Due to most of the existing graph repartitioning methods are known for poor efficiency in distributed environments. In this paper, we introduce a new graph repartitioning method with two phases in distributed environments. In the first phase, a local method is designed to identify all the potential candidate vertices that should be moved to the other partitions at once in each partition locally. In the second phase, a streaming graph processing model is adopted to reassign the candidate vertices to achieve lightweight graph repartitioning. During the reassignment of the vertex, we propose an objective function to balance both the load balance and the number of crossing edges among the distributed partitions. The experimental results with a large set of real word and synthetic graph datasets show that the communication cost can be reduced by nearly 1 to 2 orders of magnitude compared with the existing methods.

We consider an edge-weighted graph G with a designated vertex v0 such that weights of edges incident to v0 may increase or decrease. We show that, with an O(mn+n2log n) time preprocessing, a minimum cut of the current G can be computed in O(log n) time per update of weight of any edge {v0,u}.

Since their inception almost fifty years ago, hidden Markov models (HMMs) have have become the predominant methodology for automatic speech recognition (ASR) systems--today, most state-of-the-art speech systems are HMM-based. There have been a number of ways to explain HMMs and to list their capabilities, each of these ways having both advantages and disadvantages. In an effort to better understand what HMMs can do, this tutorial article analyzes HMMs by exploring a definition of HMMs in terms of random variables and conditional independence assumptions. We prefer this definition as it allows us to reason more throughly about the capabilities of HMMs. In particular, it is possible to deduce that there are, in theory at least, no limitations to the class of probability distributions representable by HMMs. This paper concludes that, in search of a model to supersede the HMM (say for ASR), rather than trying to correct for HMM limitations in the general case, new models should be found based on their potential for better parsimony, computational requirements, and noise insensitivity.