In this talk, I will briefly introduce the basic concept of machine learning and how it could be applied to fundamental research of physics. I will also present three projects in our group. We will show (1) how to use self-learning method to identify the topological phase transition from the experimental data without a priori theory; (2) how to use the random sampling neural network to calculate the energy eigenstates and their expectation in the strongly correlated regime by using data in the weakly interacting regime; (3) How to predict the long-time dynamics of a many-body system through quantum-inspired recurrent neural network. These examples show the possibility to explore important many-body problems through the application of some machine learning method.