The primary purpose of this dissertation is to address the critical issues in the developmentand design of a new class of antennas for the 5G application. Using mm-wave frequencies including 28 GHz and 38 GHz brings new challenges partially related to the high frequency of operation and partially related to the design of small low-cost phased arrays to operate at those frequencies. The first challenge in the 5G receiver and transmitter is the antenna design. When operating at high frequencies we need to overcome high path loss. To overcome this obstacle, a small antenna array with high gain and the ability to generate dynamically directional beams is desirable. Moreover, in a 5G cellular network using low cost, high efficiency, low weight, thin, and conformable (flexible) structures are desirable. The second challenge in using a flexible material is associated with its thickness which normally provides very narrowband designs. To the best of our knowledge, only a few antennas using a thin substrate with wideband and high-gain, have been reported. To address those two challenges, we present different types of compact and flexible antennas in single and linear array configurations operating at 28 GHz and 38 GHz. Second Purpose of this dissertation using Machine learning and AI in the n electromagnetic problems. To show, First we have presented an efficient scientific workflow that produces a Physics-Informed Convolutional Neural Network (CNN) model capable of running complicated electromagnetic wave propagation (at WiFi frequencies) simulations to determine the real-time Received Signal Strength Intensity (RSSI) which can be used for determining wireless communication connectivity in advanced manufacturing contexts. These simulations are done in a constant enclosed space composed of inner objects with variable position, shape, size, and reflectivity. The workflow utilizes numerical PDE simulations to generate synthetic data for both training and testing states and suggests adaptive mechanisms to incorporate real-time sensed data to improve accuracy for geolocating people (inferred from their mobile technologies such as smart watches) moving through WiFi fields. The second application that we have investigated is related to using machine learning for the Through the Wall Imaging (TWI) problem. To solve the TWI problem in presence of the wall, a Convolution Neuron Network (CNN) is proposed. The detection of the scatterers behind the wall when there is a strong reflection from the front wall is very challenging. Several microwave imaging algorithms have been introduced to extract the unknown parameters of the wall and mitigate the wall clutter to predict the location of the target. This process is very time-consuming even though it has accurate results. In this paper, we developed a method for exploiting the complex information for the TWI problem by using a CNN that accepts complex numbers from the receiver to predict wall thickness, material, and location of the target behind the wall at the same time. We show that our proposed model can predict these parameters with an accuracy of 92.6%.