A general communication network can be modeled as an interference channel, wherein multiple transmitters communicate with multiple receivers. With the inclusion of the Internet of Things (IoT) in a cellular network, e.g., in a 6G communication system, there are two major challenges: one is to allocate optimal resources (power, etc.) to each IoT node and the second is to ensure that resource allocation is energy efficient, as IoT nodes have limited energy resources. Conventional techniques, like the water-filling-based solution for power allocation work, are computationally tractable for a few users. In a typical IoT interference network, the number of users is large, and conventional approaches do not scale up as the number of users increases. In this paper, we use ensemble deep neural network (DNN) and genetic algorithm (GA)-based power control methods to address the non-convex optimization problem of maximizing the sum rate and energy efficiency for the interference channel. We also consider an interference channel with an eavesdropper and compute the secrecy sum rate and energy efficiency using the same AI methods. For the deep neural network (DNN), we have considered an unsupervised learning strategy as the ground truth is missing in the power control problem, and we directly maximize the objective function by minimizing the loss in the training phase. To improve the performance of DNN, we have used the ensemble method for deep neural networks, as it takes the least amount of time. We observe that GA and DNN outperform conventional optimization techniques like sequential quadratic programming (SQP).