Variational quantum circuits (VQC) are being used quite efficiently for Noisy Intermediate Scale Quantum Computers (which are not yet robust and fault tolerant), which have small number of qubits and shallow circuit depth. VQCs are a category of algorithms used in solving several categories of quantum machine learning problems. However, it has been discovered recently that during training these algorithms, the gradient vanishes exponentially as the qubit numbers and the circuit depth increase. This is the barren plateau problem which has a serious impact on many of the VQCs including the variational eigensolver, quantum approximate optimization algorithm, and quantum neural networks. Barren plateaus thus lead to significant challenges in scaling VQCs which are necessary to address problems involving higher qubit number and larger depth quantum circuits. In this paper, we focus on the impact of barren plateaus during training of a VQC model. We review the various algorithms which are used for training and optimizing VQCs for learning. Next, we present the various causes leading to the barren plateau problems. Finally, we review the solutions available to mitigate this problem.