Traditional history matching methods are time consuming due to their need for iterant forward reservoir simulations, which makes them impractical to deal with problems like effective monitoring well placements, quantifying importance of different source of data, etc. In my current work with Professor Durlofsky, we are trying to develop a reliable and fast method which can provide reasonably good production prediction and uncertainty quantification results. The key concept here is that we adopt a data-driven history matching approach, in which forward reservoir simulations are only used to provide a production data pool during the first stage, and the following stage will perform history matching based on the data pool generated at the first stag. This method is tremendously efficient since no sequential forward simulations are needed, which makes the overall time cost to be the same order as a single forward reservoir simulation. In real practice, all simulations at the first stage can also be paralleled, which will also accelerate the whole process.