Dimitrios Tselios

Multiple project portfolio scheduling using recurrent neural networks

This work is devoted to the portfolio project management problem and more precisely it is focused on IT portfolios’ management. The problem is modelled as a multi-purpose job shop problem. Contemporary organisations such as IT companies define a careful planning to perform a set of projects that share common resources. These projects must, therefore, be handled concurrently. In this study we reviewed the literature and developed a system based on a multi-objective problem model. We use recurrent neural network (RNN) technique to look for optimal or near-optimal solutions to the derived optimisation problem. We first apply a greedy algorithm to find an initial feasible solution. This solution is then fed to the RNN. We implemented our approach and evaluated it on some well known benchmarks. The experimental results obtained were very promising.

The Weighted Tardiness as Objective Function of a RNN Model for the Job Scheduling Problem

This paper proposes a Neural Network approach for the project portfolio management problem. The modern organizations such as the IT firms schedule and perform a set of projects that share common rare resources. Therefore, each IT organization develops a set of IT projects and it has to execute them simultaneously. In this work we reviewed the literature and extended a multi-objective system model based on the job shop scheduling problem modelling and expressed it as recurrent neural network. Moreover, we produced an example within its neural network that is focused on the Weighted Tardiness objective function. In addition, we use an initial solution by amending a greedy algorithm that has been proposed in a previous work for the Makespan objective function.

Parallelizing DBSCaN Algorithm Using MPI

The last years, huge bundles of information are extracted by computational systems and electronic devices. To exploit the derived amount of data, new innovative algorithms must be employed or the established ones maybe changed. One of the most fascinating and productive techniques, in order to locate and extract information from data repositories is clustering, and DBSCAN is a successful density based algorithm which clusters data according its characteristics. However, its main disadvantage is its severe computational complexity which proves the technique very inadequate to apply on big datasets. Although DBSCAN is a very well studied technique, a fully operational parallel version of it, has not been accepted yet by the scientific community. In this work, a three phase parallel version of DBSCAN is presented. The obtained experimental results are very promising and prove the correctness, the scalability, and the effectiveness of the technique.

Combining Distributed and Multi-core Programming Techniques to Increase the Performance of K-Means Algorithm

The last years, huge masses of data are produced or extracted by computational systems and independent electronic devices. To exploit this resource, novel methods must be employed or the established ones may be altered in order to confront the issues that arise. One of the most fruitful techniques, in order to locate and use information from data sources is clustering, and k-means is a successful representative algorithm which clustersdata according specific characteristics. However, its main disadvantage is the computational complexity which proves the techniques very unproductive to apply on big datasets. Although k-means is a very well studied technique, a fullyoperational distributed version combining the multi-core power of today machines, have not been accepted yet by the scientific community. In this work, a three phase distributed / multi-core version of k-means is presented. The obtained experimental results are very promising and prove the correctness, the scalability, and the effectiveness of the proposed technique.

Phased Method for Solving Multi-Objective MPM Job Shop Scheduling Problem

The project portfolio scheduling problem has become very popular in recent years since many modern organizations operate in multi-project and multi-objective environment. Current project oriented organizations have to design a plan in order to execute a set of projects sharing common resources such as personnel teams. This problem can be seen as an extension of the job shop scheduling problem; the multi-purpose job shop scheduling problem. In this paper, the authors propose a hybrid approach to deal with a bi-objective optimisation problem; Makespan and Total Weighted Tardiness. The approach consists of three phases; in the first phase they utilise a Genetic Algorithm (GA) to generate a set of initial solutions, which are used as inputs to recurrent neural networks (RNNs) in the second phase. In the third phase the authors apply adaptive learning rate and a Tabu Search like algorithm with the view to improve the solutions returned by the RNNs. The proposed hybrid approach is evaluated on some well-known benchmarks and the experimental results are very promising. KeywORdS Adaptive Learning, Genetic Algorithm, Job Shop Scheduling, Multi-Objective, Project Scheduling, Recurrent Neural Network

Analysis of the results obtained by intelligent method for project portfolio scheduling

A key factor at the project portfolio scheduling is about aiming at a pool of projects due the multi-objective environment in which modern organizations operate. Recent works proposed multi-objective models that are close to real world projects and they are based on intelligent methods. This work is an evaluation of this approach. More specifically, selected research questions are statistically tested, while some of them have been answered positively, proving the performance of the overall solving method.

Integrated intelligent method for solving multi-objective MPM job shop scheduling problem

The project portfolio scheduling problem has become very popular in recent years. Current project oriented organisations have to design a plan in order to execute a set of projects sharing common resources such as personnel teams. These projects must, therefore, be handled concurrently. This problem can be seen as an extension of the job shop scheduling problem; the multi-purpose job shop scheduling problem. In this paper, we propose a hybrid approach to deal with a bi-objective optimisation problem; Makespan and Total Weighted Tardiness. The approach consists of three phases; in the first phase we utilise a Genetic Algorithm (GA) to generate a set of initial solutions, which are used as inputs to recurrent neural networks (RNNs) in the second phase. In the third phase we apply adaptive learning rate and a Tabu Search like algorithm with the view to improve the solutions returned by the RNNs. The proposed hybrid approach is evaluated on some well-known benchmarks and the experimental results are very promising.

RNN Modelling for Bi-Objective MPM Job Shop Scheduling Problem

This paper presents a Recurrent Neural Network approach for the multipurpose machines Job Shop Scheduling Problem. This case of JSSP can be utilized for the modelling of project portfolio management besides the well known adoption in factory environment. Therefore, each project oriented organization develops a set of projects and it has to schedule them as a whole. In this work, we extended a bi-objective system model based on the JSSP modelling and formulated it as a combination of two recurrent neural networks. In addition, we designed an example within its neural networks that are focused on the Makespan and the Total Weighted Tardiness objectives. Moreover, we present the findings of our approach using a set of well known benchmark instances and the discussion about them and the singularity that arises.