Alain Jean-Marie LIRMM University of Montpellier, France Sylwester Kaczmarek Gdansk University of Technology, Poland. Marek Natkaniec AGH University of Science and Technology, Poland Sema Oktug Istanbul Technical University, Turkey.
An Adaptive Heuristic Approach for the Multiple Depot Automated
Transit Network Problem
1 Introduction
We study the effect of these constraints at the operational level for the multi-storage topology ATN network. Also, the proposed heuristic approach was proven to find good quality solutions in a fast computational time.
2 The Automated Transit Networks
More recently, several operational and strategic optimization studies related to ATN have been published such as simulation [7], energy minimization [10], total distance traveled [6,8], optimized operational planning [4] and so on. But from our literature review, many optimization routing models related to ATN considered a single depot network topology [4,6,10].
3 The Optimization Model
The Set of Assumptions
A literature review published in 2005 [5] states that there are more than 200 research articles related to ATN. As a result, it becomes of great interest to study optimization routing problem related to ATN based on a multiple network topology.
Graph Based Formulation
The Complexity of Our Problem
4 Genetic Algorithm Approach
Solution Representation and Evaluation Function
Each edge in the auxiliary graph represents a feasible path based on the change in question. The algorithm then uses the shortest path in the auxiliary graph to find the corresponding set of paths.
Crossover and Mutation Operators
5 Computational Results
Test Instances
We must emphasize that the SOLIS solution of LB represents the linear relaxation of the valid mathematical formulation presented in the literature [1]. Mathematical models associated with linear relaxation were implemented using IBM ILOG CPLEX Optimizer 12.2.
Result of the Genetic Algorithm
To generate the different instances, the ATN's instance generator from the literature of Mrad and Hidri [10] was adapted to our context.
6 Conclusions
Fatnassi, E., Chebbi, O., Siala, J.C.: Comparison of two mathematical formulations for the offline routing of personal rapid transit system vehicles. Mrad, M., Hidri, L.: Optimal consumed electrical energy while scheduling vehicle trips in a personal rapid transit system.
In our previous work [7], the parameters of a memetic algorithm were tuned via the Taguchi method, under a limited computational budget, using a limited number of examples from several problem domains. In this study, we further analyze and extend our previous work in order to evaluate whether we can more quickly generalize the best setting with a reduced computational time budget.
2 Hyper-Heuristics Flexible Framework (HyFlex)
The low-level heuristics (operators) in HyFlex are categorized as mutation, ruin and recreate, crossover and local search [15]. More details about the domain implementations, including low-level heuristics and initialization routines, can be found on the competition website and in [1,15].
3 Methodology
The depth of search parameter determines the number of steps the local search heuristic will go through. Based on the number of parameters and levels, a suitable orthogonal array is selected to create a design table.
4 Experimentation and Results
The combination of the best values of each parameter is predicted to be the best overall setting. The best value for the depth of run-time search parameter changes; however, it is always one of the values 0.6, 0.8 or 1.0.
5 Conclusion
These three values combined with the best values of other parameters were then tested separately on all 45 instances from 9 domains, with the aim of finding the best DoS value across all instances. According to the result of experiments, each of these three configurations found the best values for 18 instances (including bands), considering their average performance over 31 runs.
Ensemble Move Acceptance in Selection Hyper-heuristics
In [1,8], different move acceptance criteria were used in a two-stage iterative framework that switches from one move acceptance to another at each stage. The same selection hyperheuristic is then tested on thirty problem instances from six different domains from the HyFlex benchmark.
2 Group Decision Making Selection Hyper-heuristics
For example, SR G-AND denotes a selection hyperheuristic using SR as the heuristic selection method and G-AND as the shift acceptance method. Each move-making group in group decision-making tested in this study includes three move-making methods: enhancement and equivalence (IE), simulated annealing (MC), and grand plunge (GD).
3 Computational Experiments
First, the performance of all group decision hyper-heuristics is compared with each other. Ranking of selected group decision hyper-heuristics for the CHeSC 2011 competitors based on Formula1.
4 Conclusion
CF G-AND is the group decision making hyper-heuristic winner and it ranks eleventh compared to the CHeSC 2011 hyper-heuristic with a total score of 39.00. Ensemble Move Acceptance in Selection Hyper-heuristics 29. Carter, M.W., Laporte, G., Lee, S.Y.: Exam timetable: algorithmic strategies and applications.
However, instances of the same error may exist elsewhere in the source code. Extended SCAT can detect violations of derived rules throughout the source code.
2 Related Work
Then we were able to detect several new instances of the identified errors that needed to be fixed. As such, the goal of the approach proposed in this article is instead to increase recall.
3 Generating Rules from Execution Traces
Therefore, it is important to be sure that the cause of the error is application specific. The name of the method call (m) is also obtained, provided the declaration involves a method call (Line 4).
4 Industrial Case Study
NC DVP
Our ISA then recalculates the potential of each of the results based on the updated network weights and reorders them, showing the user the results that have a higher potential or score. Examples of large 1D-BPP problems cannot be solved exactly due to the intractable nature of the problem.
2 Proposed Algorithm (1D-BPP-CUDA)
But the population is required to transfer from the device to the host after the initial generation in the device to truncate and add BFD to the population. 1D-BPP 55 of each chromosome, the population set is filled with chromosomes that are in the device's memory.
3 Performance Evaluation of Experimental Results
After finding the best population size for the algorithm, we performed tests on the number of generations to observe how it affects the solution quality and execution time of the algorithm. The number of optimal solutions has an increasing pattern for Set 1 and Set 2 datasets.
4 Conclusions and Future Work
Dokeroglu, T., Cosar, A.: Optimization of one-dimensional garbage packing problem with island parallel clustering genetic algorithms. Scholl, A., Klein, R., Jurgens, C.: BISON: A fast hybrid procedure for solving the one-dimensional garbage packing problem exactly.
Approach: Artificial Bee Colony Algorithm with VIKOR Method
22] established a multi-criteria classification model called AHP-K-Veto, based on the AHP method and the K-means algorithm. Each established classification is evaluated using an estimation function based on storage cost and fillrate service level [2], which also represents the objective function of our model, by minimizing the classification cost.
2 The Proposed Work
Artifical Bee Colony Algorithm
To the best of our knowledge, the Artificial Bee Colony algorithm [16–18] and the VIKOR method have not been used to solve the ABC MCIC problem. We also describe our proposed hybrid optimization model by adapting the ABC algorithm to meet the constraints of the problem.
VIKOR
A New Hybrid Approach for ABC MCIC
When the onlooker bee moves (Eq.4), the mutation operation of ABC algorithm needs to be adjusted, because the values of the generated solution may flood the search space. Once these solutions are generated by the ABC algorithm, they will be considered as input parameters by the VIKOR method, to calculate a score for each item, determine a total ranking items and, consequently, an ABC generating classification (according to the ABC distribution). ).
3 Experimental Results
4 Conclusions
Ghorabaee Keshavarz, M., Zavadskas, E.K., Olfat, L., Turskis, Z.: Multi-criteria inventory classification using a new evaluation method based on the distance from the average solution (edas). Liu, J., Liao, X., Zhao, W., Yang, N.: A classification approach based on a better ranking model for multi-criteria abc analysis.
Android Malware Classification by Applying Online Machine Learning
Static analysis of executables using commands and modeling of malware features using permissions and API calls is presented for detecting a malware in [2,3]. In Section 3, we apply the online machine learning algorithm to classify malware samples and evaluate the results.
2 Feature Set
3 Implementation
Online Classification Algorithms
After prediction, it takes the true label yt ∈ {+1,−1} and updates its model (a.k.a. hypothesis) based on the prediction loss(yt,yˆt) which means the mismatch between the prediction and the actual class. The goal of online learning is to minimize the total number of incorrect predictions; sum (t:yt= ˆyt).
Classification Metrics
True positives (tp) refer to the number of samples in class c that are correctly classified, while true negatives (tn) are the number of samples not in class c that are correctly classified. False positives (fp) refer to the number of out-of-class and misclassified samples.
Testing Accuracy Results
Similarly, false negatives (fn) are the number of samples in class c that are misclassified. In: Proceedings of the ICTAI 2013, The IEEE 25th International Conference on Tools with Artificial Intelligence, pp.
Approaches to Evaluation of Classifiers in Authorship Attribution Domain
Among the various methods, we can mention some unsupervised ones such as cluster analysis, multidimensional scaling and principal component analysis. Tests were performed on undiscretized and discretized data using different approaches to discretize the test data sets [3].
2 Theoretical Background
Different authorship attribution tasks were categorized in [12], and three types of problems were formulated: profiling – there is no candidate proposed as an author; the needle-in-a-haystack – author of analyzed text must be selected from thousands of candidates; verification – there is a candidate that needs to be verified as author of text. Depending on the discretization methods, different problems may occur, such as unequal number of bins in training and test data, or intersection points defining boundaries of bins may differ in both datasets.
3 Experimental Setup
Finally, separate training and test sets were prepared with two classes (corresponding to two authors) in each. According to the main objective of the presented research, the classifiers were evaluated using cross-validation and test datasets.
4 Results and Discussion
Performance of classifiers for undiscretized data for evaluation performed using cross-validation and test datasets. Performance of classifiers for data discretized using supervised Kononenko MDL (above) and Fayyad & Irani MDL (below) for evaluation performed using cross-validation and test data sets.
5 Conclusions
Summarizing the observations presented, it can be noted that for almost all experiments (only one exception was observed), the evaluation performed using cross-validation provided an approximately 10% higher quality measure compared to the evaluation based on test data sets. If you choose to use this method, you should consider that the actual performance of the system is much worse than that reported by the cross-validation evaluation.
Cosine Similarity-Based Pruning for Concept Discovery
The proposed method has been implemented as an extension to an existing concept discovery system called Tabular CRIS w-EF [14,15]. The experimental results show that the proposed method reduces the number of queries executed by an average of 15% without any loss of accuracy of the systems.
2 Background
To evaluate the effectiveness of the proposed method, several experiments are performed on data sets belonging to different learning problems. The proposed method is implemented as an extension of an existing ILP-based concept discovery system called Tabular CRIS w-EF.
3 Proposed Method
With such a transformation, the proposed method can be used to prune hypotheses for ILP-based concept discovery systems operating on Prolog-like environments. In terms of algorithmic complexity, the proposed method consists of two main steps (i) matrix construction and (ii) cosine similarity calculation.
4 Experimental Results
King, R.D., Muggleton, S., Lewis, R.A., Sternberg, M.: Drug design by machine learning: the use of inductive logic programming to model the structure-activity relationships of trimethoprim analogues binding to dihydrofolate reductase. Srinivasan, A., Muggleton, S.H., Sternberg, M.J., King, R.D.: Theories for mutagenicity: a study in first-order and trait-based induction.
Ontology and Web Service Modeling Language
We believe this is due to the inherent complexity of the language, the "less than complete" state of WSML (for example, this is the main contribution of this work, which will be used in the next phase of our research, the actual design and implementation of such language.
2 Evaluation of WSMO and WSML
- General Observations
- Deficiencies in Syntax
- Logical Basis of WSMO
- Lack of a Semantics Specification for Web Service Methods/Operations
- Implementation and Tool Support
- Choreography in WSMO
- Orchestration in WSMO
- Goal Specification
- Reusing Goals Through Specialization
- Specialization Mechanism for Web Service Specifications Developing a web service specification from scratch is a very formidable task
- Missing Aggregate Function Capability
- Extra-Logical Predicates
- Multiple Functionality in a Web Service
- Automatic Mapping Between Attributes and Relations
- Error Processing
- No Agreed-Upon Semantics for WSML-Full
In WSMO, matching between a target and a web service is done by considering the pre-post conditions of the target and the web service, which is fine. Using the same terminology for both targets and web services is also misleading.
3 Related Work
WSML-full, which is a combination of WSML-DL and WSML rules, has no agreed semantics yet [9]. With no formal semantics available, it is hard to imagine how the full WSML specification could be worked out at all.
4 Conclusion and Future Work
In future work, we are planning to develop a logic-based semantic web service framework that builds on the strengths of WSMO, but at the same time corrects the weaknesses identified in this paper. This chapter is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/ . 4.0/), which permits use, duplication, adaptation, distribution, and reproduction in any medium or format, provided as long as you give proper credit to the original author(s) and source, a link to the Creative Commons license is provided and any changes made are indicated.
Weighting and Pruning of Decision Rules by Attributes and Attribute Rankings
Feature Ranking
The importance of individual attributes can be discovered by some approach that leads to their ranking, that is, to assign values of a scoring function which causes to place them in a specific order [7]. Rankings of characteristic features can be achieved through the use of statistical measures, machine learning approaches or systematic procedures [12].
Pruning of Decision Rules
When fewer rules are found, the learning rate may be shorter, but the solutions are not necessarily the best. If a larger number of rules are generated, a more thorough and in-depth analysis is enabled, but even for rule sets with small cardinalities, some quality or interestingness measures can be applied [6].
3 Experimental Setup and Obtained Results
- Input Datasets
- Rankings of Attributes
- DRSA Rule Classifiers
- Pruning of Rule Sets by Attributes
- Pruning of Rule Sets Through Rule Rankings
- Summary of the Best Results
For male authors, attribution can be increased (at a maximum by more than 4%), either by maintaining or lowering restrictions on minimal support required by rules. When rules were assessed, ranked and then selected, the quality of prediction was improved at most by more than 3 % for both datasets, and for female and male authors, datasets could be pruned by more than 29 % and 18 % of rules, respectively.
Energy Consumption Model for Data Processing and Transmission in Energy Harvesting
Wireless Sensors
Earlier works [8,9] studied the performance of an energy harvesting sensor node as a function of random data and energy flow. In this paper, the main contribution is that we consider energy consumption not only for data transmission, but also for node electronics, that is, data sensing-processing-straw in an energy-harvesting wireless sensor.
2 Mathematical Model
Stability of the System
In the steady state, the probability that separate data packets and energy packets do not exceed some final values D and E: respectively. We can thus conclude that the system with unlimited storage capacity is always stable with respect to data packets and unstable with respect to energy packets, as expected.
3 Analysis of Transmission Error Among a Set of Nodes
Networks with Restart
Recently, we studied G-networks with multiple classes where the signal is used to change the class of a customer in the queue [4]. These models still have a product shape steady-state solution under certain technical conditions on the rope loads.
2 Model Assumptions and Closed Form Solutions
Consider an arbitrary open network G with p classes of positive customers and a single class of negative customers whose effect is to restart one customer in the queue. This result is used to obtain closed solutions for some performance measures: the probability of having exact customers in the queue and the expected number of customers in the queue.
3 Examples
More formally, we say that the system is balanced if the loads for customers in service (i.e. not preparing their migration) are equal for both queues (i.e.ρ(1)1 =ρ(1)2 =ρ) and we assume that the overhead is the burden of the queues due to the migration (i.e.ρ(2)1 +ρ(2)2. The probability of acceptance for the selection depends on the class of the client and the phase of service.
4 Concluding Remarks
This aspect of the tool will be highlighted in Section 2 by presenting an example (a hysteresis queue). The technical part of the paper is as follows: in Section 2, we present how we can build a new model.
2 Building a Model with XBorne
The first dimension is the number of clients and the second dimension codes the state of the servers. Once the steady-state distribution is obtained with some numerical algorithms, the marginal distributions and some rewards are calculated by describing the states obtained by the generation method and codes provided by the modeler (and compiled) to generate the rewards. specify (see in the left part of Fig.1 the marginal distribution for the queue size).
3 Numerical Resolution
Other techniques for inputwise bounds of the stationary distribution have also been derived and implemented [3]. For example, for a matrix of rank and magnitude N, the computation of the stationary distribution requires O(N k2) operations.
4 Quasi-Lumpability
In the left part we have drawn the Markov chain representation associated with Mitran's model for a small buffer size (ie 20). On the right side of the figure, we have presented a heat diagram for the energy consumption associated with Mitran's model for all values of the U and D thresholds.
5 Simulation
Dayar, T., Pekergin, N., Youn`es, S.: Conditional stationary bounds for a subset of states in Markov chains. Fourneau, JM, Le Coz, M., Pekergin, N., Quessette, F.: An open tool to compute stochastic bounds for stationary distributions and rewards.
Evaluation of Advanced Routing Strategies with Information-Theoretic Complexity Measures
In Sect.2, we summarize the basic concepts of Gershenson and Fernandez's information theory framework [4].
2 Complexity and Information
3 Information Associated to a Routing Table
4 HR-Based Routing
1. Hierarchy and Recursion: The routing table at node 4.2 contains information on how to reach another node in the network.
5 Simulation Results
However, a small increase in ofk corresponds to a high performance increase of the HR-based routing strategy. When mandα is high, the values of H in HR-based and NoHR routing are more similar.
6 Conclusion
Stojmenovic, I.: Simulations in wireless sensor and ad hoc networks: matching and progression of models, metrics and solutions.
Performance of selection hyper-heuristics on the extended HyFlex domains 155 A selection hyper-heuristic framework operates on a single solution and iteratively selects a heuristic from a set of low-level heuristics and applies it to the candidate -solution. HyFlex [14] (Hyper-heuristicsFlexible framework) is a cross-domain heuristic search API and HyFlex v1.0 is a software framework written in Java, which provides an easy-to-use interface for developing selection hyper - heuristic search algorithms together with the implementation of various problem domains, each of which includes problem-specific components, such as solution representation and low-level heuristics.
2 Selection Hyper-heuristics for the Extended HyFlex Problem Domains
The stroke acceptance method used in [10] accepts all enhancement strokes and non-enhancement strokes with an adaptive threshold. The basic idea of the selection function hyperheuristic is to select the best low-level heuristic at each iteration.
3 Empirical Results
The fuzzy late acceptance based hyper-heuristic (F-LAHH) [8] was applied to solve MAX-SAT problems and showed promising results. The performance of the best hyper-heuristics from Table 1, SSHH is compared with methods whose performances are reported in [1], including Adap-HH, which is the winner of the CHeSC 2011 competition [13], an Evolutionary Hyper Programming. -heuristic (EPH) [12], Iterative local search with fair division with (FS-ILS) and without restart (NS-FS-ILS), Simple random moves with all (SR-AM) (marked formerly AA-HH) and Simple Random-Improving or Equal (SR-IE) (formerly ANW-HH).
Energy-Efficiency Evaluation of Computation Offloading in Personal Computing
Computational offloading first started and was mainly studied for mobile devices [1-5] due to the marked difference in computing power between mobile devices and cloud servers [6]. Energy-aware scheduling of the execution of congested computations in the cloud was studied in [15].
3 Experimental Methodology
Also, applications running in the cloud sent the actual calculation progress back to the PC after the execution was complete. Neither the PC nor the VM in the cloud performed any other user-level activity during our experiment.
4 Experimental Results on Power Consumption Vs
The results of the executions were sent back to the computer, if it was only a text output, but if the application had to write a file, it was stored in the cloud (in the VM where the application was executed) and thus the execution time we measured in the latter case does not include the download time of the resulting files. There are some natural differences in the energy consumption of a cloud infrastructure comprising 8 compute nodes in a rack depending on temperature fluctuations.
Performance Tradeoff
Power Consumption and Performance
The average power load only represents the power consumption per unit of time, and thus the total amount of energy consumed by each application depends on the execution time, as seen in the part b's (right column) of fig. 1, 2, 3 and 4. Any increases in execution time were mainly due to the lower computing power of the VM in the cloud (vCPUs vs. real CPUs).
Energy Consumption
For the sake of simplicity, we have not taken into account the energy costs incurred by the network connection to the cloud. Looking at the applications separately, the total amount of energy input by each is lower for remote execution (ranging from 0.97% to 20.28%) compared to local execution.
5 Conclusions and Future Work
Rahimi, M.R., Ren, J., Liu, C.H., Vasilakos, A.V., Venkatasubramanian, N.: Mobile cloud computing: a survey, state of the art and future directions. Sakellari, G., Loukas, G.: A survey of mathematical models, simulation approaches and testbeds used for cloud computing research.
Analysis of Transient Virtual Delay in a Finite-Buffer Queueing Model
In this section, we provide a mathematical description of the considered queuing model and introduce the necessary notations and definitions. Installation time is required for the server to reach its full capacity to process jobs, so during installation the service process is suspended.
3 Integral Equations for Transient Queueing Delay Distribution
Due to the fact that successive departures are Markov times in the evolution of the M/G/1-type system (see e.g. [1]), the continuous version of the Total Probability Law must be applied with respect to the first departure after t = 0, we get the following system of integral equations: . The last sum on the right refers to the situation where the first service completion epoch occurs after timet; in such a case, if n=K, the queue delay in time equals 0, since the "virtual" packet arriving at that time is lost due to the buffer overflow.
4 Compact Solution for Queueing Delay Transforms
Transient virtual delay analysis in a finite-buffer queuing model 181 where n≥ 1 and αi(s) is given in (16).
5 Numerical Example
Niu, Z., Guo, X., Zhou, S., Kumar, P.R.: Characterization of power-delay trade-off in hypercellular networks with base station sleep control. Sun, Q., Jin, S., Chen, C.: Energy analysis of sensor nodes in WSN based on discrete-time queuing model with setup.
Delays in IP Routers, a Markov Model
Here we introduce the details of a Markov model previously reserved for simulation models: a real distribution of IP packets and self-similar nature of packet flows. To obtain numerical results, we use standard software: HyperStar [20] to approximate measured distributions with phase types, which allows the use of Markov chains and Prism [11] to study transient states of a complex Markov model .
2 Distribution of IP Packets
With this purely technical approach we are able to construct more realistic than existing models of IP queues and delays. 1. The influence of the complexity of the Markov model of a TCP packet size on the quality of the model.
3 Self-similar Traffic
4 Remarks on Buffer Occupation and Loss Probability
5 Numerical Solutions, Transient States, Network Dynamics
6 Response Time Distribution
Average response time as a function of time and steady state distribution of response time for hyper-Erlang representation of service time distribution,H. Average response time as a function of time and steady state distribution of response time for exponential service time distribution,H.
7 Conclusions
In: INFOCOM 1991, Proceedings of the Tenth Annual Joint Conference of the IEEE Computer and Communications Societies. Rataj, A.: More flexible models with a new version of the translator from Java sources to time machines J2TADD.
The Fluid Flow Approximation of the TCP Vegas and Reno Congestion Control Mechanism
TCP Vegas was the first attempt at a completely different approach to bandwidth management and is based on congestion detection before packet loss [3]. The paper [12] shows the stability and congestion control ability of TCP Vega, but in competition with the AIMD mechanism, it cannot fully utilize the available bandwidth.
2 Fluid Flow Model of TCP NewReno and Vegas Algorithms
Moreover, we also evaluated the friendliness and fairness of the different TCP variants, as well as their ability to use the available bandwidth in the presence of both standard FIFO queues with tail drop and active queue management (AQM) mechanisms in the routers. The fluid flow approximation models are presented in Section 2, while Section 3 discusses the comparison results.
