Pre-Training Acquisition Functions by Deep Reinforcement Learning for Fixed Budget Active Learning

Abstract There are many situations in supervised learning where the acquisition of data is very expensive and sometimes determined by a user’s budget. One way to address this limitation active learning. In study, we focus on fixed budget regime propose novel algorithm for pool-based problem. The proposed method performs with pre-trained function so that maximum performance can be achieved when number acquired fixed. To implement algorithm, uses reinforcement based deep neural networks as tailored situation. By using Q-learning-based function, realize learner which selects sample annotation from pool unlabeled samples taking fixed-budget situation into account. experimentally shown comparable or superior existing methods, suggesting effectiveness approach