In applications such as facial image retrieval, given a sample image, a retrieval system

is supposed to return a set of images that are similar to the sample image. To improve

retrieval accuracy, human intervention can be invited to manually tell the system,

among the set of returned images, which ones are indeed related to the sample

image. Discuss how the user response can be specifically used to improve the

performance of the retrieval system. $($Hint: this is a question related to semi$-$

supervised and$/$or reinforcement learning.$)$