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

statistical learning. Machine learning benefits from mathematical optimization research and incorporates unsupervised learning for exploratory data analysis. Some machine learning systems utilize neural networks to simulate biological brain functioning. Predictive analytics is another term used for machine learning in solving business challenges. Learning algorithms assume that past successful methods will likely continue to be successful. Machine learning programs can perform tasks without explicit programming, learning from available data. This approach is particularly useful for increasingly complex tasks where manually designing algorithms becomes challenging. Machine learning employs techniques such as recognizing multiple valid responses and refining algorithms through practice data. The two main goals of modern machine learning are constructing data-supported models for classification and using these models to predict future events, such as identifying malignant moles or advising stock traders. c) Machine Learning as Subfield of AI Machine learning is an area of study that developed from the search for artificial intelligence. Some academics were intrigued by the idea of having machines learn from data in the early stages of artificial intelligence as a field of study. They tried using various symbolic techniques, including what was then referred to as "neural networks"- mostly perceptron and other models that were later discovered to be reimagining of the generalized linear statistics models. The use of probabilistic reasoning was widespread, particularly in automated diagnosis. A gap has emerged between AI and machine learning due to a shift towards a logical, knowledge- based approach in AI. Probabilistic systems faced challenges in data gathering and representation, leading to a decline in the popularity of statistics and the rise of expert systems in AI. Symbolic/knowledge-based learning and pattern recognition, more statistical in nature, moved beyond the realm of AI. Neural network development was abandoned by both AI and computers until researchers from other fields reintroduced it in the 1980s, known as connectionism. In the 1990s, machine learning experienced a resurgence as a distinct field, focusing on practical problem-solving using techniques from applied mathematics, statistics, and symbolic logic. Understanding the distinction between AI and machine learning is important. AI involves agents interacting with the world to learn and take actions, while machine learning learns and predicts based on passive observations. Some consider machine learning as a subset of AI, while others see it as a distinct but intelligent subset. 1. Supervised Learning In supervised learning 11 2. Unsupervised Learning , a mathematical model is created based on training data consisting of inputs and expected outputs. Each training example contains a supervisory signal in the form of desired output. Supervised learning algorithms learn a function through iterative optimization to accurately predict outputs for new inputs. Regression is used for numerical output, classification for limited set outputs, and Active Learning is another category. Similarity learning, connected to regression and classification, focuses on learning from examples using a similarity function. Applications of similarity learning include speaker verification, visual identification tracking, recommendation systems, rating, and face and identity verification. These techniques enhance machine learning's ability to make accurate predictions or outputs over time by learning from data. Unsupervised learnin g 12 3. Semi-Supervised Learning is a type of machine learning where the training algorithm does not receive labeled data but instead searches for patterns and structures in the input data. It is used for feature learning or finding hidden patterns in data. Unsupervised learning algorithms analyze input-only data to identify similarities or groupings, such as clustering. These algorithms do not rely on feedback and instead act based on commonalities found in the data. Unsupervised learning is applied in various fields, including density estimation and data summarization. Cluster analysis is a technique in unsupervised learning that divides a dataset into subsets or clusters based on predetermined criteria. Different clustering approaches make assumptions about data structure and use similarity metrics to evaluate clusters. Internal compactness and separation are measures used to assess the quality of clustering, along with estimated density and graph connectedness. Unsupervised learning plays a vital role in understanding and describing data features without relying on labeled information. Between supervised Learning (fully labeled training data) and unsupervised Learning (no labeled training data), there is semi-supervised Learning 13 11 Abhishek Vijavargia “Machine Learning using Python”, BPB Publications, 1st Edition, 2018 12 Yuxi Liu, “Python Machine Learning by Example”, 2nd Edition, PACT, 2017 . Many machine learning researchers have discovered that unlabeled data can significantly improve learning accuracy, even though some training examples lack training labels when used with a small amount of labeled data. Although the training labels in poorly supervised Learning are frequently less expensive to obtain, this results in larger functional training sets. 13 Tom Mitchel “Machine Learning”, Tata McGraW Hill, 2017. © 2023 Global Journals Global Journal of Computer Science and Technology Volume XXIII Issue II Version I 13 ( )D Year 2023 Journey of Artificial Intelligence Frontier: A Comprehensive Overview