ERN: Neural Networks & Related Topics (Topic)最新文献

{"title":"A Discrete Version of CMA-ES","authors":"E. Benhamou, J. Atif, R. Laraki, A. Auger","doi":"10.2139/ssrn.3307212","DOIUrl":"https://doi.org/10.2139/ssrn.3307212","url":null,"abstract":"Modern machine learning uses more and more advanced optimization techniques to find optimal hyper parameters. Whenever the objective function is non-convex, non continuous and with potentially multiple local minima, standard gradient descent optimization methods fail. A last resource and very different method is to assume that the optimum(s), not necessarily unique, is/are distributed according to a distribution and iteratively to adapt the distribution according to tested points. These strategies originated in the early 1960s, named Evolution Strategy (ES) have culminated with the CMA-ES (Covariance Matrix Adaptation) ES. It relies on a multi variate normal distribution and is supposed to be state of the art for general optimization program. However, it is far from being optimal for discrete variables. In this paper, we extend the method to multivariate binomial correlated distributions. For such a distribution, we show that it shares similar features to the multi variate normal: independence and correlation is equivalent and correlation is efficiently modeled by interaction between different variables. We discuss this distribution in the framework of the exponential family. We prove that the model can estimate not only pairwise interactions among the two variables but also is capable of modeling higher order interactions. This allows creating a version of CMA ES that can accomodate efficiently discrete variables. We provide the corresponding algorithm and conclude.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"233 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132029283","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Forecasting Realized Volatility With Kernel Ridge Regression","authors":"B. LeBaron","doi":"10.2139/ssrn.3229272","DOIUrl":"https://doi.org/10.2139/ssrn.3229272","url":null,"abstract":"This paper explores a common machine learning tool, the kernel ridge regression, as applied to financial volatility forecasting. It is shown that kernel ridge provides reliable forecast improvements to both a linear specification, and a fitted nonlinear specification which represents well known empirical features from volatility modeling. Therefore, the kernel ridge specification is still finding some nonlinear improvements that are not part of the usual volatility modeling toolkit. Various diagnostics show it to be a reliable and useful tool. Finally, the results are applied in a dynamic volatility control trading strategy. The kernel ridge results again show improvements over linear modeling tools when applied to building a dynamic strategy.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134041659","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"K-Means Clustering Von Self-Organizing Maps: Eine Empirische Studie Zum Informationsgehalt Der Selbsteinstufung Von Hedge-Fonds (K-means Clustering of Self-Organizing Maps: An Empirical Study on the Information Content of the Self-Classification of Hedge Funds)","authors":"Marcus Deetz","doi":"10.2139/ssrn.3218895","DOIUrl":"https://doi.org/10.2139/ssrn.3218895","url":null,"abstract":"<b>German Abstract:</b> Mit der Implementierung des 2-stufigen Ansatzes nach VesantoAlhoniemi (2000) erweitert der vorliegende Artikel das in der Hedge-Fonds Literatur zur Klassifikation mit Self-Organizing Maps üblicherweise gewählte Vorgehen der visiuellen Auswertung der Kohonen-Karten und stellt damit ein automatisiertes Verfahren vor, welches eine konsistente Zusammenfassung benachbarter Output-Units und damit eine objektive Klassifizierung garantiert. Die empirische Anwendung dieses Verfahrens resultiert in einer Reduktion der durch die Datenbank vorgegebenen Strategiegruppen. Damit einher geht eine ebenfalls deutliche Reduzierung des Davies-Bouldin Indexes (DBI) der SOM-Partitionierungen. Da eine geringe Streuung innerhalb der Cluster und große Abstände zwischen den Clustern zu kleinen DBIs führen, ist eine Minimierung dieser Größe erwünscht. Diese signifikant bessere Partitionierung der SOMs gegenüber der auf Eigenangaben beruhenden Einteilung der Hedge-Fonds in das durch den Datenbankanbieter vorgegebene Kategorisierungsschema ist in allen untersuchten Datensamples (Robustheitsanalysen) zu beobachten. Letztendlich kann keine der originär 23 Strategiegruppen empirisch validiert werden. Ferner kann keine stabile Klassifizierung festgestellt werden. Sowohl die Anzahl der empirischen bestimmten Kategorien (SOM-Cluster) als auch die Zusammensetzungen dieser Cluster weichen in den untersuchten Subsamples stark voneinander ab. Damit bestätigt das vorliegende Resultate im Wesentlichen die Ergebnisse und Schlussfolgerungen in der Literatur, wonach die originäre, auf Selbsteinstufung beruhende Strategiebezeichnungen der Datenbankanbieter irreführend sind und somit regelmäßig keinen Informationsgehalt aufweisen. <b>English Abstract:</b> With the implementation of the 2-step approach according to Vesanto &amp; Alhoniemi (2000), this article extends the procedure of visual evaluation of the Kohonen Maps usually chosen in the hedge fund literature for classification with Self-Organizing Maps. It introduces an automated procedure which guarantees a consistent combination of adjacent output units and thus an objective classification. The empirical application of this method results in a reduction of the strategy groups specified by the database. This is also accompanied by a significant reduction in the Davies Bouldin Index (DBI) of the SOM partitions. Since a small dispersion within the clusters and large distances between the clusters lead to small DBIs, a minimization of this measure is desired. This significantly better partitioning of SOMs in comparison to the classification of hedge funds into the categorization scheme specified by the database provider can be observed in all examined data samples (robustness analyses). Ultimately, none of the original 23 strategy groups can be empirically validated. Furthermore, no stable classification can be found. Both the number of empirically determined categories (SOM clusters) and the composition of these c","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129570331","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Data Driven Neural Network Approach to Optimal Asset Allocation for Target Based Defined Contribution Pension Plans","authors":"Yuying Li, P. Forsyth","doi":"10.2139/ssrn.3192132","DOIUrl":"https://doi.org/10.2139/ssrn.3192132","url":null,"abstract":"A data driven Neural Network (NN) optimization framework is proposed to determine optimal asset allocation during the accumulation phase of a defined contribution pension scheme. In contrast to parametric model based solutions computed by a partial differential equation approach, the proposed computational framework can scale to high dimensional multi-asset problems. More importantly, the proposed approach can determine the optimal NN control directly from market returns, without assuming a particular parametric model for the return process. We validate the proposed NN learning solution by comparing the NN control to the optimal control determined by solution of the Hamilton-Jacobi-Bellman (HJB) equation. The HJB equation solution is based on a double exponential jump model calibrated to the historical market data. The NN control achieves nearly optimal performance. An alternative data driven approach (without the need of a parametric model) is based on using the historic bootstrap resampling data sets. Robustness is checked by training with a blocksize different from the test data. In both two and three asset cases, we compare performance of the NN controls directly learned from the market return sample paths and demonstrate that they always significantly outperform constant proportion strategies.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122188430","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Optimization of Constrained Liquidity Management","authors":"Maged S. Tawfik","doi":"10.2139/ssrn.3064047","DOIUrl":"https://doi.org/10.2139/ssrn.3064047","url":null,"abstract":"A consistent framework for optimal liquidity management is presented. This framework optimizes the cost of covering expected cashflow gaps without violating regulatory and business constraints. Anticipated economic value loss, cashflow loss, and adverse market impact are the major drivers of cost. The notion of a deployable liquidity resource, which is subsequently extend to the notion of a dated liquidity strategy, is introduced. A formalization of LCR as a typical regulatory constraints is presented and included in the formulation. The formulation includes a general arbitrage free market impact function. A decoupling between liquidity risk management and that of market and credit risks is assumed. Both linear and quadratic programming approaches for solving the resulting optimization problem are derived. This is followed by the introduction of a novel mapping algorithm which transforms the linear program to a network flow problem that is more efficiently solvable via the network simplex algorithm. Next an algorithm for generating a plausible starting point for the iterative optimization problem, is given. This is shown to be already optimal under the risk neutral measure. Finally, heuristics that can help speed up the satisfaction of regulatory constraints are discussed. Throughout the presentation attention is given to algorithmic complexity issues.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115093365","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Walled Buildings, Sustainability, and Housing Prices: An Artificial Neural Network Approach","authors":"R. Li, Ka Yi Cheng, M. Shoaib","doi":"10.3390/SU10041298","DOIUrl":"https://doi.org/10.3390/SU10041298","url":null,"abstract":"Various researchers have explored the adverse effects of walled buildings on human health. However, few of them have examined the relationship between walled buildings and private housing estates in Hong Kong. This study endeavors to fill the research gap by exploring the connections among walled-building effects, housing features, macroeconomic factors, and housing prices in private housing estates. Specifically, it reveals the relationship between walled buildings and housing prices. Eight privately owned housing estates are selected with a total of 11,365 observations. Results are analyzed to study the factors that affect the housing price. Firstly, unit root tests are carried out to evaluate if the time series variables follow the unit root process. Secondly, the relationship between walled buildings and housing price is examined by conducting an artificial neural network. We assumed that the housing price reduces due to walled-building effects, given that previous literature showed that heat island effect, and blockage of natural light and views, are common in walled-building districts. Moreover, we assume that housing price can also be affected by macroeconomic factors and housing features, and these effects vary among private housing estates. We also study these impacts by using the two models. Recommendations and possible solutions are suggested at the end of the research paper.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"02 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129537965","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Predicting Jump Arrivals in Stock Prices Using Neural Networks with Limit Order Book Data","authors":"Milla Mäkinen, Alexandros Iosifidis, M. Gabbouj, J. Kanniainen","doi":"10.2139/ssrn.3165408","DOIUrl":"https://doi.org/10.2139/ssrn.3165408","url":null,"abstract":"This paper proposes a new method for predicting jump arrivals in stock markets with high-frequency limit order book data. We introduce a new model architecture, based on Convolutional Long Short-Term Memory with attention, to apply time series representation learning with memory and to focus the prediction attention on the most important features to improve performance. Using order book data on five liquid U.S. stocks, we provide empirical evidence on the efficacy of the proposed approach. We find that the proposed approach with an attention mechanism outperforms the multi-layer perceptron network as well as the convolutional neural network and Long Short-Term memory model. The use of limit order book data was found to improve the performance of the proposed model in jump prediction, either clearly or marginally, depending on the underlying stock.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131202320","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Predicting Stock Returns Using Neural Networks","authors":"Murat Aydogdu","doi":"10.2139/ssrn.3141492","DOIUrl":"https://doi.org/10.2139/ssrn.3141492","url":null,"abstract":"A single hidden layer neural network can be trained to predict whether a stock will be in the top, middle, or bottom third of sample stocks based on its return over the next month based on return, trading volume, and volatility measures available at the end of this month. In my preliminary work using S&P 500 stocks, the network has limited success in predicting which stocks are likely to go up but the prediction strength is not strong enough to help build profitable portfolios. While neural networks have pushed artificial intelligence forward in many fields, and while the investment industry has been shifting more towards quantitative prediction using neural networks and other machine learning models, their place in empirical finance research has been limited. My work aims to contribute to this growing literature.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133228984","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Artificial Intelligence Can Make Our Jail System More Efficient, Equitable and Just","authors":"Arthur L. Rizer, Caleb Watney","doi":"10.2139/ssrn.3129576","DOIUrl":"https://doi.org/10.2139/ssrn.3129576","url":null,"abstract":"Artificial intelligence (AI), and algorithms more broadly, hold great promise for making our criminal justice system more efficient, equitable, and just. Many of these systems are already in place today, assisting with tasks such as risk assessment and case management. In the popular media, these tools have been compared to dystopian science-fiction scenarios run awry. But while these comparisons may succeed in luring readers, the reality of how AI is used in the criminal justice context-at least in its current form-is a bit more mundane. The courts are not at the precipice of replacing jurists with black-robed robots or arresting people before they commit a crime. However, there are real concerns about how effectively and transparently these systems operate, or how they might subtly distort outcomes, without adequate scrutiny. \u0000This article contends that AI can play a critical role in achieving fairer and more efficient pretrial and jail systems, in particular through risk assessment software. Unlike other applications of risk assessment AI, such as for sentencing or parole, pretrial applications have relatively simple goals, involve fewer complex legal questions, and have outcomes that are quicker and easier to measure. Thus, it is likely that the pretrial and jail stages will be the testbed for broader deployment of AI technology in the justice system. \u0000Of course, AI will not (and should not) supplant human judgment any time soon. A machine cannot yet read a defendant's demeanor or assess the full context of facts the way an experienced judge can. But AI can counter certain human biases and, if deployed in a transparent manner, can help advise judges in ways that will produce better outcomes-such as reduced crime rates and lower jail populations. \u0000This article will differentiate between the various types of algorithms and explain current capabilities, as well as give an overview of current pretrial and jail system trends. Next, we give a brief overview of the history of risk assessment tools, their current uses in the pretrial and jail systems, and the potential for further reform using more advanced algorithms. In addition, the article will discuss the relevant legal framework as well as governance capabilities across state, municipal, and federal jurisdictions. We then will attempt to consider the most prominent critiques of algorithms in the jail system, especially in risk assessment. Finally, the article will look at potential policy and legal solutions for the effective stewardship and deployment of algorithms in the pretrial and jail systems.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-02-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123612161","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Brain Information Optimization and Ethical Behavior","authors":"Emmanuel Chauvet","doi":"10.14704/NQ.2018.16.3.1158","DOIUrl":"https://doi.org/10.14704/NQ.2018.16.3.1158","url":null,"abstract":"Neural networks are tackled through probabilities for neurons to be activated by other neurons. They are represented by doubly stochastic matrices, named brain matrices, the polytope of which is the convex hull of the permutation matrices which are vertices of this Birkhoff polytope. Each permutation matrix enables to identify loops of neurons associated with a given neurotransmitter. The entropy of evolution of one network is defined and a short study of the optimal information transport in this network leads to consider two thresholds that give rise to questioning about the foundations of classical psychoanalysis within the construction of an extended and more realistic matrix of the neural network. A parallel is emphasized between the expansions in permutation matrices of the brain matrix and the quantum measurement theory through the collapse of the wave function. At a higher scale all the neural networks can be integrated in a global model that can be studied on the same ground as individual brain matrices or through specific thresholds in order to define the origins of ethical behaviors as well as what can lead to mental disability.","PeriodicalId":114865,"journal":{"name":"ERN: Neural Networks & Related Topics (Topic)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-11-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121005313","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}