Data Analytics and Machine Learning Assignment Sample

Data Analytics and Machine Learning Assignment Sample

Introduction

Game Theory is related to humans and machines from entertainment to business. In the artificial intelligence era where machines can learn and develop without being programmed or instructions through the machine learning process. Machine Learning controls the majority of artificial intelligence advancements and applications (Merrick and Taly, 2019). Machine Learning algorithms make patterns or built models based on huge amounts of data. When multi-objective optimization problems occur in machine learning, game theory can give firm solutions to the program. This presents is about game theory and how it works for some of the machine learning problems. For this assessment five articles will be assessed for a depth learning of the machine and game theory. The research is not limited to such models as some other literatures will be included as well.

Looking for Help With Assignments in the UK? Look no further than Native Assignment Help. Our team of experienced professionals is dedicated to providing top-notch assistance to students across the UK, ensuring they excel in their academic endeavours.

Discussion

Game Theory in Machine Learning

Game Theory is playing a crucial role in economics, social sciences, biology, and in artificial intelligence field. This theory is useful for planning and developing Artificial Models in the technology world. Game theory models are increasingly useful mechanisms for machine learning like multi-agent, artificial intelligence systems including imitation and reinforcement learning (Lasaulce and Tembine, 2011). With game theory, a multi-agent system can interact with its participants.

Game theory proclaims some functional forms as well which assist machine learning. The functional form is defined a pair of < N, v>. In this form, N presents finite set of players and v functional from define 2 the power N to R with v=0. Under this subset of S of N proclaims coalition. Moreover, v (S) worth the coalition S in the game of v. Thus, the game evaluates< N, v> with the function of v but in no confusion case with the player. Further the player set N denoted with G raise to power N. The set N is equivalent to {1, 2, …, n}.

Game theory is a branch of Mathematics that contrive the rules and outcomes (Davis, 2012). This theory is essentially required to enable or develop the environment of a multi-agent system, where different AI factors required to interact or compete in a pattern for achieving a target with accuracy or efficiency. A good way to start with principles of game theory is to involve AI Systems and understands the different aspect of games which encounter every day in the life of economics, business, finance, telecommunications, transport, insurance, war &defence, law ethic, political science, agriculture, environment, biology, medicine, sports, psychology, sociology, and social interactions.

• There are many areas where the combination of game theory and artificial intelligence have the power to provide the real-world solutions in different areas from cybersecurity to healthcare systems. Game theory and AI is a road to approach software systems rather than a particular technique.
• In the AI environment for using game theory, it should include more than one participant. Game Theory can be used in an efficient manner while work on multi-participant model. Mechanism design and Participant Design these two fundamentals steps are active in architecting game dynamics in an AI System.
• Participant Design: With game theory participants elements interact to predesign the environment and enhance to get the maximum results.
• Mechanism Design: In mechanism design, game theory use inverse concept for designing a game for a group or simulation of intelligent participant.

Five different kinds of games which examples of game theory.

Game theory is a process where multiple agents will communicate and compete to achieve the desired task. It also works on the interaction patterns Multi-Agent AI Systems, and understand the multiple kinds of factors and dynamics (Colman, 2016). It is a crucial element to develop well organized and productive gamified AI Systems Model.

Symmetric and Asymmetric Games

Chess is a good example of a symmetric game where each player has to achieve the same goal and task completion depends on strategies. Symmetry is one of the simplest categorizations in the real world (Curiel, 2013). The use of mathematical strategies work as symmetry as participants have different and conflicting goals. The finance sector is an example of the asymmetric game in which planner has different goals and get the result from a different desired task.

For a symmetric game, some retailers want to maximise their profit. The profit maximisation can be denoted as πrðp; τ; wÞ which further describe as max p;τ πrðp; τ; wÞ ¼ max p;τ ðpwÞDðpÞkτ2 þcfτDðpÞ:. On the other hand, manufacturer profit maximisation denoted as πmðw; p; τÞ, this further describe as:

Perfect and Imperfect Information Games

This is an important classification of task that is dependent on the availability of the type of information. In perfect information, games provide a platform where each player can view an analysis of the other player's actions. Game theory categorized as imperfect information where each player's actions are hidden from others.

As per this picture, Nash Equilibrium states that mixed strategies with game proclaim a finite player. Each player choosepure strategy with at least one Nash equilibrium.

Simultaneous and Sequential Games

There is so much difference between simultaneous and sequential games. Simulation game is a pattern where a model runs and things get explain in a relevant manner. It also depends on a person aspect and strategy to work. On the other side in sequential games player action depend other's player previous actions.

Co-operative and Non-Cooperative Games

Cooperative games provide a platform which enhance the probability to maximize the results as negotiation is a good example (Chalkiadakis, et. al., 2011). Non- cooperative games are less focused on alliance and modules. The cooperative game theory possesses an example of five people A,B,C,D, and E.

All of them have decided to start a business and initiate some funds to process activities. An assess of the situation and business, they determine that company can achieve a profit of 100 which further distributed in a equal manner. Two of the partners determine that a cooperation of work between only two can earn a profit of more than 40. Then three partners determine that their cooperation of work can generate an revenue of 70. Two partners offer to provide a profit of 71 and share 29 between two. Thus, such issues can be resolved with assistance and support of cooperative game theory.

Zero-Sum vs Non-Zero Games

 Zero-sum games provide a platform where a player gains something that a loss to the other players. Board games are the example of zero-sum games where a player always loses for other players but in Environment of Non-Zero Games, where multiple players gains from one player's actions. Non-Zero games example be like where multiple participants join to enhance the business market with finance interactions.

What is the big and broad idea of this specific machine learning model?

Researchers have to build many of machine learning model and to solve a problem using machine learning choose one or many of these algorithms. Machine learning model are explicit in the problem-solving approach (Samek, et. al., 2017). Model-based machine learning can be used in any problem, its approach that doesn't require to learn a huge amount of data of machine learning algorithms and techniques. There is always assumption build applications in any algorithms. Model-based machine learning used in the task using the platform to plan for enhancing the accuracy and security. The machine learning algorithms are defined as follow:

• Classification
• Regression
• Clustering
• Dimensionality Reduction
• Deep Learning

1. Classification

Classification is the process of forecast the type of class of an object. The result variable is coming under a classified variable. Check or predict an email is spam or not is a classification task (Strumbelj and Kononenko, 2010). There are some machine learning models for classification problems.

• SVM: involves binary or multiclass classification.
• Logistic Regression: linear model in ML for binary classification.
• K-Nearest neighbors algorithms - simple but computationally extensive.
• Ensembles - a group of multiple machine learning participants models join together for the desired output.

Machine learning presents different definitions where objective just predict. The first aspect states feature space A as cartesian product of n feature which further presents sets of N = {1,2,...,n}): A = A1 ×A2 ×...×An. For describing the situation, feature values are ignored and define a subspace as AS = A ′ 1 ×A ′ 2 ×...×A ′ n , where A ′ i = Ai if i∈ S and A ′ i = {ε}. Thus S⊂ N, as S not ignored AN = A.

2. Regression

Regression describes the environment in which regression provides a set of problems where the end variable can take values without any errors (Rezek, et. al., 2008).  Like airline price comes under in the regression task. there are some regression models used in machine learning.

• Decision Tree Regression
• SVM Regression
• Linear Regression
• Ridge Regression

(Figure 1: The Danger of extrapolation in regression, 2012)

(Source:Montgomery, et. al., 2012)

The data collected on the x and y define in the form of interval x2 is less than and equal to x and x3 is greater than and equal to x. This linear regression is a good example of true relationship and if equation use to predict values of y for regressor variables in similar region than it not performs well. And the range of x states equation error.

3. Clustering

Identify similar objects not possible manually as clustering is the process of grouping the same objects automatically. Without homogeneous data, there are no possibilities to build effective and efficient machine learning models. Clustering gives an environment to achieve the task more smartly. There are some well-known and widely used clustering models in machine learning.

1. DBSCAN
2. K Means
3. K Means ++
4. K Medoids
5. Agglomerative clustering
6. Dimensionality Reduction

Dimensionality is the figure of forecast variables used to predict open variable or predefined target. there is a group of huge number datasets of variables in the machine learning model. Too many datasets of variables do not contribute equally or efficiently to achieve the defined task. some generally used models for dimensionality reduction.

1. PCA - it builds short numbers of new variables from large numbers of predictors.
2. TSNE - Give lower-dimensional connection to Top Dimensional data points.
3. Deep Learning

 Deep learning build models to deal with neural networks in machine learning. deep learning is a subset or part of Machine learning which based on the architecture of neural networks, there are some important deep learning models:

Multi-Layer perceptron

 A multilayer perceptron is an artificial neural network. A multi-Layer perceptron combined of a minimum of three layers of nodes: an input layer, a hidden layer, and an output layer. hidden and output nodes are neurons that work on nonlinear functions. MLP is now deemed characteristic of fully combined layers or nodes. not successful is that the all-over total layers. inefficient there is redundancy in high dimensions. Lightweight MLP can get high accuracy with the MNIST datasheet.

Convolution Neural Networks

The incumbent computer vision algorithms image net competitions. each layer is around image according to a certain size and stride. layers also sparsely connected due to fully connected. matrices as well as vectors as inputs, every node doesn't connect to every other node (Sun, et. al., 2012). Efficient parameters sharing that the filter can be discovered more than one part of data or image. Tensorflows work high-level API to give developers an easy deep learning framework being applied to an MLP in Keras.

Recurrent Neural Networks

 Advantages or properties of recurrent Neural Networks their diversity of applications. when dealing with recurrent neural networks various inputs and output elements.

                        a). Sentiment Classification - the process of simply classifying on data into positive and negative sentiments. where input about lengths, output of type, and size.

                        b). image captioning - for an image that needs a textual description. so only input - the image and combination of words as output. the output is a   description of different lengths but the image might be fixed or the same size.

                        c). Language Translation - the language translation process to provide an offer where a particular data or text wish to translate in other languages, every language own style, semantics.

Boltzmann machine

Energy Models to associate scalar energy with the variable of interest. there are no output layers like classifiers, which cannot learn oneself map or pattern from independent variables or target variables. these properties make the machine learning model non-deterministic.

1. Autoencoders

Autoencoders are a machine learning model. Autoencoders (AE) is a group of neural networks for which the input and output both are the same. In Autoencoders input into a latent space representation and redevelop the output through latent space representation.

What has been achieved so far on this model with a chronological summary?

Algorithms have been discovered more than half-decade of the century ago in game theory and used in machine learning is mathematical equations in models. Researchers use algorithms which is simple and powerful, to way understand how models works and genetic diversity. Algorithms have been used to solve problems in models, zero-sum games, and other computer science problems, which is used to determine agent weight possible strategies when making a process of decisions (Gao, et. al., 2017). Games theorists assume that typically referred to in the economics.

This formulated by a simple process that players are rational. the importance of game theory in human-centered disciplines, this kind of rhetoric has improving magnet of diverse, the connection of web rationality western cultural tradition, use the concept economic rationality is technical, not normative Morgenstern original data of game theory, in the phrase rationality are no boundaries, rational expectations are beliefs that reflect accurately weighted all info available to the agent. Game theorists use of the concept need not implicate the ideology. economic rationality in some cases internal computations performed an agent and in other perspectives of economic rationality simply be embodied in dispositions natural, culture and market selection, in particular, imply no deliberation, conscious or otherwise act.

What Type of machine learning problems can be solved using this specific model?

The new time of Artificial Intelligence (AI) is much successful by the achievements of machine learning and deep learning which works on a massive amount of information and data patterns. Artificial Intelligence is not only about understanding real-world problems but also provide solutions for problems and making smart decisions based on understanding. In the challenging decision-making environment game theory is a situated pattern strategic connectivity between many decision-makers and participants. to develop and building a model for the real world in the AI system, the process of learning with game theory for strategy development. these models are accepting and fulfill the global world challenges with impressive social effectiveness like infrastructure system, cybersecurity, security, developments, etc., Strategic decision making is the solution to these problems. there are many ways to explore research and develop solutions for such problems but here are three main directions;

1. Data-Based Game Theory: Humans are so intelligent there is not any doubt that Humans are creating their competitors as AI because humans' decision-making capacity or accuracy not always correctly or perfectly relational. Beneficial thing is that there are tons of data available for use in Machine Learning Models or Building applications. some data problems occur when modeling the strategic decision making these patterns in game theory. these problems are how to capitalize on the amount of information, how to develop behaviours and work of model of human players, how to develop algorithms compute strategies depend on the models, algorithms patterns for creating solutions, use of mathematical programming for technical.
2. Learning Powered Strategy in Big Models: The combination of equilibrium strategy and the optimal strategy is challenging for developing or creating Big Models. In other sets of conditions more than two players, endless action patterns, interactions between participants, and of course no sustainability in the models. some depend on the position of techniques, with machine learning skills, deep reinforcement use to solve complex games. In the game models, the combinations of strategies are more difficult and difficult. some defending-attacking patterns modeling games are relevant to the situation’s strategies can secure and the sustainability of models.
3. Game Parameters from Start to End: When players value unknown in predefined situations then game parameters become the more challenging and crucial plan. To differentiate problems and give to them solutions learning all aspects of game parameters. this game solver is commonly used in deep neural network systems.

What are the open challenges in this model?

In the world, for the technology era, many challenges are customers' possible application of their innovative technology. Machine learning face just inverse, entrepreneurs, designers, and managers present capabilities of machine learning. because buzz about technology people assume that AI as a magic wand that will solve all problems. in automatic face recognition, analysis of financial risk less than in a few seconds, that's not easy at all. the commercial aspect of Machine learning, deep learning methods are new. the required properly organized and prepared data to accurately desired results. Enterprises working on machine learning applications need to invest time, resources, and risks. neural networks have millions of parameters of millions. the network is capable of remembering the training set and giving answers to efficient accuracy.

• The Black Box problem: The initial stages of machine learning to relatively simple, shallow methods, a decision tree algorithm strictly according to the rule’s supervisors taught. after analyzing large sets of pieces of information, data neural networks with super accuracy (Castelvecchi, 2016). Artificial intelligence supervisors the input and the output decision make. How engineers can understand how forecasting work and made, it's difficult to understand the complete model works.
• Talent deficit: There are many jobs in the machine learning industry, but there are few specialists that can develop this technology. Some scientist who understands machine learning experienced knowledge of engineering.
• Data is not Free: To develop a machine learning model needs a huge amount of data, it’s easy to afford data but it needs time to collect a sufficient amount of data, and of course, Data is very expensive.

Conclusion

Machine learning is a set of techniques of train machines to perform the desired task as well as the human mind can do, even faster and better than a human being. for example, AI beat human champions in games like chess and many more. now machines can be trained in human activities and serve better lives for humans. Machine learning can be categorized as supervised or unsupervised learning. less amount of data then goes for supervised learning but if you have large data sets unsupervised would generally fine performance and good results. deep learning techniques is best for a huge amount of data.

Reinforcement learning and deep reinforcement learning are detailed above and neural networks and their applications. when it comes to developers or build machine learning models there are choices of available, most of them connect your requirement easily and provide a scenario of AI Models. We can imagine complexity in developing this kind of models and applications but lots AI and Machine learning techniques to make applications. There are many choices of languages, IDES, and platforms to build models. in languages there are Python, R, Matlab, Octave, Julia, C++, C, etc., in Integrated Development Systems (IDEs) R Studio, Pycharm, iPython/Jupyter Notebook, Julia, Spyder, Anaconda, Rodeo, Google-Colab, and some platforms where Machine learning applications can be deployed like Microsoft Azure, IBM, Google Cloud, Amazon, MlFlow, etc. learning and practicing each machine learning techniques also necessary, The Subject is broad, if you deal with the dept, you will grabs as soon as your capabilities. every topic is independent and different from each other.

Reference

Castelvecchi, D., 2016. Can we open the black box of AI?. Nature News, 538(7623), p.20.

Chalkiadakis, G., Elkind, E. and Wooldridge, M., 2011. Computational aspects of cooperative game theory. Synthesis Lectures on Artificial Intelligence and Machine Learning, 5(6), pp.1-168.

Colman, A.M., 2016. Game theory and experimental games: The study of strategic interaction. Elsevier.

Curiel, I., 2013. Cooperative game theory and applications: cooperative games arising from combinatorial optimization problems (Vol. 16). Springer Science & Business Media.

Davis, M.D., 2012. Game theory: a nontechnical introduction. Courier Corporation.

Gao, W., Kwong, S. and Jia, Y., 2017. Joint machine learning and game theory for rate control in high efficiency video coding. IEEE Transactions on Image Processing, 26(12), pp.6074-6089.

Houston, A.I., Székely, T. and McNamara, J.M., 2013. The parental investment models of Maynard Smith: a retrospective and prospective view. Animal Behaviour, 86(4), pp.667-674.

Lasaulce, S. and Tembine, H., 2011. Game theory and learning for wireless networks: fundamentals and applications. Academic Press.

Merrick, L. and Taly, A., 2019. The Explanation Game: Explaining Machine Learning Models with Cooperative Game Theory. arXiv preprint arXiv:1909.08128.

Othman, M.F. and Shazali, K., 2012. Wireless sensor network applications: A study in environment monitoring system. Procedia Engineering, 41, pp.1204-1210.

Rezek, I., Leslie, D.S., Reece, S., Roberts, S.J., Rogers, A., Dash, R.K. and Jennings, N.R., 2008. On similarities between inference in game theory and machine learning. Journal of Artificial Intelligence Research, 33, pp.259-283.

Samek, W., Wiegand, T. and Müller, K.R., 2017. Explainable artificial intelligence: Understanding, visualizing and interpreting deep learning models. arXiv preprint arXiv:1708.08296.

Strumbelj, E. and Kononenko, I., 2010. An efficient explanation of individual classifications using game theory. The Journal of Machine Learning Research, 11, pp.1-18.

Sun, X., Liu, Y., Li, J., Zhu, J., Chen, H. and Liu, X., 2012. Feature evaluation and selection with cooperative game theory. Pattern recognition, 45(8), pp.2992-3002.

Sun, X., Liu, Y., Li, J., Zhu, J., Liu, X. and Chen, H., 2012. Using cooperative game theory to optimize the feature selection problem. Neurocomputing, 97, pp.86-93.

Montgomery, D.C., Peck, E.A. and Vining, G.G., 2012. Introduction to linear regression analysis (Vol. 821). John Wiley & Sons.