Publikační činnost Katedry systémového inženýrství / Publications of Department of System Engineering (157)
http://hdl.handle.net/10084/100633
Kolekce obsahuje bibliografické záznamy publikační činnosti (článků) akademických pracovníků Katedry systémového inženýrství (157) v časopisech registrovaných ve Web of Science od roku 2003 po současnost.2018-09-21T00:52:03ZAn LP-based hyperparameter optimization model for language modeling
http://hdl.handle.net/10084/127003
An LP-based hyperparameter optimization model for language modeling
Rahnama, Amir Hossein Akhavan; Toloo, Mehdi; Zaidenberg, Nezer Jacob
In order to find hyperparameters for a machine learning model, algorithms such as grid search or random search are used over the space of possible values of the models' hyperparameters. These search algorithms opt the solution that minimizes a specific cost function. In language models, perplexity is one of the most popular cost functions. In this study, we propose a fractional nonlinear programming model that finds the optimal perplexity value. The special structure of the model allows us to approximate it by a linear programming model that can be solved using the well-known simplex algorithm. To the best of our knowledge, this is the first attempt to use optimization techniques to find perplexity values in the language modeling literature. We apply our model to find hyperparameters of a language model and compare it to the grid search algorithm. Furthermore, we illustrate that it results in lower perplexity values. We perform this experiment on a real-world dataset from SwiftKey to validate our proposed approach.
2018-01-01T00:00:00ZA simplification generalized returns to scale approach for selecting performance measures in data envelopment analysis
http://hdl.handle.net/10084/126099
A simplification generalized returns to scale approach for selecting performance measures in data envelopment analysis
Toloo, Mehdi; Allahyar, Maryam
Toloo and Tichy (2015) with the aim of holding the rule of thumb in data envelopment analysis, developed a pair of models which optimally chooses some inputs and outputs among selective measures under variable returns to scale assumption. Their approach involves a lower bound for the input and output weights in the multiplier model and a penalty term in the objective function of envelopment model. These models possess an epsilon which on the one hand turns the selecting envelopment model non-linear and on the other hand increases the required computational burden for solving the selecting multiplier models. Selecting an improper value for the epsilon may cause the infeasibility and unboundedness issues for the multiplier and envelopment model, respectively. This paper demonstrates that the method of Toloo and Tichy (2015) is valid even with excluding the epsilon. The method is extended to generalized returns to scale model which considers other returns to scale assumptions, i.e. non-increasing, constant, and non-decreasing. The obtained results point out that the simplified approach is more stable and more reliable and substantially reduces the required calculations.
2018-01-01T00:00:00ZCosts of quality or quality costs
http://hdl.handle.net/10084/125822
Costs of quality or quality costs
Řeháček, Petr
Costs of quality or quality costs do not mean the use of expensive or very highly quality materials to manufacture a product. The term refers to the costs that are incurred to prevent, detect and remove defects from products. There are four categories: internal failure costs (costs associated with defects found before the customer receives the product or service), external failure costs (costs associated with defects found after the customer receives the product or service), appraisal costs (costs incurred to determine the degree of conformance to quality requirements) and prevention costs (costs incurred to keep failure and appraisal costs to a minimum). Cost of quality is a methodology that allows an organization to determine the extent to which its resources are used for activities that prevent poor quality, that appraise the quality of the organization's products or services, and that result from internal and external failures. Having such information allows an organization to determine the potential savings to be gained by implementing process improvements.
2018-01-01T00:00:00ZIntegrated data envelopment analysis: Linear vs. nonlinear model
http://hdl.handle.net/10084/125753
Integrated data envelopment analysis: Linear vs. nonlinear model
Mahdiloo, Mahdi; Toloo, Mehdi; Thach-Thao Duong; Saen, Reza Farzipoor; Tatham, Peter
This paper develops a relationship between two linear and nonlinear data envelopment analysis (DEA) models which have previously been developed for the joint measurement of the efficiency and effectiveness of decision making units (DMUs). It will be shown that a DMU is overall efficient by the nonlinear model if and only if it is overall efficient by the linear model. We will compare these two models and demonstrate that the linear model is an efficient alternative algorithm for the nonlinear model. We will also show that the linear model is more computationally efficient than the nonlinear model , it does not have the potential estimation error of the heuristic search procedure used in the nonlinear model and it determines global optimum solutions rather than the local optimum. Using 11 different data sets from published papers and also 1000 simulated sets of data we will explore and compare these two models. Using the data set that is most frequently used in the published papers it is shown that the nonlinear mosel with a step size equal to 0.00001 reauires running 1,955,573 linear problems (LPs) to measure the efficiency of 24 DMUs compared to only 24 LPs required for the linear model. Similarly for a very small data set which consists of only 5 DMUs the nonlinear model requires running 7861 LPs with step size equal to 0.0001 whereas the linear model needs just 5 LPs.
2018-01-01T00:00:00Z