Filtering with Abstract Particles
Jacob Steinhardt, Percy Liang · 2014
By using particles, beam search and sequential Monte Carlo can approximate distributions in an extremely flexible manner. However, they can suffer from sparsity and inadequate cover-age on large state spaces. We present a new fil-tering method for discrete spaces that addresses this issue by using “abstract particles, ” each of which represents an entire region of state space. These abstract particles are combined into a hier-archical decomposition, yielding a compact and flexible representation. Empirically, our method outperforms beam search and sequential Monte Carlo on both a text reconstruction task and a multiple object tracking task.