In the coming era of 5G, heterogeneous cellular networks (HetNets) are adopted to meet the rapidly increasing of user demand and the amount of access devices. Resource allocation is a key issue in HetNets, which is considered to be a NP-hard problem. Therefore it is difficult to effectively achieve the accurate solution. In this paper, resource allocation and user association are taken into consideration together and are solved jointly by using automatic differentiation (AD). Automatic differentiation is a series of techniques for accurately and efficiently evaluating derivatives of numeric functions which are expressed as computer programs. AD is used generally in the field of deep learning recently for example the back propagation algorithm in artificial neural network. In this study, we first model the problems of resource allocation and user association as a mixed integer inequality constrained optimization problem. Then the constraints and objective function are constructed as a dynamic computational graph where the solutions for resource allocation and user association are the parameters in the graph. A loss function is defined as the combination of the constraints and objective function, which can be iteratively optimizing based on automatic differentiation scheme. Numerical results show the effectiveness and efficiency of the proposed algorithm. Higher network throughput and higher frequency reusing can be achieved compared to the max SINR and SINR bias method.