eduzhai > Applied Sciences > Engineering >

Constrained Active Classification Using Partially Observable Markov Decision Processes

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: In this work, we study the problem of actively classifying the attributes ofdynamical systems characterized as a finite set of Markov decision process(MDP) models. We are interested in finding strategies that actively interactwith the dynamical system and observe its reactions so that the attribute ofinterest is classified efficiently with high confidence. We present adecision-theoretic framework based on partially observable Markov decisionprocesses (POMDPs). The proposed framework relies on assigning a classificationbelief (a probability distribution) to the attributes of interest. Given aninitial belief, confidence level over which a classification decision can bemade, a cost bound, safe belief sets, and a finite time horizon, we computePOMDP strategies leading to classification decisions. We present two differentalgorithms to compute such strategies. The first algorithm computes the optimalstrategy exactly by value iteration. To overcome the computational complexityof computing the exact solutions, we propose a second algorithm is based onadaptive sampling to approximate the optimal probability of reaching aclassification decision. We illustrate the proposed methodology using examplesfrom medical diagnosis and privacy-preserving advertising.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×