Probabilistic latent semantic analysis (PLSA), also known as probabilistic latent semantic indexing (PLSI, especially in information retrieval circles) is a statistical technique for the analysis of two-mode and co-occurrence data. In effect, one can derive a low-dimensional representation of the observed variables in terms of their affinity to certain hidden variables, just as in latent semantic analysis, from which PLSA evolved. Compared to standard latent semantic analysis which stems from linear algebra and downsizes the occurrence tables (usually via a singular value decomposition), probabilistic latent semantic analysis is based on a mixture decomposition derived from a latent class model. == Model == Considering observations in the form of co-occurrences ( w , d ) {\displaystyle (w,d)} of words and documents, PLSA models the probability of each co-occurrence as a mixture of conditionally independent multinomial distributions: P ( w , d ) = ∑ c P ( d ) P ( c | d ) P ( w | c ) = P ( d ) ∑ c P ( c | d ) P ( w | c ) {\displaystyle P(w,d)=\sum _{c}P(d)P(c|d)P(w|c)=P(d)\sum _{c}P(c|d)P(w|c)} with c {\displaystyle c} being the words' topic. Note that the number of topics is a hyperparameter that must be chosen in advance and is not estimated from the data. The first formulation is the symmetric formulation, where w {\displaystyle w} and d {\displaystyle d} are both generated from the latent class c {\displaystyle c} in similar ways (using the conditional probabilities P ( d | c ) {\displaystyle P(d|c)} and P ( w | c ) {\displaystyle P(w|c)} ), whereas the second formulation is the asymmetric formulation, where, for each document d {\displaystyle d} , a latent class is chosen conditionally to the document according to P ( c | d ) {\displaystyle P(c|d)} , and a word is then generated from that class according to P ( w | c ) {\displaystyle P(w|c)} . Although we have used words and documents in this example, the co-occurrence of any couple of discrete variables may be modelled in exactly the same way. So, the number of parameters is equal to c d + w c {\displaystyle cd+wc} . The number of parameters grows linearly with the number of documents. In addition, although PLSA is a generative model of the documents in the collection it is estimated on, it is not a generative model of new documents. Their parameters are learned using the EM algorithm. == Application == PLSA may be used in a discriminative setting, via Fisher kernels. PLSA has applications in information retrieval and filtering, natural language processing, machine learning from text, bioinformatics, and related areas. It is reported that the aspect model used in the probabilistic latent semantic analysis has severe overfitting problems. == Extensions == Hierarchical extensions: Asymmetric: MASHA ("Multinomial ASymmetric Hierarchical Analysis") Symmetric: HPLSA ("Hierarchical Probabilistic Latent Semantic Analysis") Generative models: The following models have been developed to address an often-criticized shortcoming of PLSA, namely that it is not a proper generative model for new documents. Latent Dirichlet allocation – adds a Dirichlet prior on the per-document topic distribution Higher-order data: Although this is rarely discussed in the scientific literature, PLSA extends naturally to higher order data (three modes and higher), i.e. it can model co-occurrences over three or more variables. In the symmetric formulation above, this is done simply by adding conditional probability distributions for these additional variables. This is the probabilistic analogue to non-negative tensor factorisation. == History == This is an example of a latent class model (see references therein), and it is related to non-negative matrix factorization. The present terminology was coined in 1999 by Thomas Hofmann.
Truth discovery
Truth discovery (also known as truth finding) is the process of choosing the actual true value for a data item when different data sources provide conflicting information on it. Several algorithms have been proposed to tackle this problem, ranging from simple methods like majority voting to more complex ones able to estimate the trustworthiness of data sources. Truth discovery problems can be divided into two sub-classes: single-truth and multi-truth. In the first case only one true value is allowed for a data item (e.g birthday of a person, capital city of a country). While in the second case multiple true values are allowed (e.g. cast of a movie, authors of a book). Typically, truth discovery is the last step of a data integration pipeline, when the schemas of different data sources have been unified and the records referring to the same data item have been detected. == General principles == The abundance of data available on the web makes more and more probable to find that different sources provide (partially or completely) different values for the same data item. This, together with the fact that we are increasing our reliance on data to derive important decisions, motivates the need of developing good truth discovery algorithms. Many currently available methods rely on a voting strategy to define the true value of a data item. Nevertheless, recent studies, have shown that, if we rely only on majority voting, we could get wrong results even in 30% of the data items. The solution to this problem is to assess the trustworthiness of the sources and give more importance to votes coming from trusted sources. Ideally, supervised learning techniques could be exploited to assign a reliability score to sources after hand-crafted labeling of the provided values; unfortunately, this is not feasible since the number of needed labeled examples should be proportional to the number of sources, and in many applications the number of sources can be prohibitive. == Single-truth vs multi-truth discovery == Single-truth and multi-truth discovery are two very different problems. Single-truth discovery is characterized by the following properties: only one true value is allowed for each data item; different values provided for a given data item oppose to each other; values and sources can either be correct or erroneous. While in the multi-truth case the following properties hold: the truth is composed by a set of values; different values could provide a partial truth; claiming one value for a given data item does not imply opposing to all the other values; the number of true values for each data item is not known a priori. Multi-truth discovery has unique features that make the problem more complex and should be taken into consideration when developing truth-discovery solutions. The examples below point out the main differences of the two methods. Knowing that in both examples the truth is provided by source 1, in the single truth case (first table) we can say that sources 2 and 3 oppose to the truth and as a result provide wrong values. On the other hand, in the second case (second table), sources 2 and 3 are neither correct nor erroneous, they instead provide a subset of the true values and at the same time they do not oppose the truth. == Source trustworthiness == The vast majority of truth discovery methods are based on a voting approach: each source votes for a value of a certain data item and, at the end, the value with the highest vote is select as the true one. In the more sophisticated methods, votes do not have the same weight for all the data sources, more importance is indeed given to votes coming from trusted sources. Source trustworthiness usually is not known a priori but estimated with an iterative approach. At each step of the truth discovery algorithm the trustworthiness score of each data source is refined, improving the assessment of the true values that in turn leads to a better estimation of the trustworthiness of the sources. This process usually ends when all the values reach a convergence state. Source trustworthiness can be based on different metrics, such as accuracy of provided values, copying values from other sources and domain coverage. Detecting copying behaviors is very important, in fact, copy allows to spread false values easily making truth discovery very hard, since many sources would vote for the wrong values. Usually systems decrease the weight of votes associated to copied values or even don’t count them at all. == Single-truth methods == Most of the currently available truth discovery methods have been designed to work well only in the single-truth case. Below are reported some of the characteristics of the most relevant typologies of single-truth methods and how different systems model source trustworthiness. === Majority voting === Majority voting is the simplest method, the most popular value is selected as the true one. Majority voting is commonly used as a baseline when assessing the performances of more complex methods. === Web-link based === These methods estimate source trustworthiness exploiting a similar technique to the one used to measure authority of web pages based on web links. The vote assigned to a value is computed as the sum of the trustworthiness of the sources that provide that particular value, while the trustworthiness of a source is computed as the sum of the votes assigned to the values that the source provides. === Information-retrieval based === These methods estimate source trustworthiness using similarity measures typically used in information retrieval. Source trustworthiness is computed as the cosine similarity (or other similarity measures) between the set of values provided by the source and the set of values considered true (either selected in a probabilistic way or obtained from a ground truth). === Bayesian based === These methods use Bayesian inference to define the probability of a value being true conditioned on the values provided by all the sources. P ( v ∣ ψ ( o ) ) = P ( ψ ( o ) ∣ v ) ⋅ P ( v ) P ( ψ ( o ) ) {\displaystyle P(v\mid \psi (o))={\frac {P(\psi (o)\mid v)\cdot P(v)}{P(\psi (o))}}} where v {\displaystyle \textstyle v} is a value provided for a data item o {\displaystyle \textstyle o} and ψ ( o ) {\displaystyle \textstyle \psi (o)} is the set of the observed values provided by all the sources for that specific data item. The trustworthiness of a source is then computed based on the accuracy of the values that provides. Other more complex methods exploit Bayesian inference to detect copying behaviors and use these insights to better assess source trustworthiness. == Multi-truth methods == Due to its complexity, less attention has been devoted to the study of the multi-truth discovery Below are reported two typologies of multi-truth methods and their characteristics. === Bayesian based === These methods use Bayesian inference to define the probability of a group of values being true conditioned on the values provided by all the data sources. In this case, since there could be multiple true values for each data item, and sources can provide multiple values for a single data item, it is not possible to consider values individually. An alternative is to consider mappings and relations between set of provided values and sources providing them. The trustworthiness of a source is then computed based on the accuracy of the values that provides. More sophisticated methods also consider domain coverage and copying behaviors to better estimate source trustworthiness. === Probabilistic Graphical Models based === These methods use probabilistic graphical models to automatically define the set of true values of given data item and also to assess source quality without need of any supervision. == Applications == Many real-world applications can benefit from the use of truth discovery algorithms. Typical domains of application include: healthcare, crowd/social sensing, crowdsourcing aggregation, information extraction and knowledge base construction. Truth discovery algorithms could be also used to revolutionize the way in which web pages are ranked in search engines, going from current methods based on link analysis like PageRank, to procedures that rank web pages based on the accuracy of the information they provide.
Master data
Master data represents "data about the business entities that provide context for business transactions". The most commonly found categories of master data are parties (individuals and organisations, and their roles, such as customers, suppliers, employees), products, financial structures (such as ledgers and cost centres) and locational concepts. Master data should be distinguished from reference data. While both provide context for business transactions, reference data is concerned with classification and categorisation, while master data is concerned with business entities. Master data is, by its nature, almost always non-transactional in nature. There exist edge cases where an organization may need to treat certain transactional processes and operations as "master data". This arises, for example, where information about master data entities, such as customers or products, is only contained within transactional data such as orders and receipts and is not housed separately. ISO 8000 is the international standard for data quality and data portability in master data. == Alternative definition == An alternative definition of the term master data is that it represents the business objects that contain the most valuable, agreed upon information shared across an organization. In this sense, it gives context to business activities and transactions, answering questions like who, what, when and how as well as expanding the ability to make sense of these activities through categorizations, groupings and hierarchies. It can cover relatively static reference data, transactional, unstructured, analytical, hierarchical and metadata. What constitutes master data under this definition is therefore not about an essential quality of the data (e.g. it is a business entity that provides context for business transactions), but rather about the context in which the organisation has decided to treat the data. == Externally-defined master data == For most organisations, most or all master data is defined and managed within that organisation. Some master data, however, may be externally defined and managed. This represents the single source of basic business data used across a marketplace, regardless of organisation or location. Thus, it can be used by multiple enterprises within a value chain, facilitating "integration of multiple data sources and literally [putting] everyone in the market on the same page." An example of market master data is the Universal Product Code (UPC) found on consumer products. == Master data management == Curating and managing master data is key to ensuring its quality and thus fitness for purpose. All aspects of an organisation, operational and analytical, are greatly dependent on the quality of an organization's master data. Master Data is therefore the focus of the information technology (IT) discipline of master data management (MDM). Without this discipline in place, organisations commonly encounter difficulties with having multiple versions of "the truth" about a business entity, both within individual applications, and distributed across applications.
Iteration
Iteration means repeating a process to generate a (possibly unbounded) sequence of outcomes. Each repetition of the process is a single iteration, and the outcome of each iteration is the starting point of the next iteration. In mathematics and computer science, iteration (along with the related technique of recursion) is a standard element of algorithms. == Mathematics == In mathematics, iteration may refer to the process of iterating a function, i.e. applying a function repeatedly, using the output from one iteration as the input to the next. Iteration of apparently simple functions can produce complex behaviors and difficult problems – for examples, see the Collatz conjecture and juggler sequences. Another use of iteration in mathematics is in iterative methods which are used to produce approximate numerical solutions to certain mathematical problems. Newton's method is an example of an iterative method. Manual calculation of a number's square root is a common use and a well-known example. == Computing == In computing, iteration is a technique that marks out of a block of statements within a computer program for a defined number of repetitions. That block of statements is said to be iterated. A computer programmer might also refer to that block of statements as an iteration. === Implementations === Loops constitute the most common language constructs for performing iterations. The following pseudocode "iterates" three times the line of code between begin & end through a for loop, and uses the values of i as increments. It is permissible, and often necessary, to use values from other parts of the program outside the bracketed block of statements, to perform the desired function. Iterators constitute alternative language constructs to loops, which ensure consistent iterations over specific data structures. They can eventually save time and effort in later coding attempts. In particular, an iterator allows one to repeat the same kind of operation at each node of such a data structure, often in some pre-defined order. Iteratees are purely functional language constructs, which accept or reject data during the iterations. === Relation with recursion === Recursions and iterations have different algorithmic definitions, even though they can generate identical results. The primary difference is that recursion can be a solution without prior knowledge as to how many times the action must repeat, while a successful iteration requires that foreknowledge. Some types of programming languages, known as functional programming languages, are designed such that they do not set up a block of statements for explicit repetition, as with the for loop. Instead, those programming languages exclusively use recursion. Rather than call out a block of code to repeate a pre-defined number of times, the executing code block instead "divides" the work into a number of separate pieces, after which the code block executes itself on each individual piece. Each piece of work is divided repeatedly until the "amount" of work is as small as possible, at which point the algorithm does that work very quickly. The algorithm then "reverses" and reassembles the pieces into a complete whole. The classic example of recursion is in list-sorting algorithms, such as merge sort. The merge sort recursive algorithm first repeatedly divides the list into consecutive pairs. Each pair is then ordered, then each consecutive pair of pairs, and so forth until the elements of the list are in the desired order. The code below is an example of a recursive algorithm in the Scheme programming language that outputs the same result as the pseudocode under the previous heading. == Education == In some schools of pedagogy, iterations are used to describe the process of teaching or guiding students to repeat experiments, assessments, or projects, until more accurate results are found, or the student has mastered the technical skill. This idea is found in the old adage, "Practice makes perfect." In particular, "iterative" is defined as the "process of learning and development that involves cyclical inquiry, enabling multiple opportunities for people to revisit ideas and critically reflect on their implication." Unlike computing and math, educational iterations are not predetermined; instead, the task is repeated until success according to some external criteria (often a test) is achieved.
Artificial intelligence industry in Italy
The artificial intelligence industry in Italy is growing and supports industrial development. In 2024 it reached a new record, reaching 1.2 billion euros with a growth of +58% compared to 2023. While in 2025, the growth of artificial intelligence in the industrial application was even greater than in 2024 both in terms of value and application to industrial sectors. == History == The roots of AI research in Italy extend back to the 1970s, when Italian scholars began exploring automated reasoning, programming language semantics, and pattern recognition. Researchers such as those involved in early projects at the National Research Council and various universities laid the groundwork for subsequent academic and industrial developments in the field. During this period, the focus was predominantly on developing algorithms for automated theorem proving and building systems to reason about complex mathematical problems. This era witnessed the birth of methodologies that would later influence numerous AI subfields, from natural language processing (NLP) to robotics. === Institutional milestones and academic contributions === A turning point in the Italian AI landscape was the formation of the Italian Association for Artificial Intelligence (AIxIA) in 1988. Founded by academics, including Luigia Carlucci Aiello, the association established a platform for collaboration between universities, research centers, and industry. Led by Aiello, AIIA played a role in promoting research, organizing national conferences, and fostering international partnerships that connected Italy's AI community to global networks. At the same time, professors such as Roberto Navigli and numerous practitioners contributed to the advancement of AI in Italy. Navigli has worked in multilingual NLP, including the creation of BabelNet, and led the Minerva project. === Industrial AI === Over recent decades, numerous national and European initiatives supported by funding from programs such as the National Recovery and Resilience Plan (PNRR) have spurred the transition from theoretical research to practical applications. Industrial sectors including manufacturing, banking, and healthcare increasingly embraced AI-driven automation, while research institutions collaborated with industrial partners to deploy cutting-edge solutions. In recent years, Italy has also seen the establishment of specialized research centers and institutes aimed at bridging the gap between academic innovation and industrial application. These initiatives indicate a broader national commitment to integrating AI into the fabric of Italian industry. == Recent developments == === Emergence of generative AI === A landmark in Italy's modern AI evolution is the development of Minerva AI. Developed by the Sapienza NLP research group at Sapienza University of Rome and led by Professor Roberto Navigli, Minerva represents the first family of large language models (LLMs) trained from scratch with a primary focus on the Italian language. ==== Minerva 7B ==== The latest iteration, Minerva 7B, has 7 billion parameters and has been trained on an extensive corpus of over 1.5 trillion words. By using advanced instruction tuning techniques, Minerva 7B is able to produce highly accurate, coherent, and contextually sensitive responses addressing common issues such as hallucinations and inappropriate content generation. This breakthrough sets a benchmark for transparent, open-source AI development in the country. Minerva's development, carried out within the FAIR (Future Artificial Intelligence Research) project in collaboration with CINECA and supported by supercomputing resources like the Leonardo (supercomputer), aligns closely with Italy's cultural and linguistic heritage. === Establishment of AI4I === The recent establishment of the Istituto Italiano per l’Intelligenza Artificiale (AI4I) is part of Italy's strategy to improve its industrial competitiveness in AI. This dedicated institute aims to bridge the gap between research institutions and industrial enterprises; promote training and R&D support to nurture the next generation of Italian AI experts; and enhance national competitiveness. This initiative is expected to serve as a hub for applied AI research, driving innovations that are tailored to the specific needs of Italian industry and public administration. === Benefits of InvestAI === Italy's AI industry stands to benefit from the European InvestAI initiative, a plan unveiled at the recent AI Action Summit in Paris. InvestAI is an effort by the European Commission to mobilize €200 billion for AI investments, with a dedicated €20 billion fund earmarked for building AI gigafactories. These gigafactories are planned as large-scale hubs for training advanced, complex AI models using approximately 100,000 last-generation AI chips. For Italy, this investment presents several major opportunities: Access to State-of-the-Art Infrastructure: Italian companies, research institutions, and start-ups can leverage the gigafactories’ immense computational resources, enabling them to train highly sophisticated language models and other AI systems. Enhanced Competitiveness and Collaboration: With InvestAI's layered funding model where EU funds help de-risk private investments Italian firms can access capital more readily. This will bolster public–private partnerships and create a more dynamic AI ecosystem that spans from academic research to industrial applications. Alignment with National and Regional Initiatives: The Istituto Italiano per l’Intelligenza Artificiale (AI4I), based in Turin, is already recognized as a strategic asset by both Italy and the European Union. As the main recipient of InvestAI funds in Italy, AI4I will play a pivotal role in implementing these investments locally, fostering innovation in sectors like manufacturing, healthcare and aerospace. Commission President Ursula von der Leyen emphasized that InvestAI is designed to democratize AI innovation throughout Europe by ensuring that even smaller companies have access to high-performance computing power. For Italy, this means not only keeping pace with global leaders but also harnessing European-scale investments to transform its AI industry and drive economic growth.
Human–robot collaboration
Human-Robot Collaboration is the study of collaborative processes in human and robot agents work together to achieve shared goals. Many new applications for robots require them to work alongside people as capable members of human-robot teams. These include robots for homes, hospitals, and offices, space exploration and manufacturing. Human-Robot Collaboration (HRC) is an interdisciplinary research area comprising classical robotics, human-computer interaction, artificial intelligence, process design, layout planning, ergonomics, cognitive sciences, and psychology. Industrial applications of human-robot collaboration involve Collaborative Robots, or cobots, that physically interact with humans in a shared workspace to complete tasks such as collaborative manipulation or object handovers. == Collaborative Activity == Collaboration is defined as a special type of coordinated activity, one in which two or more agents work jointly with each other, together performing a task or carrying out the activities needed to satisfy a shared goal. The process typically involves shared plans, shared norms and mutually beneficial interactions. Although collaboration and cooperation are often used interchangeably, collaboration differs from cooperation as it involves a shared goal and joint action where the success of both parties depend on each other. For effective human-robot collaboration, it is imperative that the robot is capable of understanding and interpreting several communication mechanisms similar to the mechanisms involved in human-human interaction. The robot must also communicate its own set of intents and goals to establish and maintain a set of shared beliefs and to coordinate its actions to execute the shared plan. In addition, all team members demonstrate commitment to doing their own part, to the others doing theirs, and to the success of the overall task. == Theories Informing Human-Robot Collaboration == Human-human collaborative activities are studied in depth in order to identify the characteristics that enable humans to successfully work together. These activity models usually aim to understand how people work together in teams, how they form intentions and achieve a joint goal. Theories on collaboration inform human-robot collaboration research to develop efficient and fluent collaborative agents. === Belief Desire Intention Model === The belief-desire-intention (BDI) model is a model of human practical reasoning that was originally developed by Michael Bratman. The approach is used in intelligent agents research to describe and model intelligent agents. The BDI model is characterized by the implementation of an agent's beliefs (the knowledge of the world, state of the world), desires (the objective to accomplish, desired end state) and intentions (the course of actions currently under execution to achieve the desire of the agent) in order to deliberate their decision-making processes. BDI agents are able to deliberate about plans, select plans and execute plans. === Shared Cooperative Activity === Shared Cooperative Activity defines certain prerequisites for an activity to be considered shared and cooperative: mutual responsiveness, commitment to the joint activity and commitment to mutual support. An example case to illustrate these concepts would be a collaborative activity where agents are moving a table out the door, mutual responsiveness ensures that movements of the agents are synchronized; a commitment to the joint activity reassures each team member that the other will not at some point drop his side; and a commitment to mutual support deals with possible breakdowns due to one team member's inability to perform part of the plan. === Joint Intention Theory === Joint Intention Theory proposes that for joint action to emerge, team members must communicate to maintain a set of shared beliefs and to coordinate their actions towards the shared plan. In collaborative work, agents should be able to count on the commitment of other members, therefore each agent should inform the others when they reach the conclusion that a goal is achievable, impossible, or irrelevant. == Approaches to Human-Robot Collaboration == The approaches to human-robot collaboration include human emulation (HE) and human complementary (HC) approaches. Although these approaches have differences, there are research efforts to develop a unified approach stemming from potential convergences such as Collaborative Control. === Human Emulation === The human emulation approach aims to enable computers to act like humans or have human-like abilities in order to collaborate with humans. It focuses on developing formal models of human-human collaboration and applying these models to human-computer collaboration. In this approach, humans are viewed as rational agents who form and execute plans for achieving their goals and infer other people's plans. Agents are required to infer the goals and plans of other agents, and collaborative behavior consists of helping other agents to achieve their goals. === Human Complementary === The human complementary approach seeks to improve human-computer interaction by making the computer a more intelligent partner that complements and collaborates with humans. The premise is that the computer and humans have fundamentally asymmetric abilities. Therefore, researchers invent interaction paradigms that divide responsibility between human users and computer systems by assigning distinct roles that exploit the strengths and overcome the weaknesses of both partners. == Key Aspects == Specialization of Roles: Based on the level of autonomy and intervention, there are several human-robot relationships including master-slave, supervisor–subordinate, partner–partner, teacher–learner and fully autonomous robot. In addition to these roles, homotopy (a weighting function that allows a continuous change between leader and follower behaviors) was introduced as a flexible role distribution. Establishing shared goal(s): Through direct discussion about goals or inference from statements and actions, agents must determine the shared goals they are trying to achieve. Allocation of Responsibility and Coordination: Agents must decide how to achieve their goals, determine what actions will be done by each agent, and how to coordinate the actions of individual agents and integrate their results. Shared context: Agents must be able to track progress toward their goals. They must keep track of what has been achieved and what remains to be done. They must evaluate the effects of actions and determine whether an acceptable solution has been achieved. Communication: Any collaboration requires communication to define goals, negotiate over how to proceed and who will do what, and evaluate progress and results. Adaptation and learning: Collaboration over time require partners to adapt themselves to each other and learn from one's partner both directly or indirectly. Time and space: The time-space taxonomy divides human-robot interaction into four categories based on whether the humans and robots are using computing systems at the same time (synchronous) or different times (asynchronous) and while in the same place (collocated) or in different places (non-collocated). Ergonomics: Human factors and ergonomics are one of the key aspects for a sustainable human-robot collaboration. The robot control system can use biomechanical models and sensors to optimize various ergonomic metrics, such as muscle fatigue.
Kullback–Leibler Upper Confidence Bound
In multi-armed bandit problems, KL-UCB (for Kullback–Leibler Upper Confidence Bound) is a UCB-type algorithm that is asymptotically optimal, in the sense that its regret matches the problem-dependent Lai-Robbins lower bound. == Multi-armed bandit problem == The Multi-armed bandit problem is a sequential game where one player has to choose at each turn between K {\displaystyle K} actions (arms). Behind every arm a {\displaystyle a} there is an unknown distribution ν a {\displaystyle \nu _{a}} that lies in a set D {\displaystyle {\mathcal {D}}} known by the player (for example, D {\displaystyle {\mathcal {D}}} can be the set of Gaussian distributions or Bernoulli distributions). At each turn t {\displaystyle t} the player chooses (pulls) an arm a t {\displaystyle a_{t}} , he then gets an observation X t {\displaystyle X_{t}} of the distribution ν a t {\displaystyle \nu _{a_{t}}} . === Regret minimization === The goal is to minimize the regret at time T {\displaystyle T} that is defined as R T := ∑ a = 1 K Δ a E [ N a ( T ) ] {\displaystyle R_{T}:=\sum _{a=1}^{K}\Delta _{a}\mathbb {E} [N_{a}(T)]} where μ a := E [ ν a ] {\displaystyle \mu _{a}:=\mathbb {E} [\nu _{a}]} is the mean of arm a {\displaystyle a} μ ∗ := max a μ a {\displaystyle \mu ^{}:=\max _{a}\mu _{a}} is the highest mean Δ a := μ ∗ − μ a {\displaystyle \Delta _{a}:=\mu ^{}-\mu _{a}} N a ( t ) {\displaystyle N_{a}(t)} is the number of pulls of arm a {\displaystyle a} up to turn t {\displaystyle t} The player has to find an algorithm that chooses at each turn t {\displaystyle t} which arm to pull based on the previous actions and observations ( a s , X s ) s < t {\displaystyle (a_{s},X_{s})_{s