Vassil Alexandrov

Parallel genetic algorithms for stock market trading rules

Abstract Finding the best trading rules is a well-known problem in the field of technical analysis of stock markets. One option is to employ genetic algorithms, as they offer valuable characteristics towards retrieving a “good enough” solution in a timely manner. However, depending on the problem size, their application might not be a viable option as the iterative search through a multitude of possible solutions does take considerable time. Even more so if a variety of stocks are to be analysed.In this paper we concentrate on the enhancement of a previously published genetic algorithm for the optimisation of technical trading rules, using example data from the Madrid Stock Exchange General Index (IGBM).

On scalability behaviour of Monte Carlo sparse approximate inverse for matrix computations

This paper presents a Monte Carlo SPAI pre-conditioner. In contrast to the standard deterministic SPAI pre-conditioners that use the Frobenius norm, a Monte Carlo alternative that relies on the use of Markov Chain Monte Carlo (MCMC) methods to compute a rough matrix inverse (MI) is given. Monte Carlo methods enable a quick rough estimate of the non-zero elements of the inverse matrix with a given precision and certain probability. The advantage of this method is that the same approach is applied to sparse and dense matrices and that complexity of the Monte Carlo matrix inversion is linear of the size of the matrix. The behaviour of the proposed algorithm is studied, its performance is investigated and a comparison with the standard deterministic SPAI, as well as the optimized and parallel MSPAI version is made. Further Monte Carlo SPAI and MSPAI are used for solving systems of linear algebraic equations (SLAE) using BiCGSTAB and a comparison of the results is made.

Parallel Regularized Multiple-criteria Linear Programming

In this paper, we proposed a new parallel algorithm: Parallel Regularized Multiple-Criteria Linear Programming (PRMCLP) to overcome the computing and storage requirements increased rapidly with the number of training samples. Firstly, we convert RMCLP model into a unconstrained optimization problem, and then split it into several parts, and each part is computed by a single processor. After that, we analyze each parts result for next cycle going. By doing this, we are be able to obtain the final optimization solution of the whole classification problem. All experiments in public datasets show that our method greatly increases the training speed of RMCLP in the help of multiple processors.

Towards scalable mathematics and scalable algorithms for extreme scale computing

ScalA Workshop series is a premier forum for presentation of esearch results on novel scalable scientific algorithms needed to nable key science applications and to exploit the computational ower of large-scale systems. This is especially true for the curent tier of leading petascale machines and the road to exascale omputing as HPC systems continue to scale up in compute node nd processor core count. These extreme scale systems require ovel scientific algorithms to hide network and memory latency, ave very high computation/communication overlap, have minial communication, have no synchronization points. Although this special issue focuses on selected papers from the calable Algorithms for Large Scale Systems Workshop (ScalA’11), eld at Supercomputing 2011, as well as selected papers of some of he keynotes of ScalA’10 workshop, the author wants to draw the ttention to the wider picture and the strategic issues it addresses in erms of the novel scalable mathematical methods and algorithms eeded for extreme-scale computing. In fact on the very first ScalA orkshop [1] the participants and the experts involved highlighted hat in order to achieve overall efficient scalability, scalability at all hree levels is required:

Facilitating analysis of Monte Carlo dense matrix inversion algorithm scaling behaviour through simulation

Abstract With the latest developments in the area of advanced computer architectures, we are already seeing large-scale machines at petascale level and are faced with the exascale computing challenge. All these require scalability at system, algorithmic and mathematical model levels. In particular, efficient scalable algorithms are required to bridge the performance gap. Being able to predict application demeanour, performance and scalability of currently used software on new supercomputers of different architectures, varying sizes, and utilising distinct ways of intercommunication, can be of great benefit for researchers as well as application developers. This paper is concerned with scaling characteristics of Monte Carlo based algorithms for matrix inversion. The algorithmic behaviour on both, a shared memory and a large-scale cluster system will be predicted with the help of an extreme-scale high-performance computing (HPC) simulator.

Route to exascale: Novel mathematical methods, scalable algorithms and Computational Science skills

Abstract This editorial outlines the research context, the needs and challenges on the route to exascale. In particular the focus is on novel mathematical methods and mathematical modeling approaches together with scalable scientific algorithms that are needed to enable key science applications at extreme-scale. This is especially true as HPC systems continue to scale up in compute node and processor core count. These extreme-scale systems require novel mathematical methods to be developed that lead to scalable scientific algorithms to hide network and memory latency, have very high computation/communication overlap, have minimal communication, have fewer synchronization points. It stresses the need of scalability at all levels, starting from mathematical methods level through algorithmic level, and down to systems level in order to achieve overall scalability. It also points out that with the advances of Data Science in the past few years the need of such scalable mathematical methods and algorithms able to handle data and compute intensive applications at scale becomes even more important. The papers in the special issue are selected to address one or several key challenges on the route to exascale.

Scalable Stochastic and Hybrid Methods and Algorithms for Extreme Scale Computing

Abstract Novel mathematics and mathematical modelling approaches together with scalable algorithms are needed to enable key applications at extreme-scale. This is especially true as HPC systems continue to scale up in compute node and processor core count. At the moment computational scientists are at the critical point/threshold of novel mathematics development as well as large-scale algorithm development and re-design and implementation that will affect most of the application areas. Thus the paper will focus on the mathematical and algorithmic challenges and approaches towards exascale and beyond and in particular on stochastic and hybrid methods that in turn lead to scalable scientific algorithms with minimal or no global communication, hiding network and memory latency, have very high computation/communication overlap, have no synchronization points.

A framework for parallel genetic algorithms for distributed memory architectures

Genetic algorithms are metaheuristic search methods, based on the principles of biological evolution and genetics. Through a heuristic search they are able to find good solutions in acceptable time. However, with the increase of the complexity of the fitness landscape and the size of the search space their runtime increases rapidly. Using parallel implementations of genetic algorithms in order to harness the power of modern computational platforms, is a powerful approach to mitigating this issue. In this paper several parallel implementations ranging from MPI to hybrid MPI/OpenMP and MPI/OmpSs are made. These implementations are optimized for execution on tightly coupled distributed memory systems. We address issues that arise when running a distributed genetic algorithm and present an adaptive migration scheme. Comparison of their efficiency is also made.

Quantifying Uncertainty in Phylogenetic Studies of the Slavonic Languages

Abstract We describe the application of Bayesian methods to accommodate the uncertainty problem in phylogenetic reconstruction with an example of the Slavonic languages family. Comparative studies of languages have lots in common with evolutionary biology studies. Stable linguistic characters (e.g. word forms from the basic vocabulary, grammar characters) can be used to construct DNA-like sequences that the phylogenetic reconstruction methods can then be applied to. Linguistic data is known to be a subject of noise and error of different kinds causing conflicting signals and uncertainty within a phylogeny. Bayesian methods help to quantify the uncertainty. The comparison with the Damerau-Levenshtein distance-based tree is also given.

Investigating scaling behaviour of monte carlo codes for dense matrix inversion

With the latest developments in the area of advanced computer architectures, we are already seeing large-scale machines at petascale level and are faced with the exascale computing challenge. All these require scalability at system, algorithmic and mathematical model level. In particular, efficient scalable algorithms are required to bridge the performance gap. Being able to predict application demeanour, performance and scalability of currently used software on new supercomputers of different architectures, varying sizes, and utilising alternative ways of intercommunication, can be of great benefit for researchers as well as application developers. This paper is concerned with scaling characteristics of Monte Carlo based algorithms for matrix inversion. The algorithmic behaviour on large-scale systems will be predicted with the help of an extreme-scale high-performance computing (HPC) simulator.