We have previously described trading systems based on unsupervised learning approaches such as reinforcement learning and genetic algorithms which take as input a collection of commonly used technical indicators and g...
详细信息
ISBN:
(纸本)3540440259
We have previously described trading systems based on unsupervised learning approaches such as reinforcement learning and genetic algorithms which take as input a collection of commonly used technical indicators and generate profitable trading decisions from them. This article demonstrates the advantages of applying evolutionary algorithms to the reinforcement learning problem using a hybrid credit assignment approach. In earlier work, the temporal difference reinforcement learning approach suffered from problems with overfitting the in-sample data. This motivated the present approach. Technical analysis has been shown previously to have predictive value regarding future movements of foreign exchange prices and this article presents methods for automated high-frequency FX trading based on evolutionary reinforcement learning about signals from a variety of technical indicators. These methods are applied to CBPUSD, USDCHF and USDJPY exchange rates at various frequencies. Statistically significant profits are made consistently at transaction costs of up to 4bp for the hybrid system while the standard RL is only able to trade profitably up to about 1bp slippage per trade.
We present a general framework for interactive visualization and analysis of multi-dimensional data points. The proposed model is a hierarchical extension of the latent trait family of models developed in [4] as a gen...
详细信息
ISBN:
(纸本)3540440259
We present a general framework for interactive visualization and analysis of multi-dimensional data points. The proposed model is a hierarchical extension of the latent trait family of models developed in [4] as a generalization of GTM to noise models from the exponential family of distributions. As some members of the exponential family of distributions are suitable for modeling discrete observations, we give a brief example of using our methodology in interactive visualization and semantic discovery in a corpus of text-based documents. We also derive formulas for computing local magnification factors of latent trait projection manifolds.
Previous studies on frequent pattern discovery from temporal sequence mainly consider finding global patterns, where every record in a sequence contributes to support the patterns. In this paper, we present a novel pr...
详细信息
ISBN:
(纸本)3540440259
Previous studies on frequent pattern discovery from temporal sequence mainly consider finding global patterns, where every record in a sequence contributes to support the patterns. In this paper, we present a novel problem class that is the discovery of local sequential patterns, which only a subsequence of the original sequence exhibits. The problem has a two-dimensional solution space consisting of patterns and temporal features, therefore it is impractical that use traditional methods on this problem directly in terms of either time complexity or result validity. Our approach is to maintain a suffix-tree-like index to support efficiently locating and counting of local patterns. Based on the index, a method is proposed for discovering such patterns. We have analyzed the behavior of the problem and evaluated the performance of our algorithm on both synthetic and real data. The results correspond with the definition of our problem and verify the superiority of our method.
The predominate weakness in the creation of decision trees is the strict partitions which are selected by the induction algorithm. To overcome this problem the theories of fuzzy logic have been applied to generate sof...
详细信息
ISBN:
(纸本)3540440259
The predominate weakness in the creation of decision trees is the strict partitions which are selected by the induction algorithm. To overcome this problem the theories of fuzzy logic have been applied to generate soft thresholds leading to the creation of fuzzy decision trees, thus allowing cases passing through the tree for classification to be assigned partial memberships down all paths. A challenging task is how these resultant membership grades are combined to produce an overall outcome. A number of theoretical fuzzy inference techniques exist, yet they have not been applied extensively in practical situations and are often domain dependent. Thus the overall classification success of the fuzzy trees has a high dependency on the optimization of the strength of the fuzzy intersection and union operators that are applied. This paper investigates a new, more general approach to combining membership grades using neural-fuzzy inference. Comparisons are made between using the fuzzy-neural approach and the use of pure fuzzy inference trees.
Image retrieval by subjective content has been recently addressed by the Kansei engineering community in Japan. Such information retrieval systems aim to include subjective aspects of the users in the querying criteri...
详细信息
ISBN:
(纸本)3540440259
Image retrieval by subjective content has been recently addressed by the Kansei engineering community in Japan. Such information retrieval systems aim to include subjective aspects of the users in the querying criteria. While many techniques have been proposed in modeling such users' aspects, little attention has been placed on analyzing the amount of information involved in this modeling process and the multi-interpretation of such information. We propose a data warehouse as a support for the mining of the multimedia user feedback. A unique characteristic of our data warehouse lays in its ability to manage multiple hierarchical descriptions of images. Such characteristic is necessary to allow the mining of such data, not only at different levels of abstraction, but also according to multiple interpretation of their content. The proposed data warehouse has been used to support the adaptation of web-based image retrieval systems by impression words.
In this paper we propose the use of dominant point method for Chinese character recognition. We compare the performance of three classifiers on the same inputs;a statistical linear classifier, a machine learning C4.5 ...
详细信息
data cleaning is an important part of the knowledge discovery process. The principal causes of data anomalies include incomplete information, absence of a unique identifier across multiple databases, inconsistent data...
详细信息
ISBN:
(纸本)3540440259
data cleaning is an important part of the knowledge discovery process. The principal causes of data anomalies include incomplete information, absence of a unique identifier across multiple databases, inconsistent data, existence of data entry errors and logically incorrect data. This situation is further exacerbated while integrating data from multiple, disparate data sources. Since data quality is directly related with the quality of services in data-driven applications, such as medical informatics, a reliable data cleaning solution, which allows rapid and precise detection of invalid data, is needed. Most existing data cleaning solutions are domain specific, time-consuming and do not easily accommodate logical validations. In this paper, we propose a Fuzzy rule-based framework, which is domain independent, flexible and easily accommodates physical as well as logical validations. We have implemented existing cleaning strategies (i.e. Sorted Neighborhood Method), and enhanced them by using state-of-the-art algorithms (i.e. Rete, Bigram). As proof-of-concept, our prototype system was applied to real patient data. Experimental results illustrate that our framework is extensible and allows rapid detection of invalid data with high precision.
作者：
Wright, W. Andy
Sowerby Building PO Box 5 Bristol United Kingdom
An alternative approach to learning decision strategies in multi-state multiple agent systems is presented here. The method, which uses a game theoretic construction which is model free and does not rely on direct com...
详细信息
Viewed as a promising application of neural networks, financial time series forecasting was studied in the literature of neural nets and machine learning. The recently developed Temporal Factor Analysis (TFA) model ma...
详细信息
The purpose of a classification algorithm is to predict the class label of a new instance based on the analysis of a training dataset. Many classification algorithms work most naturally with nominal attributes. Howeve...
详细信息
暂无评论