The aim of the thesis was to create a tool for time-series prediction based on deep learning. The first part of the work is a brief description of deep learning and its comparison to classical machine learning. In the next section contains brief analysis of some tools, that are already used for time-series forecasting. The last part is focused on the analysis of the problem as well as on the actual creation of the program.

The diploma thesis deals with the financial analysis of the selected company. On the analysis base of ratio financial indicators, time series, regression analysis and with using of the Maple system, the past and actual financial situation have been evaluated and future potential development of the company has been predicted.

Diploma thesis deals with the financial analysis of a banking entity operating in Czech Republic and the analysis of time series of selected indicators. The first part gives methodology of the work and theoretical bases for processing financial analysis and time series. The second practical part consists of analysis itself. Based on the results, the state of banking company is evaluated in the final part and the possibilities and proposal are introduced for entity operating in the banking market.

博士（商学） / 甲第544号 / 3, 175p / Hitotsubashi University(一橋大学)

The American high school schedule of single-period classes has remained mostly unchanged for over one hundred years. In response to societal changes and reform movements, the secondary school schedule is receiving renewed attention. Block scheduling, the use of extended periods of time for learning, is one response to school restructuring in Virginia and throughout the nation. In Virginia, the 7A/B block schedule is used by 23.3% of the high schools. Although advocates have convinced school boards to adopt this schedule, there is little hard data available to assess its efficacy. In this study the relationship between two types of schedules(7A/B block and 7-period traditional) and student achievement at the eleventh grade was examined. No differences were found between the two schedules for achievement as measured by the subscales of the eleventh grade Tests of Achievement and Proficiency. / Ed. D.

Karst aquifers are productive groundwater systems around the world, supplying approximately 25% of the world's drinking water. However, they are highly vulnerable to contamination due to rapid groundwater transit in the transmission zone (KWI 2006). The epikarst, also known as the subcutaneous zone, is an interface between the soil overburden and the transmission zone. The epikarst is considered a critical zone as it can control hydrologic and geochemical characteristics of recharge to the underlying karst aquifer.  The overall goal of this thesis is to utilize time series hydrologic and geochemical data collected at James Cave, Virginia, to examine the influence of epikarst on the quantity, quality, and rates of recharge to aquifers in Appalachian karst. Results of this study indicate a strong seasonality of both the hydrology and geochemistry of recharge. The conceptual model of the epikarst developed in this study identifies three hydrologic seasons: recharge, recession, and baseflow. Seasonality of recharge geochemistry coincides with these three hydrologic seasons.  These results have implications for management of karst aquifers.  First, recharge to Appalachian karst aquifers is seasonal, reaching a maximum during the winter-early spring; the onset of recharge depends on antecedent climatic conditions.  Second, water that infiltrates into the epikarst will have seasonally variable residence times due to changes in hydrologic storage; these variations in attenuation affect geochemical reactions in the epikarst, which can influence recharge quality. Overall, these results point to the complex influence of epikarst on karst recharge, which necessitates collection of long-term and high resolution datasets. / Master of Science

Thesis (Sc. D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering, 1956. / "June, 1956." / Includes bibliographical references (leaf 113). / Introduction: A physically realizable nonlinear system, like a linear one, is a system whose present output is a function of the past of its input. We may regard the system as a computer that operates on the past of one time function to yield the present value of another time function. Mathematically we say that the system performs a transformation on the past of its input to yield its present output. When this transformation is linear (the case of linear systems) we can take advantage of the familiar convolution integral to obtain the present output from the past of the input and the system is said to be characterized by its response to an impulse. That is, the response of a linear system to an impulse is sufficient to determine its response to any input. When the transformation is nonlinear we no longer have a simple relation like the convolution integral relating the output to the past of the input and the system can no longer be characterized by its response to an impulse since superposition does not apply. Wiener has shown, however, that we can characterize a nonlinear system by a set of coefficients and that these coefficients can be determined from a knowledge of the response of the system to shot noise excitation. Thus, shot noise occupies the same position as a probe for investigating nonlinear systems that the impulse occupies as a probe for investigating linear systems. The first section of this thesis is devoted to the Wiener theory of nonlinear system characterization. Emphasis is placed on important concepts of this theory that are used in succeeding chapters to develop a theory for determining optimum nonlinear systems. / by Amar Gopal Bose. / Sc.D.

Time series are very common in presenting collected data such as economic indicators, natural phenomenon, control engineering data, among others. In the last decade, the interest in time series data mining increased as the amount of collected data increased dramatically. Standard approaches for time series classification are based on collecting distance measures, such as the Euclidian distance (ED) and dynamic time warping (DTW) along with 1-NN classifier for further classification. Recently, more advanced types of classification were found, introducing primitives (named time series shapelet) that consistently represent a certain class. The time series shapelet is a small sub-section of the entire time series, which is “particularly discriminating”. It appears that shapelets based classification produces higher accuracies on some data sets, based on the fact that the global features are more sensitive to noise than locals. Despite its advantages, the time series shapelets classification has an apparent disadvantage: very slow training time. This work attempts to improve the training time for the originally proposed time series shapelets classification algorithm and introduces a new approach for time series classification based on concatenated decision tree paths. First, the classical algorithm for time series classification based on shapelets, is significantly improved in terms of the training time. The improvement is based on using randomly generated sequences tuned in a particle-swarm-optimization (PSO) environment, instead of using sub-series from the original time series. Second, a new highly accurate classification method, based on concatenated decision tree paths, is introduced. The approach builds a unique representative pattern of a certain class based on the taken paths in a pool of decision trees. Third, the proposed method has been successfully extended for a 2-class-labels classification problem where only one decision tree can be built. A variety of 2-class-labels decision trees were built based on different splitting criterion (distance to a random shapelet); thus- increasing the pool of decision trees and increasing the overall accuracy. Fourth, the proposed method was successfully applied on two classes image classification problem, by converting the image into time series. An accuracy of around 95% was achieved for the pedestrian detection case from the Daimler database.

No description available.

In many applications involving data collected over time, it is important to get timely estimates and adjustments of the parameters associated with a dynamic model. When the dynamics of the model must be updated, time and computational simplicity are important issues. When the dynamic system is not linear the problem of adaptation and response to feedback are exacerbated. A linear approximation of the process at various levels or “states” may approximate the non-linear system. In this case the approximation is linear within a state and transitions from state to state over time. The transition probabilities are parametrized as a Markov chain, and the within-state dynamics are modeled by an AR time series model. However, in order to make the estimates available almost instantaneously, least squares and weighted least squares estimates are used. This is a modification of the cluster-weighted models proposed by Gershenfeld, Schoner, and Metois (1999). A simulation study compares the models and explores the adequacy of least squares estimators.