Multiple Search Paths Research Articles

LOGER: A Learned Optimizer Towards Generating Efficient and Robust Query Execution Plans

Query optimization based on deep reinforcement learning (DRL) has become a hot research topic recently. Despite the achieved promising progress, DRL optimizers still face great challenges of robustly producing efficient plans, due to the vast search space for both join order and operator selection and the highly varying execution latency taken as the feedback signal. In this paper, we propose LOGER, a learned optimizer towards generating efficient and robust plans, aiming at producing both efficient join orders and operators. LOGER first utilizes Graph Transformer to capture relationships between tables and predicates. Then, the search space is reorganized, in which LOGER learns to restrict specific operators instead of directly selecting one for each join, while utilizing DBMS built-in optimizer to select physical operators under the restrictions. Such a strategy exploits expert knowledge to improve the robustness of plan generation while offering sufficient plan search flexibility. Furthermore, LOGER introduces ε -beam search, which keeps multiple search paths that preserve promising plans while performing guided exploration. Finally, LOGER introduces a loss function with reward weighting to further enhance performance robustness by reducing the fluctuation caused by poor operators, and log transformation to compress the range of rewards. We conduct experiments on Join Order Benchmark (JOB), TPC-DS and Stack Overflow, and demonstrate that LOGER can achieve a performance better than existing learned query optimizers, with a 2.07x speedup on JOB compared with PostgreSQL.

A Localization Method of Ant Colony Optimization in Nonuniform Space.

The purpose of geographic location selection is to make the best use of space. Geographic location selection contains a large amount of spatiotemporal data and constraints, resulting in too many solutions. Therefore, this paper adopts the ant colony algorithm in the meta-heuristic search method combined with the incomplete quadtree to improve the searchability of the space. This paper proposes an improved ant colony algorithm in nonuniform space to solve the P-center facility location problem. The geographic space is divided by the incomplete quadtree, and the ant colony path is constructed on the level of the quadtree division. Ant colonies can leave pheromones on multiple search paths, and optimized quadtree encoding in nonuniform space stores pheromone matrices and distance matrices. The algorithm proposed in this paper improves the pheromone diffusion algorithm and the optimization objective at the same time to update the pheromone in the nonuniform space and obtain the ideal solution. The results show that the algorithm has excellent performance in solving the location problem with good convergence accuracy and calculation time.

A mathematical model and a parallel multiple search path simulated annealing for an integrated distributed layout design and machine cell formation

Facility layout problem is a well-researched problem of finding configurations of departments and machines on a plant floor with the objective of improving material handling efficiency. With this objective, different techniques of configuring facilities have been documented in literature. Among them are cellular and distributed layouts. Cellular layouts are applicable in scenarios where demand and product mix are relatively stable and rational part families/machine cells can be identified. With this assumption, the literature provides many techniques for their design. Distributed layout, on the other hand, are recommended in volatile environments where product demand and mix are changing very rapidly. However, we argue that a real-life scenario may lay within the spectrum of these two extremes. In this paper, we attempt to bridge this gap by developing a mathematical model that integrates distributed layout design and machine cell formation with an objective to minimize a weighted sum of material handling and inter cellular movement costs. Through distributing the machines over the shop floor, the model attempts to minimize material handling cost. By identifying possible machine cells and part families, it attempts to minimize inter cellular movements. At the same time, the model ensures that machines that belong to the same cell are laid out on contiguous physical locations so that the advantages of cellular manufacturing systems can be fully exploited. Operations sequence, alternative routing, workload balancing among cells and other pragmatic issues are also incorporated in the model. We developed a parallel multiple search path simulated annealing to solve the proposed model efficiently. Several numerical examples are presented to illustrated the model and the computational performance of the developed algorithm.

New Improvements in Parallel Implementation of N-FINDR Algorithm

Endmember extraction (EE) is the first step in hyperspectral data unmixing. N-FINDR is one of the most commonly used EE algorithms. Nevertheless, its computational complexity is high, particularly, for a large data set. Following a parallel version of N-FINDR, i.e., P-FINDR, further improvements are presented in this paper. First, generic endmember re-extraction operation (GERO) and multiple search paths are introduced such that multiple endmembers are extracted in parallel. Second, by making full use of the advantages of the proposed algorithms, two extended schemes, i.e., extended mapping rule and multiple-stage GERO are presented, which can reduce synchronous cost and provide steady parallel performance. In experiments, the proposed algorithms have been quantitatively evaluated. The results demonstrate that they can outperform the conventional parallel computing and do not degrade the quality of EE.

Methodology for determining the optimal operating strategies for a chilled-water-storage system—Part I: Theoretical model

This article proposes a new methodology for determining the optimal operating strategies for a chilled-water storage system under a time-of-use electricity rate structure. It is based on a new classification of operating strategies and an investigation of multiple search paths. Each operating strategy consists of a control strategy and the maximum number of chillers running during the off-peak and on-peak periods. For each month, the operating strategy with the lowest monthly billing cost and minimal water level higher than the set-point is selected as the optimal operating strategy for the current month. A system model is developed to simulate the tank water level at the end of each time step and the system total power during each time step. This model includes six sub-models: plant, thermal energy storage tank, loop, chiller, control strategy, and nonplant power. This method considers the interaction among plant, tank, and loop, and it is easy for users to apply it to different projects, modify individual models, or implement new sub-models. The final results will be a table showing the monthly control strategy and maximal number of chillers staged-on during the off-peak and on-peak periods, an operation schedule that is easy for the operators to follow.

Time Efficient Indexing Technique for Querying Location Dependent Data in Wireless Broadcast

Wireless communication and mobile computing have gained a lot concentration from the computer and communication research society. Wireless dissemination has emerged as an outstanding and scalable method to broadcast information to huge number of clients equipped with hand held devices. In practice, wireless broadcast faces challenge from issues like timely delivery of data, error free quality of service, low battery capacity and limited bandwidth. Different applications need to curb out different challenges. To support location based services in wireless data broadcast system a distributed spatial index (DSI) is popular. DSI indexing technique is in recent times has been becoming popular as reduce access time and energy consumption in location based queries because it has a linear yet fully distributed structure that facilitates multiple search path to be naturally mixed together by sharing links. Here, in this paper we address an issue where error free timely delivery of information is major concern. Such state of affairs generally appears in an inhouse environment where recharging of battery is not a serious concern. We have proposed a Data Replicated Distributed Spatial Index (DR-DSI) which is extension of Distributed Spatial Index to get advantages of data replication to reduce access latency.

Multiple Search Paths and the General-To-Specific Methodology

Multiple Search Paths and the General-To-Specific Methodology

Large‐scale identification of genetic design strategies using local search

In the past decade, computational methods have been shown to be well suited to unraveling the complex web of metabolic reactions in biological systems. Methods based on flux–balance analysis (FBA) and bi-level optimization have been used to great effect in aiding metabolic engineering. These methods predict the result of genetic manipulations and allow for the best set of manipulations to be found computationally. Bi-level FBA is, however, limited in applicability because the required computational time and resources scale poorly as the size of the metabolic system and the number of genetic manipulations increase. To overcome these limitations, we have developed Genetic Design through Local Search (GDLS), a scalable, heuristic, algorithmic method that employs an approach based on local search with multiple search paths, which results in effective, low-complexity search of the space of genetic manipulations. Thus, GDLS is able to find genetic designs with greater in silico production of desired metabolites than can feasibly be found using a globally optimal search and performs favorably in comparison with heuristic searches based on evolutionary algorithms and simulated annealing.

Improvement of Continuous Dynamic Programming for Human Gesture Recognition

This paper describes an improved Continuous Dynamic Programming (CDP) for human gesture recognition. In order to cope with the variation in the amplitude of the input signal, a reference pattern set containing patterns with different mean amplitude is used instead of a single reference pattern. Furthermore, a new multiple search path determination strategy is developed in place of the traditional single one to deal with the variation in the speed of the input gesture. Also, a matching error controlled path determination method is newly introduced to reduce the influence of noise. The experimental results show this improved CDP can give not only stable identification results compared with the conventional one, but also the degree of the size and speed of the gesture to be distinguished.