Global Journal of Computer Science and Technology, D: Neural & Artificial Intelligence, Volume 23 Issue 2

may not require every agent to take other agents' decisions into account. i) Learning with Complete Data Learning can involve supervised learning and inductive learning, which focus on learning functions from examples. Supervised learning relies on feedback from a teacher or environment to improve performance. Classification and regression are types of learning for discrete valued functions. Inductive learning aims to find a reliable theory that supports the given examples. Occam's razor suggests favoring simpler models with fewer assumptions to avoid overfitting. j) Learning with Hidden Variables: The Expectation- Maximized Algorithm The cortex is constantly learning and inferring elements that produce sensory information. New learning algorithms and techniques have been presented in recent years that enable neural network models to learn these properties from real-world photos, text, audio signals, etc. The Expectation-Maximization (EM) algorithm, which is used to calculate the local maximum likelihood estimates (MLE) or maximum a posteriori estimates (MAP) for unobservable variables in statistical models, is described as the combination of many unsupervised machine learning algorithms. k) Statistical Learning Method Data comprehension technologies encompass statistical learning, which can be categorized into supervised and unsupervised learning. Supervised learning involves predicting or estimating an output based on inputs, while inferential statistics and descriptive statistics are commonly used in data analysis. Machine learning utilizes statistical methods, linear algebra, and calculus for improvement. Statistical learning plays a crucial role in various fields, such as research, business, and industry. An example application is predicting the likelihood of a patient experiencing another heart attack after being hospitalized for one. l) Reinforcement Lear Reinforcement learnin g 32 32 Andrew Barto and Richard S. Sutton, “Reinforcement Learning: An Introduction”, 2018 is a machine learning method that uses rewards and penalties to train agents. It involves perceiving and understanding the environment, taking actions, and learning from mistakes. Positive reinforcement learning includes two forms: 1) Markov Decision Process and 2) Q-learning. Unlike supervised learning, which relies on example data, reinforcement learning involves interaction with the environment. It finds applications in trajectory optimization, motion planning, dynamic pathing, controller optimization, and scenario-based learning policies for autonomous driving. For example, it can be used to learn automated parking strategies. Reinforce- ment learning is a powerful tool for decision-making and optimization, and it is widely utilized in machine learning applications today. V. C ommunication Artificial intelligence and communication go hand in hand. AI's defining test case and experimental data come from communication, especially interpersonal conversational interaction. Level 1: We will initially receive meeting bots that resemble the command-and-control bots we presently use; we will not have to press any buttons; instead, we may say to a bot, "Join the meeting," and it will set everything up for us. These meeting bots will require active activation and will have limited language and knowledge of context. Streamlining the mechanical processes, we all detest, like dialing complex conference numbers, will make meetings more comfortable. Level 2: Understanding of Natural Language: Beyond simple voice recognition, bots that comprehend the context and know the status of a meeting they are in will begin to appear. It will be possible for us to command, "Remind Sam to send this presentation to the team." On- demand Level 2 meeting bots can understand simple linguistic connections, monitor meeting activity (e.g., who is present, what file is being exhibited), and manage more complex aspects of professional interactions. Level 3: Semantic Comprehension and Domain Knowledge: A meeting bot at this level can tell us, "I have analyzed your meeting, and here is a summary of the key points." A meeting bot at this level will listen to meetings and be able to tell what subjects are being addressed. It will provide its analysis following a discussion, which can aid our memory of important spoken issues. These bots will collect word clouds from meeting recordings and perform sentiment analysis to create summaries of what happened. They can include company- and domain-specific knowledge bases, such as jargon dictionaries and FAQs, in their analysis for greater accuracy. A Level 3 bot will provide more than just operational ease; it will assist participants in achieving their goals by cognitively processing some of the meeting content. Although there are now some reliable post-meeting support tools that summarize subjects and sentiment, it will still be five years before we can construct Level 3 bots that are trustworthy enough to begin releasing goods. Speech-to-text conversion (which machines can now do) is a much simpler task than analyzing human intent from human speech, and it is also outside the capabilities of current natural-language technology. However, this is the way we are going. © 2023 Global Journals Global Journal of Computer Science and Technology Volume XXIII Issue II Version I 18 ( ) Year 2023 D Journey of Artificial Intelligence Frontier: A Comprehensive Overview