Page 34 - IJAMD-1-1

International Journal of AI
for Material and Design ML for quality improvement in L-PBF

without utilizing labeled data. Semi-supervised learning wherein the acquisition of new knowledge from prior
combines supervised learning and unsupervised learning, experiences facilitates the subsequent delivery of
utilizing a limited amount of labeled data and a more appropriate solutions in new circumstances, leveraging the
substantial volume of unlabeled data for model training. learned knowledge or methodology. However, machines
RL, on the other hand, lacks supervised signals, but it or models must learn from the provided data due to the
has measurable feedback. Besides the learning approach, absence of prior experiences. The general steps of supervised
considerable attention is directed toward hyperparameters learning are shown in Figure 3. To be more concrete, the
and appropriate optimization methods in ML methods. training dataset consists of “examples (x)” and “labels
Details of these learning processes are discussed in the (y).” The labels of the data act as instructions for guiding
subsequent sections. the machine to construct an accurate methodology. In
supervised learning, two main tasks prevail: classification
2.1. Learning approaches and regression. The main difference between these two
2.1.1. Supervised learning tasks lies in the nature of the variable y. In the regression
task, y represents a continuous real value, whereas in the
Supervised learning is a significant branch within the field classification task, y takes the form of a discrete category. For
of ML, finding widespread application across diverse real- instance, predicting the actual density or porosity values of
world scenarios. The fundamental principle of supervised fabricated products in L-PBF is a regression task, whereas
learning involves emulating the human learning process, distinguishing between poor and good quality in printed

Figure 2. Basic steps and branches of machine learning.

Figure 3. Steps of supervised learning.
Abbreviation: ML: Machine learning.

Volume 1 Issue 1 (2024) 28 https://doi.org/10.36922/ijamd.2301