On Compressibility and Acceleration of Orthogonal NMF for POMDP Compression

摘要: State space compression is one of the recently proposed approaches for improving POMDP's tractability. Despite its initial success, it still carries two intrinsic limitations. First, not all POMDP problems can be compressed equally well. Also, cost computing itself may become significant as size problem scaled up. In this paper, we address issues with respect to an orthogonal non-negative matrix factorization compression. particular, first propose eigenvalue analysis evaluate compressibility a and determine effective range dimension reduction. incorporate interior-point gradient acceleration into NMF derive accelerated version minimize overhead. The validity has been evaluated empirically. demonstrated in speeding up policy computation set robot navigation related problems.