检索结果-内蒙古大学图书馆

Building Uncertainty Models on Top of Black-Box Predictive APIs

IEEE ACCESS 2020年 8卷 121344-121356页

作者： Brando, Axel Torres-Latorre, Clara Rodriguez-Serrano, Jose A. Vitria, Jordi Univ Barcelona UB Dept Matemat & Informat Barcelona 08007 Spain BBVA Data & Analyt Madrid 28050 Spain

With the commoditization of machine learning, more and more off-the-shelf models are available as part of code libraries or cloud services. Typically, data scientists and other users apply these models as "black boxes" within larger projects. In the case of regressing a scalar quantity, such APIs typically offer a predict() function, which outputs the estimated target variable (often referred to as Oy or, in code, y_hat). However, many real-world problems may require some sort of deviation interval or uncertainty score rather than a single point-wise estimate. In other words, a mechanism is needed with which to answer the question "How confident is the system about that prediction?" Motivated by the lack of this characteristic in most predictive APIs designed for regression purposes, we propose a method that adds an uncertainty score to every black-box prediction. Since the underlying model is not accessible, and therefore standard Bayesian approaches are not applicable, we adopt an empirical approach and fit an uncertainty model using a labelled dataset (x, y) and the outputs (y) over cap of the black box. In order to be able to use any predictive system as a black box and adapt to its complex behaviours, we propose three variants of an uncertainty model based on deep networks. The first adds a heteroscedastic noise component to the black-box output, the second predicts the residuals of the black box, and the third performs quantile regression using deep networks. Experiments using real financial data that contain an in-production black-box system and two public datasets (energy forecasting and biology responses) illustrate and quantify how uncertainty scores can be added to black-box outputs.

关键词： Aleatoric uncertainty deep learning neural networks regression problems

来源：评论

学校读者我要写书评

暂无评论

Uncertainty Estimation for Black-Box Classification Models: A Use Case for Sentiment Analysis 9th

Uncertainty Estimation for Black-Box Classification Models: ...

引用

9th Iberian Conference on Pattern Recognition and Image Analysis (IbPRIA)

作者： Mena, Jose Brando, Axel Pujol, Oriol Vitria, Jordi Eurecat Ctr Tecnol Catalunya Barcelona Spain BBVA Analyt Data & Analyt Madrid Spain Univ Barcelona Barcelona Spain

ISBN: (纸本)9783030313326;9783030313319

With the advent of new pre-trained word embedding models like ELMO, GPT or BERT, that leverage transfer-learning to deliver high-quality prediction systems, natural language processing (NLP) methods are reaching or even overtaking human baselines in some applications. The basic principle of these successful models is to train a model to solve a given NLP task, mainly Language Modelling, using significant volumes of data like the whole Wikipedia. The model is then fine-tuned to solve another NLP task, requiring fewer domain-specific data to achieve state-of-the-art accuracies. The method proposed in the present work assists the practitioner in evaluating the quality of the transferred classification models when applied to new data domains. In this case, we consider the original model as a black box. No matter how complex the original model may be, the method only requires access to the output layer to train a measure of the uncertainty associated with the predictions of the original model. This measure of uncertainty is a measure of how well the black-box model accommodates to the new data. Later on, we show how a rejection system can use this uncertainty to improve its accuracy, effectively enabling the practitioner to find the best trade-off between the quality of the model and the number of rejected cases.

关键词： Sentiment analysis Transfer learning Uncertainty estimation Natural Language Processing

来源：评论

学校读者我要写书评

暂无评论

Sampling Unknown Decision Functions to Build Classifier Copies 1

引用

17th International Conference on Modeling Decisions for Artificial Intelligence (MDAI)

作者： Unceta, Irene Palacios, Diego Nin, Jordi Pujol, Oriol BBVA Data & Analyt Barcelona Spain Univ Ramon Llull ESADE Barcelona Spain Univ Barcelona Barcelona Spain

ISBN: (数字)9783030575243

ISBN: (纸本)9783030575243;9783030575236

Copies have been proposed as a viable alternative to endow machine learning models with properties and features that adapt them to changing needs. A fundamental step of the copying process is generating an unlabelled set of points to explore the decision behavior of the targeted classifier throughout the input space. In this article we propose two sampling strategies to produce such sets. We validate them in six well-known problems and compare them with two standard methods in terms of both their accuracy performance and their computational cost.

关键词： Bayesian sampling Classification Copies Synthetic data

来源：评论

学校读者我要写书评

暂无评论

Machine Learning in Finance 21

Machine Learning in Finance

引用

27th ACM SIGKDD International Conference on Knowledge Discovery and data Mining (KDD)

作者： Kumar, Senthil Akoglu, Leman Chawla, Nitesh Rodriguez-Serrano, Jose A. Faruquie, Tanveer Nagrecha, Saurabh Capital One Mclean VA 22102 USA Carnegie Mellon Univ Pittsburgh PA 15213 USA Univ Notre Dame Notre Dame IN 46556 USA BBVA Data & Analyt Madrid Spain eBay San Jose CA USA

ISBN: (纸本)9781450383325

The finance industry is constantly faced with an ever evolving set of challenges including credit card fraud, identity theft, network intrusion, money laundering, human trafficking, and illegal sales of firearms. There are also newly emerging threats such as fake news in financial media that can lead to distortions in trading strategies and investment decisions. In addition, traditional problems such as customer analytics, forecasting, and recommendations take on a unique flavor when applied to financial data. A number of new ideas are emerging to tackle all these problems including semi-supervised learning methods, deep learning algorithms, network/graph based solutions as well as linguistic approaches. These methods must often be able to work in real-time and be able handle large volumes of data. The purpose of this workshop is to bring together researchers and practitioners to discuss both the problems faced by the financial industry and potential solutions. We have invited regular papers, positional papers and extended abstracts of work in progress. We have also encouraged short papers from financial industry practitioners that introduce domain specific problems and challenges to academic researchers. This event is the fourth in a sequence of finance related workshops we have organized at KDD since 2017.

关键词： Finance Fraud Fairness in lending Financial graphs Early detection of emerging phenomenon Forecasting

来源：评论

学校读者我要写书评

暂无评论

Determining the usual environment of cardholders as a key factor to measure the evolution of domestic tourism

引用

INFORMATION TECHNOLOGY & TOURISM 2019年第1期21卷 23-43页

作者： Romero Palop, Juan de Dios Murillo Arias, Juan Bodas-Sagi, Diego J. Valero Lapaz, Heribert BBVA Data & Analyt Madrid Spain

Domestic tourism is harder to analyse compared to international tourism due to its smaller data footprint generation, as most times private means of transport are used, no border is crossed, and no lodging is registered. Digital data sources can be a useful, but still underused, complement to official survey-based statistics to fill this lack of reliable information. The present paper covers a research gap in the use of card transactions data (on site payments and cash withdrawals) to provide an innovative methodology to enhance vision on domestic tourism dynamics. The chosen approach is based on the United Nations World Tourism Organization definition of 'usual environment': "the geographical area (though not necessarily a contiguous one) within which an individual conducts his/her regular life routines" Upon this premise, a methodology has been developed in order to use transactional footprints of cardholders to delineate their usual environment, and subsequently to classify transactions as 'touristic' or 'non-touristic'. So as to ensure scalability, the resulting procedure is non- territory reliant, and can therefore be adapted to different geographies by varying one single parameter. Some practical applications are described in Sect. 5 through two use cases carried out in Spain and Mexico by bbva.

关键词： Usual environment Big data Domestic tourism Digital footprint Payments Applied data science

来源：评论

学校读者我要写书评

暂无评论

Copying Machine Learning Classifiers

引用

IEEE ACCESS 2020年 8卷 160268-160284页

作者： Unceta, Irene Nin, Jordi Pujol, Oriol BBVA Data & Analyt Barcelona 08018 Spain Univ Barcelona Dept Math & Comp Sci Barcelona 08007 Spain Univ Ramon Llull ESADE Dept Operat Innovat & Data Sci Sant Cugat Del Valles 08172 Spain

We study copying of machine learning classifiers, an agnostic technique to replicate the decision behavior of any classifier. We develop the theory behind the problem of copying, highlighting its properties, and propose a framework to copy the decision behavior of any classifier using no prior knowledge of its parameters or training data distribution. We validate this framework through extensive experiments using data from a series of well-known problems. To further validate this concept, we use three different use cases where desiderata such as interpretability, fairness or productivization constrains need to be addressed. Results show that copies can be exploited to enhance existing solutions and improve them adding new features and characteristics.

关键词： Training data Machine learning data models Training Buildings Production Predictive models Applied machine learning classification copying differential replication fidelity

来源：评论

学校读者我要写书评

暂无评论

Reinforcement learning for pricing strategy optimization in the insurance industry

引用

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE 2019年 80卷 8-19页

作者： Krasheninnikova, Elena Garcia, Javier Maestre, Roberto Fernandez, Fernando BBVA Data & Analyt Madrid Spain Univ Carlos III Madrid Getafe Spain

Pricing is a fundamental problem in the banking sector, and is closely related to a number of financial products such as credit scoring or insurance. In the insurance industry an important question arises, namely: how can insurance renewal prices be adjusted? Such an adjustment has two conflicting objectives. On the one hand, insurers are forced to retain existing customers, while on the other hand insurers are also forced to increase revenue. Intuitively, one might assume that revenue increases by offering high renewal prices, however this might also cause many customers to terminate their contracts. Contrarily, low renewal prices help retain most existing customers, but could negatively affect revenue. Therefore, adjusting renewal prices is a non-trivial problem for the insurance industry. In this paper, we propose a novel modelization of the renewal price adjustment problem as a sequential decision problem and, consequently, as a Markov decision process (MDP). In particular, this study analyzes two different strategies to carry out this adjustment. The first is about maximizing revenue analyzing the effect of this maximization on customer retention, while the second is about maximizing revenue subject to the client retention level not falling below a given threshold. The former case is related to MDPs with a single criterion to be optimized. The latter case is related to Constrained MDPs (CMDPs) with two criteria, where the first one is related to optimization, while the second is subject to a constraint. This paper also contributes with the resolution of these models by means of a model-free Reinforcement Learning algorithm. Results have been reported using real data from the insurance division of bbva, one of the largest Spanish companies in the banking sector.

关键词： Pricing strategy optimization Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Uncertainty Modelling in Deep Networks: Forecasting Short and Noisy Series

Uncertainty Modelling in Deep Networks: Forecasting Short an...

引用

European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in databases (ECML PKDD)

作者： Brando, Axel Rodriguez-Serrano, Jose A. Ciprian, Mauricio Maestre, Roberto Vitria, Jordi BBVA Data & Analyt Barcelona Spain BBVA Data & Analyt Madrid Spain Univ Barcelona Dept Matemat & Informat Barcelona Spain

ISBN: (纸本)9783030109974;9783030109967

Deep Learning is a consolidated, state-of-the-art Machine Learning tool to fit a function y = f(x) when provided with large data sets of examples {(x(i), y(i))}. However, in regression tasks, the straight-forward application of Deep Learning models provides a point estimate of the target. In addition, the model does not take into account the uncertainty of a prediction. This represents a great limitation for tasks where communicating an erroneous prediction carries a risk. In this paper we tackle a real-world problem of forecasting impending financial expenses and incomings of customers, while displaying predictable monetary amounts on a mobile app. In this context, we investigate if we would obtain an advantage by applying Deep Learning models with a Heteroscedastic model of the variance of a network's output. Experimentally, we achieve a higher accuracy than non-trivial baselines. More importantly, we introduce a mechanism to discard low-confidence predictions, which means that they will not be visible to users. This should help enhance the user experience of our product.

关键词： Deep Learning Uncertainty Aleatoric models Time-series

来源：评论

学校读者我要写书评

暂无评论

Using Copies to Remove Sensitive data: A Case Study on Fair Superhero Alignment Prediction 9th

Using Copies to Remove Sensitive Data: A Case Study on Fair ...

引用

9th Iberian Conference on Pattern Recognition and Image Analysis (IbPRIA)

作者： Unceta, Irene Nin, Jordi Pujol, Oriol BBVA Data & Analyt Barcelona Spain Univ Barcelona Dept Math & Comp Sci Barcelona Spain Univ Ramon Llull ESADE Barcelona Spain

ISBN: (纸本)9783030313326;9783030313319

Ensuring classification models are fair with respect to sensitive data attributes is a crucial task when applying machine learning models to real-world problems. Particularly in company production environments, where the decision output by models may have a direct impact on individuals and predictive performance should be maintained over time. In this article, build upon [17], we propose copies as a technique to mitigate the bias of trained algorithms in circumstances where the original data is not accessible and/or the models cannot be re-trained. In particular, we explore a simple methodology to build copies that replicate the learned decision behavior in the absence of sensitive attributes. We validate this methodology in the low-sensitive problem of superhero alignment. We demonstrate that this naive approach to bias reduction is feasible in this problem and argue that copies can be further exploited to embed models with desiderata such as fair learning.

关键词： Fairness Superhero alignment Bias reduction Copying classifiers

来源：评论

学校读者我要写书评

暂无评论

A comparative analysis of recommender systems based on item aspect opinions extracted from user reviews

引用

USER MODELING AND USER-ADAPTED INTERACTION 2019年第2期29卷 381-441页

作者： Hernandez-Rubio, Maria Cantador, Ivan Bellogin, Alejandro BBVA Data & Analyt Madrid Spain Univ Autonoma Madrid Dept Ingn Informat Madrid Spain

In popular applications such as e-commerce sites and social media, users provide online reviews giving personal opinions about a wide array of items, such as products, services and people. These reviews are usually in the form of free text, and represent a rich source of information about the users' preferences. Among the information elements that can be extracted from reviews, opinions about particular item aspects (i.e., characteristics, attributes or components) have been shown to be effective for user modeling and personalized recommendation. In this paper, we investigate the aspect-based top-N recommendation problem by separately addressing three tasks, namely identifying references to item aspects in user reviews, classifying the sentiment orientation of the opinions about such aspects in the reviews, and exploiting the extracted aspect opinion information to provide enhanced recommendations. Differently to previous work, we integrate and empirically evaluate several state-of-the-art and novel methods for each of the above tasks. We conduct extensive experiments on standard datasets and several domains, analyzing distinct recommendation quality metrics and characteristics of the datasets, domains and extracted aspects. As a result of our investigation, we not only derive conclusions about which combination of methods is most appropriate according to the above issues, but also provide a number of valuable resources for opinion mining and recommendation purposes, such as domain aspect vocabularies and domain-dependent, aspect-level lexicons.

关键词： Recommender systems Aspect-based recommendation Sentiment analysis Opinion mining Aspect extraction User reviews

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：