We consider the problem of detecting anomalies among a given set of processes\nusing their noisy binary sensor measurements. The noiseless sensor measurement\ncorresponding to a normal process is 0, and the measurement is 1 if the process\nis anomalous. The decision-making algorithm is assumed to have no knowledge of\nthe number of anomalous processes. The algorithm is allowed to choose a subset\nof the sensors at each time instant until the confidence level on the decision\nexceeds the desired value. Our objective is to design a sequential sensor\nselection policy that dynamically determines which processes to observe at each\ntime and when to terminate the detection algorithm. The selection policy is\ndesigned such that the anomalous processes are detected with the desired\nconfidence level while incurring minimum cost which comprises the delay in\ndetection and the cost of sensing. We cast this problem as a sequential\nhypothesis testing problem within the framework of Markov decision processes,\nand solve it using the actor-critic deep reinforcement learning algorithm. This\ndeep neural network-based algorithm offers a low complexity solution with good\ndetection accuracy. We also study the effect of statistical dependence between\nthe processes on the algorithm performance. Through numerical experiments, we\nshow that our algorithm is able to adapt to any unknown statistical dependence\npattern of the processes.\n