China
Africa
Explore chapters and articles related to this topic
Introduction
Published in Joseph Y.-T. Leung, Handbook of SCHEDULING, 2004
The idea behind an online algorithm is that the algorithm does not have access to the entire input instance as it makes its decisions. For a thorough introduction to online algorithms that extends beyond online scheduling algorithms, please refer to the book by Borodin and El-Yaniv [1] and the collection of surveys [2]. In scheduling, we model a range of different environments which differ in the way the information is released. These are discussed below, as well as some additional notations specific to online problems that we use to supplement the standard three-field notation introduced in Chapter 1.
The Kernel-Based Online Anomaly Detection Algorithm
Published in Mohiuddin Ahmed, Abu S. S. M. Barkat Ullah, Al-Sakib Khan Pathan, Security Analytics for the Internet of Everything, 2020
Salva Daneshgadeh, Tarem Ahmed, Al-Sakib Khan Pathan
Online algorithms have the ability to learn from a newly arriving data instance, without re-training the whole data obtained to-date from initiation. As online algorithms involve real-time operations, the computational and storage complexities of the algorithms (in terms of both time and memory) are required to not grow with time as the size of the whole (to-date) dataset grows, and preferably be small.
Lightweight Embedded Systems
Published in Vojin G. Oklobdzija, Digital Design and Fabrication, 2017
Foad Dabiri, Tammara Massey, Ani Nahapetian, Majid Sarrafzadeh, Roozbeh Jafari
A power minimization for online algorithm is presented that carries out DVS and tailors the algorithms to today's real-world processors. The online algorithm schedules tasks without information about future tasks, which is often the case in real-time systems. It utilizes discrete voltage values, as found in today's dynamically variable voltage processors.
Exact reoptimisation under gradual look-ahead for operational control in production and logistics
Published in International Journal of Systems Science: Operations & Logistics, 2023
The most prominent concept for measuring online algorithm performance is competitive analysis (Borodin & El-Yaniv, 2005). An online algorithm is called c-competitive () in a minimisation problem, if on any problem instance the objective of the online algorithm is at most c times the objective of an optimal offline algorithm. Hence, competitive analysis is a worst-case analysis. It was first carried out by Graham (1966) for list scheduling. The idea was revived for data communication structures by Sleator and Tarjan (1985) and from then on experienced increasing attention. The monograph by Borodin and El-Yaniv (2005) is entirely devoted to the discipline and applies it to different settings from packing, to data structures, to financial problems. Similarly, the collection edited by Fiat and Woeginger (1998) surveys online problems (both from computer science and classical combinatorial optimisation) addressed by competitive analysis. Several refinements were proposed to ease the worst-case orientation. Surveys on such alternatives are given by Boyar et al. (2015) and Dorrigiv (2010). Different approaches are suggested such as restricting the power of the online adversary issuing the input sequence, increasing the resources of the online algorithm, perturbing the input element order or integrating probabilistic information.
An investigation on environmental pollution due to essential heavy metals: a prediction model through multilayer perceptrons
Published in International Journal of Phytoremediation, 2023
Murat Sari, Ibrahim Ertugrul Yalcin, Mahmut Taner, Tahir Cosgun, Ibrahim Ilker Ozyigit
The gradient descent algorithm can achieve the optimum solution with two different implementations: batch or online. The online algorithm updates the synaptic weights after each single training data record; in other words, it utilizes information from one record at one time. Online training continuously updates the weights until one of the stopping rules is met. Whenever all the records are utilized once and none of the stopping criteria is fulfilled, then the process is reiterated over the data records. Online training has superiorities over batch mode for relatively larger datasets with associated predictors; that is if there are a large number of records and too many inputs, and their values are not independent of each other, then online training could attain a plausible answer than batch training within a remarkably less amount of time. Since the number of collected samples is 800, the batch mode is utilized to obtain the results of this paper.
Joint production and maintenance operations in smart custom-manufacturing systems
Published in IISE Transactions, 2019
Jin Xu, Hoang M. Tran, Natarajan Gautam, Satish T. S. Bukkapatnam
The Fixed Speed Approach requires one revealed workload, and we see from Table 4 below a big improvement of performance it makes (94.38% of optimal offline) compared to the Stochastic Optimal Policy (83.33% of optimal offline). Ability to acquire workload distribution beforehand is also important. It helps improve the performance of online algorithms. If workload is revealed, however, the distribution of future workload is unknown, the optimal control policy for this case would be myopic as well: we choose the maximal speed (given by ) to process the current job w. We call it Simple Approach (SA) (a simplified data-driven method) which is shown in Table 4. When using Simple Approach we assume workloads are drawn from a single uniform distribution; however our online algorithms can be modified accordingly if workload distributions are changed.