Abstract The consensus algorithm has been popular in current distributed systems as it is more effective solving server unreliability. It ensures a group of servers can form coordinated system, and the entire system continues to work when part service point fails. Raft well-known widely used algorithm, but built-in purpose comprehensibility, always compromised terms performance trade-off. In th...

We analyze a data-processing system with n clients producing jobs which are processed in batches by m parallel servers; the throughput critically depends on batch size and corresponding sub-additive speedup function that arises due to overhead amortization. In practice, optimization relies numerical searches for optimal is computationally cumbersome. this paper, we model terms of closed queuein...

We demonstrate that training ResNet-50 on ImageNet for 90 epochs can be achieved in 15 minutes with 1024 Tesla P100 GPUs. This was made possible by using a large minibatch size of 32k. To maintain accuracy with this large minibatch size, we employed several techniques such as RMSprop warm-up, batch normalization without moving averages, and a slow-start learning rate schedule. This paper also d...

We present investigation of gene expression profiles by means of singular spectrum analysis (SSA). The biological problem under investigation is the decomposition of bicoid protein profiles of Drosophila melanogaster into the sum of a signal and noise, where the former consists of an exponential-in-distance pattern and is close to constant nonspecific component, or "background." The signal proc...

False alarms triggered by security sensors incur high costs for all parties involved. According to police reports, a large majority of alarms are false. Recent advances in machine learning can enable automatically classifying alarms. However, building a scalable alarm verification system is a challenge, since the system needs to: (1) process thousands of alarms in real-time, (2) classify false ...

We study the problem of planning the production of new and recovering defective items of the same product manufactured on the same facility. Items of the product are produced in batches. The processing of a batch includes two stages. In the first work stage, all items of a batch are manufactured and good quality items go to the inventory to satisfy given demands. In the second rework stage, som...

A constraint propagation (filtering) turned out to be an efficient method how to reduce search space for backtrack-based algorithms. This paper brings several improvements of filtering methods for batch processing with sequence dependent setup times. The first one is domain filtering based on a precedence graph—such filtering can be useful even when no precedences are given at the start. I desc...