firms which specialise in 'social' intranets, organisations are beginning to think of how their intranets can become a 'communication hub' for their entire team. The actual implementation would include steps such as securing senior management support and funding, conducting a business requirement analysis and identifying users' information needs. From the technical perspective, there would need to be a coordinated installation of the web server and user access netw
Actor-critic algorithm
The actor-critic algorithm (AC) is a family of reinforcement learning (RL) algorithms that combine policy-based RL algorithms such as policy gradient methods, and value-based RL algorithms such as value iteration, Q-learning, SARSA, and TD learning. An AC algorithm consists of two main components: an "actor" that determines which actions to take according to a policy function, and a "critic" that evaluates those actions according to a value function. Some AC algorithms are on-policy, some are off-policy. Some apply to either continuous or discrete action spaces. Some work in both cases. == Overview == The actor-critic methods can be understood as an improvement over pure policy gradient methods like REINFORCE via introducing a baseline. === Actor === The actor uses a policy function π ( a | s ) {\displaystyle \pi (a|s)} , while the critic estimates either the value function V ( s ) {\displaystyle V(s)} , the action-value Q-function Q ( s , a ) , {\displaystyle Q(s,a),} the advantage function A ( s , a ) {\displaystyle A(s,a)} , or any combination thereof. The actor is a parameterized function π θ {\displaystyle \pi _{\theta }} , where θ {\displaystyle \theta } are the parameters of the actor. The actor takes as argument the state of the environment s {\displaystyle s} and produces a probability distribution π θ ( ⋅ | s ) {\displaystyle \pi _{\theta }(\cdot |s)} . If the action space is discrete, then ∑ a π θ ( a | s ) = 1 {\displaystyle \sum _{a}\pi _{\theta }(a|s)=1} . If the action space is continuous, then ∫ a π θ ( a | s ) d a = 1 {\displaystyle \int _{a}\pi _{\theta }(a|s)da=1} . The goal of policy optimization is to improve the actor. That is, to find some θ {\displaystyle \theta } that maximizes the expected episodic reward J ( θ ) {\displaystyle J(\theta )} : J ( θ ) = E π θ [ ∑ t = 0 T γ t r t ] {\displaystyle J(\theta )=\mathbb {E} _{\pi _{\theta }}\left[\sum _{t=0}^{T}\gamma ^{t}r_{t}\right]} where γ {\displaystyle \gamma } is the discount factor, r t {\displaystyle r_{t}} is the reward at step t {\displaystyle t} , and T {\displaystyle T} is the time-horizon (which can be infinite). The goal of policy gradient method is to optimize J ( θ ) {\displaystyle J(\theta )} by gradient ascent on the policy gradient ∇ J ( θ ) {\displaystyle \nabla J(\theta )} . As detailed on the policy gradient method page, there are many unbiased estimators of the policy gradient: ∇ θ J ( θ ) = E π θ [ ∑ 0 ≤ j ≤ T ∇ θ ln π θ ( A j | S j ) ⋅ Ψ j | S 0 = s 0 ] {\displaystyle \nabla _{\theta }J(\theta )=\mathbb {E} _{\pi _{\theta }}\left[\sum _{0\leq j\leq T}\nabla _{\theta }\ln \pi _{\theta }(A_{j}|S_{j})\cdot \Psi _{j}{\Big |}S_{0}=s_{0}\right]} where Ψ j {\textstyle \Psi _{j}} is a linear sum of the following: ∑ 0 ≤ i ≤ T ( γ i R i ) {\textstyle \sum _{0\leq i\leq T}(\gamma ^{i}R_{i})} . γ j ∑ j ≤ i ≤ T ( γ i − j R i ) {\textstyle \gamma ^{j}\sum _{j\leq i\leq T}(\gamma ^{i-j}R_{i})} : the REINFORCE algorithm. γ j ∑ j ≤ i ≤ T ( γ i − j R i ) − b ( S j ) {\textstyle \gamma ^{j}\sum _{j\leq i\leq T}(\gamma ^{i-j}R_{i})-b(S_{j})} : the REINFORCE with baseline algorithm. Here b {\displaystyle b} is an arbitrary function. γ j ( R j + γ V π θ ( S j + 1 ) − V π θ ( S j ) ) {\textstyle \gamma ^{j}\left(R_{j}+\gamma V^{\pi _{\theta }}(S_{j+1})-V^{\pi _{\theta }}(S_{j})\right)} : TD(1) learning. γ j Q π θ ( S j , A j ) {\textstyle \gamma ^{j}Q^{\pi _{\theta }}(S_{j},A_{j})} . γ j A π θ ( S j , A j ) {\textstyle \gamma ^{j}A^{\pi _{\theta }}(S_{j},A_{j})} : Advantage Actor-Critic (A2C). γ j ( R j + γ R j + 1 + γ 2 V π θ ( S j + 2 ) − V π θ ( S j ) ) {\textstyle \gamma ^{j}\left(R_{j}+\gamma R_{j+1}+\gamma ^{2}V^{\pi _{\theta }}(S_{j+2})-V^{\pi _{\theta }}(S_{j})\right)} : TD(2) learning. γ j ( ∑ k = 0 n − 1 γ k R j + k + γ n V π θ ( S j + n ) − V π θ ( S j ) ) {\textstyle \gamma ^{j}\left(\sum _{k=0}^{n-1}\gamma ^{k}R_{j+k}+\gamma ^{n}V^{\pi _{\theta }}(S_{j+n})-V^{\pi _{\theta }}(S_{j})\right)} : TD(n) learning. γ j ∑ n = 1 ∞ λ n − 1 1 − λ ⋅ ( ∑ k = 0 n − 1 γ k R j + k + γ n V π θ ( S j + n ) − V π θ ( S j ) ) {\textstyle \gamma ^{j}\sum _{n=1}^{\infty }{\frac {\lambda ^{n-1}}{1-\lambda }}\cdot \left(\sum _{k=0}^{n-1}\gamma ^{k}R_{j+k}+\gamma ^{n}V^{\pi _{\theta }}(S_{j+n})-V^{\pi _{\theta }}(S_{j})\right)} : TD(λ) learning, also known as GAE (generalized advantage estimate). This is obtained by an exponentially decaying sum of the TD(n) learning terms. === Critic === In the unbiased estimators given above, certain functions such as V π θ , Q π θ , A π θ {\displaystyle V^{\pi _{\theta }},Q^{\pi _{\theta }},A^{\pi _{\theta }}} appear. These are approximated by the critic. Since these functions all depend on the actor, the critic must learn alongside the actor. The critic is learned by value-based RL algorithms. For example, if the critic is estimating the state-value function V π θ ( s ) {\displaystyle V^{\pi _{\theta }}(s)} , then it can be learned by any value function approximation method. Let the critic be a function approximator V ϕ ( s ) {\displaystyle V_{\phi }(s)} with parameters ϕ {\displaystyle \phi } . The simplest example is TD(1) learning, which trains the critic to minimize the TD(1) error: δ i = R i + γ V ϕ ( S i + 1 ) − V ϕ ( S i ) {\displaystyle \delta _{i}=R_{i}+\gamma V_{\phi }(S_{i+1})-V_{\phi }(S_{i})} The critic parameters are updated by gradient descent on the squared TD error: ϕ ← ϕ − α ∇ ϕ ( δ i ) 2 = ϕ + α δ i ∇ ϕ V ϕ ( S i ) {\displaystyle \phi \leftarrow \phi -\alpha \nabla _{\phi }(\delta _{i})^{2}=\phi +\alpha \delta _{i}\nabla _{\phi }V_{\phi }(S_{i})} where α {\displaystyle \alpha } is the learning rate. Note that the gradient is taken with respect to the ϕ {\displaystyle \phi } in V ϕ ( S i ) {\displaystyle V_{\phi }(S_{i})} only, since the ϕ {\displaystyle \phi } in γ V ϕ ( S i + 1 ) {\displaystyle \gamma V_{\phi }(S_{i+1})} constitutes a moving target, and the gradient is not taken with respect to that. This is a common source of error in implementations that use automatic differentiation, and requires "stopping the gradient" at that point. Similarly, if the critic is estimating the action-value function Q π θ {\displaystyle Q^{\pi _{\theta }}} , then it can be learned by Q-learning or SARSA. In SARSA, the critic maintains an estimate of the Q-function, parameterized by ϕ {\displaystyle \phi } , denoted as Q ϕ ( s , a ) {\displaystyle Q_{\phi }(s,a)} . The temporal difference error is then calculated as δ i = R i + γ Q θ ( S i + 1 , A i + 1 ) − Q θ ( S i , A i ) {\displaystyle \delta _{i}=R_{i}+\gamma Q_{\theta }(S_{i+1},A_{i+1})-Q_{\theta }(S_{i},A_{i})} . The critic is then updated by θ ← θ + α δ i ∇ θ Q θ ( S i , A i ) {\displaystyle \theta \leftarrow \theta +\alpha \delta _{i}\nabla _{\theta }Q_{\theta }(S_{i},A_{i})} The advantage critic can be trained by training both a Q-function Q ϕ ( s , a ) {\displaystyle Q_{\phi }(s,a)} and a state-value function V ϕ ( s ) {\displaystyle V_{\phi }(s)} , then let A ϕ ( s , a ) = Q ϕ ( s , a ) − V ϕ ( s ) {\displaystyle A_{\phi }(s,a)=Q_{\phi }(s,a)-V_{\phi }(s)} . Although, it is more common to train just a state-value function V ϕ ( s ) {\displaystyle V_{\phi }(s)} , then estimate the advantage by A ϕ ( S i , A i ) ≈ ∑ j ∈ 0 : n − 1 γ j R i + j + γ n V ϕ ( S i + n ) − V ϕ ( S i ) {\displaystyle A_{\phi }(S_{i},A_{i})\approx \sum _{j\in 0:n-1}\gamma ^{j}R_{i+j}+\gamma ^{n}V_{\phi }(S_{i+n})-V_{\phi }(S_{i})} Here, n {\displaystyle n} is a positive integer. The higher n {\displaystyle n} is, the more lower is the bias in the advantage estimation, but at the price of higher variance. The Generalized Advantage Estimation (GAE) introduces a hyperparameter λ {\displaystyle \lambda } that smoothly interpolates between Monte Carlo returns ( λ = 1 {\displaystyle \lambda =1} , high variance, no bias) and 1-step TD learning ( λ = 0 {\displaystyle \lambda =0} , low variance, high bias). This hyperparameter can be adjusted to pick the optimal bias-variance trade-off in advantage estimation. It uses an exponentially decaying average of n-step returns with λ {\displaystyle \lambda } being the decay strength. == Variants == Asynchronous Advantage Actor-Critic (A3C): Parallel and asynchronous version of A2C. Soft Actor-Critic (SAC): Incorporates entropy maximization for improved exploration. Deep Deterministic Policy Gradient (DDPG): Specialized for continuous action spaces.
Social influence bias
The social influence bias is an asymmetric herding effect on online social media platforms which makes users overcompensate for negative ratings but amplify positive ones. Driven by the desire to be accepted within a specific group, it surrounds the idea that people alter certain behaviors to be like those of the people within a group. Therefore, it is a subgroup term for various types of cognitive biases. Some social influence bias types include the bandwagon effect, authority bias, groupthinking effect, social comparison bias, social media bias and more. Understanding these biases helps us understand the term overall. However, the composition of the term "social influence bias" requires critical examination to understand the way that it affects individuals' and groups' lives. The term "influence" has 2 different types of stigma. For one, it surrounds the idea that people show their true inner selves when "under the influence". On the other end, it also proposes the idea that people are not their own selves when "under the influence". These tend to be constructions made by people, which also tend to fit the situation based on their own perspectives. So, even in social terms, it requires both sides to be examined to understand whether we truly are affected by context, or we remain to be and behave in terms of our own selves. The term "influence" doesn't necessarily say that there lies greater strength in our inner self's desires and decisions, nor does it say that external factors have the greater power. In a similar manner, both social and non-social judgments are to be associated with anxiety, but the same can't necessarily be said in the case of social conformity. So, the gray areas within this topic beg the question, "What does social influence bias say about us, and does it affect us all in the same way?" == Social media bias == Media bias is reflected in search systems in social media. Kulshrestha and her team found through research in 2018 that the top-ranked results returned by these search engines can influence users' perceptions when they conduct searches for events or people, which is particularly reflected in political bias and polarizing topics. Fueled by confirmation bias, online echo chambers allow users to be steeped within their own ideology. Because social media is tailored to your interests and your selected friends, it is an easy outlet for political echo chambers. Social media bias is also reflected in hostile media effect. Social media has a place in disseminating news in modern society, where viewers are exposed to other people's comments while reading news articles. In their 2020 study, Gearhart and her team showed that viewers' perceptions of bias increased and perceptions of credibility decreased after seeing comments with which they held different opinions. == In research context == In observational data, how social influence affects collected judgment is challenging to fully understand. Positive social influence can accumulate and result in a rating bubble, while negative social influence is neutralized by crowd correction. This phenomenon was first described in a paper written by Lev Muchnik, Sinan Aral and Sean J. Taylor in 2014, then the question was revisited by Cicognani et al., whose experiment reinforced Munchnik's and his co-authors' results. == Relevance == Online customer reviews are trusted sources of information in various contexts such as online marketplaces, dining, accommodation, movies, or digital products. However, these online ratings are not immune to herd behavior, which means that subsequent reviews are not independent from each other. As on many such sites, preceding opinions are visible to a new reviewer, he or she can be heavily influenced by the antecedent evaluations in his or her decision about the certain product, service or online content. This form of herding behavior inspired Muchnik, Aral and Taylor to conduct their experiment on influence in social contexts. == Experimental design == Muchnik, Aral, and Taylor designed a large-scale randomized experiment to measure social influence on user reviews. The experiment was conducted on social news aggregation website like Reddit. The study lasted for 5 months, the authors randomly assigned 101 281 comments to one of the following treatment groups: up-treated (4049), down-treated (1942), or control (the proportions reflect the observed ratio of up-and down-votes. Comments which fell to the first group were given an up-vote upon the creation of the comment, the second group got a down-vote upon creation, the comments in the control group remained untouched. A vote is equivalent to a single rating (+1 or -1). As other users are unable to trace a user’s votes, they were unaware of the experiment. Due to randomization, comments in the control and the treatment group were not different in terms of expected rating. The treated comments were viewed more than 10 million times and rated 308 515 times by successive users. == Results == The up-vote treatment increased the probability of up-voting by the first viewer by 32% over the control group, while the probability of down-voting did not change compared to the control group, which means that users did not correct the random positive rating. The upward bias remained inplace for the observed 5-month period. The accumulating herding effect increased the comment’s mean rating by 25% compared to the control group comments. Positively manipulated comments did receive higher ratings at all parts of the distribution, which means that they were also more likely to collect extremely high scores. The negative manipulation created an asymmetric herd effect: although the probability of subsequent down-votes was increased by the negative treatment, the probability of up-voting also grew for these comments. The community performed a correction which neutralized the negative treatment and resulted non-different final mean ratings from the control group. The authors also compared the final mean scores of comments across the most active topic categories on the website. The observed positive herding effect was present in the "politics," "culture and society," and "business" subreddits, but was not applicable for "economics," "IT," "fun," and "general news".- == Implications == The skewed nature of online ratings makes review outcomes different to what it would be without the social influence bias. In a 2009 experiment by Hu, Zhang and Pavlou showed that the distribution of reviews of a certain product made by unconnected individuals is approximately normal, however, the rating of the same product on Amazon followed a J-Shaped distribution with twice as much five-star ratings than others. Cicognani, Figini and Magnani came to similar conclusions after their experiment conducted on a tourism services website: positive preceding ratings influenced raters' behavior more than mediocre ones. Positive crowd correction makes community-based opinions upward-biased.