Название: Machine Learning: Multi-Agent Technologies
Автор: Igor Sheremet, Andries Engelbrecht
Издательство: ITexLi
Год: 2023
Страниц: 109
Язык: английский
Формат: pdf (true)
Размер: 15.2 MB

Multi-agent systems (MASs) are defined as a group of interacting entities or agents sharing a common environment that changes over time, with capabilities of perception and action, and the mechanisms for their coordination provide a modern perspective on systems that traditionally were regarded as centralized. The main characteristics of agents are learning and adaptation. In the last few years, MASs have received tremendous attention from scholars in different fields. However, there are still challenges faced by MASs and their integration with Machine Learning (ML) methods. The primary goal of the study is to provide a broad review of the current developments in the field of MASs combined with ML methods. First, we present features of MASs considering the ML perspective. Second, we provide a classification of applications of MASs combined with ML methods. Third, we present a density map of applications in E-learning, manufacturing, and commerce. We expect this study to serve as a comprehensive resource for researchers and practitioners in the area.

Machine Learning (ML) is a subset of Artificial Intelligence (AI) that concerns the development of algorithms, which allows the machine to learn via inductive inference based on observation data that represent incomplete information about statistical phenomena. To carry out the learning process an algorithm is used based on examples of the task we want to solve (data) and letting the computer find patterns and make inferences that optimize the decision-making according to a user-defined objective. Based on the training strategy, ML can be divided into three classical categories with different learning approaches: supervised learning, unsupervised learning, and reinforcement learning. The first one includes classification and regression tasks, in the second one the widely used task is clustering, and the third one consists of the process of training a model on a series of actions that lead to a particular outcome, where the system receives rewards for performing well and punishment for performing poorly directly from its environment. In the last years, MASs integrated with ML have received tremendous attention from scholars in different fields such as Computer Science, Engineering, Mathematics, Material Science, Neuroscience, Energy, Physics and Astronomy, Social Sciences, Environmental Sciences, Business, Management, and Accounting. The overview of MASs integrated with ML in these fields will be presented in the following sections. MASs have been used in areas of e-learning, manufacturing, and commerce combining mathematical methods, optimization methods, Markov processes, learning algorithms, and Artificial Intelligence techniques.

Scalability in training large numbers of deep reinforcement learning agents, which must decide on actions jointly, is a major issue that becomes apparent in many real-life problems. This issue is related to numerous aspects of deep multiagent reinforcement learning (DMARL), such as assignment of credits to the learners for their choices, assumptions regarding homogeneity or interchangeability of the agents, society structure due to interaction of agents’ decisions, agents’ communication requirements, abilities, and constraints. In the Chapter 2, we provide a review of deep multiagent reinforcement learning (DMARL) methods, examining their ability to scale up to large agent populations. “Large” here could mean anything from hundreds up to several thousands of agents.

Contents
1. A State-of-the-Art Survey on Various Domains of Multi-Agent Systems and Machine Learning
2. Deep Multiagent Reinforcement Learning Methods Addressing the Scalability Challenge
3. Role of an Optimal Multiagent Scheduling in Different Applications Using ML
4. On an Approach to Knowledge Management and the Development of the Knowledge-Вased Multi-Agent System
5. Modeling Electric Vehicle Charging Station Behavior Using Multiagent System
6. Approximate Dynamic Programming: An Efficient Machine Learning Algorithm