Applied AI letters最新文献_第6页

Generating and Evaluating Explanations of Attended and Error-Inducing Input Regions for VQA Models VQA模型的关注和误差诱导输入区域的生成和评价解释

Applied AI letters Pub Date : 2021-03-26 DOI: 10.22541/au.162464902.28050142/v1

Arijit Ray, Michael Cogswell, Xiaoyu Lin, Kamran Alipour, Ajay Divakaran, Yi Yao, Giedrius Burachas

{"title":"Generating and Evaluating Explanations of Attended and Error-Inducing Input Regions for VQA Models","authors":"Arijit Ray, Michael Cogswell, Xiaoyu Lin, Kamran Alipour, Ajay Divakaran, Yi Yao, Giedrius Burachas","doi":"10.22541/au.162464902.28050142/v1","DOIUrl":"https://doi.org/10.22541/au.162464902.28050142/v1","url":null,"abstract":"Attention maps, a popular heatmap-based explanation method for Visual\u0000Question Answering (VQA), are supposed to help users understand the\u0000model by highlighting portions of the image/question used by the model\u0000to infer answers. However, we see that users are often misled by current\u0000attention map visualizations that point to relevant regions despite the\u0000model producing an incorrect answer. Hence, we propose Error Maps that\u0000clarify the error by highlighting image regions where the model is prone\u0000to err. Error maps can indicate when a correctly attended region may be\u0000processed incorrectly leading to an incorrect answer, and hence, improve\u0000users’ understanding of those cases. To evaluate our new explanations,\u0000we further introduce a metric that simulates users’ interpretation of\u0000explanations to evaluate their potential helpfulness to understand model\u0000correctness. We finally conduct user studies to see that our new\u0000explanations help users understand model correctness better than\u0000baselines by an expected 30% and that our proxy helpfulness metrics\u0000correlate strongly (rho>0.97) with how well users can\u0000predict model correctness.","PeriodicalId":72253,"journal":{"name":"Applied AI letters","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"45352065","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}

引用次数: 1

Hierarchical spline for time series prediction: An application to naval ship engine failure rate 层次样条时间序列预测在舰船发动机故障率中的应用

Applied AI letters Pub Date : 2021-03-24 DOI: 10.1002/ail2.22

Hyunji Moon, Jinwoo Choi

引用次数: 2

Cognitive analysis in sports: Supporting match analysis and scouting through artificial intelligence 体育认知分析:通过人工智能支持比赛分析和球探

Applied AI letters Pub Date : 2021-03-14 DOI: 10.1002/ail2.21

Joe Pavitt, Dave Braines, Richard Tomsett

{"title":"Cognitive analysis in sports: Supporting match analysis and scouting through artificial intelligence","authors":"Joe Pavitt, Dave Braines, Richard Tomsett","doi":"10.1002/ail2.21","DOIUrl":"10.1002/ail2.21","url":null,"abstract":"In elite sports, there is an opportunity to take advantage of rich and detailed datasets generated across multiple threads of the sporting business. Challenges currently exist due to time constraints to analyse the data, as well as the quantity and variety of data available to assess. Artificial Intelligence (AI) techniques can be a valuable asset in assisting decision makers in tackling such challenges, but deep AI skills are generally not held by those with rich experience in sporting domains. Here, we describe how certain commonly available AI services can be used to provide analytic assistance to sports experts in exploring, and gaining insights from, typical data sources. In particular, we focus on the use of Natural Language Processing and Conversational Interfaces to provide users with an intuitive and time-saving toolkit to explore their datasets and the conclusions arising from analytics performed on them. We show the benefit of presenting powerful AI and analytic techniques to domain experts, showing the potential for impact not only at the elite level of sports, where AI and analytic capabilities may be more available, but also at a more grass-roots level where there is generally little access to specialist resources. The work described in this paper was trialled with Leatherhead Football Club, a semi-professional team that, at the time, were based in the English 7th tier of football.","PeriodicalId":72253,"journal":{"name":"Applied AI letters","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-03-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1002/ail2.21","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"93915417","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 4

Towards an affordable magnetomyography instrumentation and low model complexity approach for labour imminency prediction using a novel multiresolution analysis 使用新颖的多分辨率分析，实现负担得起的磁断层成像仪器和低模型复杂性的劳动迫切性预测方法

Applied AI letters Pub Date : 2021-02-09 DOI: 10.22541/AU.161289481.19912239/V1

E. Nsugbe, I. Sanusi

引用次数: 14

Deep Imputation on Large-Scale Drug Discovery Data 大规模药物发现数据的深度归算

Applied AI letters Pub Date : 2021-01-20 DOI: 10.22541/AU.161111205.55340339/V2

Benedict W J Irwin, T. Whitehead, Scott Rowland, Samar Y. Mahmoud, G. Conduit, M. Segall

{"title":"Deep Imputation on Large-Scale Drug Discovery Data","authors":"Benedict W J Irwin, T. Whitehead, Scott Rowland, Samar Y. Mahmoud, G. Conduit, M. Segall","doi":"10.22541/AU.161111205.55340339/V2","DOIUrl":"https://doi.org/10.22541/AU.161111205.55340339/V2","url":null,"abstract":"More accurate predictions of the biological properties of chemical\u0000compounds would guide the selection and design of new compounds in drug\u0000discovery and help to address the enormous cost and low success-rate of\u0000pharmaceutical R&D. However this domain presents a significant\u0000challenge for AI methods due to the sparsity of compound data and the\u0000noise inherent in results from biological experiments. In this paper, we\u0000demonstrate how data imputation using deep learning provides substantial\u0000improvements over quantitative structure-activity relationship (QSAR)\u0000machine learning models that are widely applied in drug discovery. We\u0000present the largest-to-date successful application of deep-learning\u0000imputation to datasets which are comparable in size to the corporate\u0000data repository of a pharmaceutical company (678,994 compounds by 1166\u0000endpoints). We demonstrate this improvement for three areas of practical\u0000application linked to distinct use cases; i) target activity data\u0000compiled from a range of drug discovery projects, ii) a high value and\u0000heterogeneous dataset covering complex absorption, distribution,\u0000metabolism and elimination properties and, iii) high throughput\u0000screening data, testing the algorithm’s limits on early-stage noisy and\u0000very sparse data. Achieving median coefficients of determination,\u0000R, of 0.69, 0.36 and 0.43 respectively across these\u0000applications, the deep learning imputation method offers an unambiguous\u0000improvement over random forest QSAR methods, which achieve median\u0000R values of 0.28, 0.19 and 0.23 respectively. We also\u0000demonstrate that robust estimates of the uncertainties in the predicted\u0000values correlate strongly with the accuracies in prediction, enabling\u0000greater confidence in decision-making based on the imputed values.","PeriodicalId":72253,"journal":{"name":"Applied AI letters","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"44697723","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}

引用次数: 5

Heritage connector: A machine learning framework for building linked open data from museum collections 遗产连接器:一个机器学习框架，用于从博物馆藏品中构建链接的开放数据

Applied AI letters Pub Date : 2021-01-06 DOI: 10.22541/au.160994838.81187546/v1

Kalyan Dutia, John Stack

{"title":"Heritage connector: A machine learning framework for building linked open data from museum collections","authors":"Kalyan Dutia, John Stack","doi":"10.22541/au.160994838.81187546/v1","DOIUrl":"https://doi.org/10.22541/au.160994838.81187546/v1","url":null,"abstract":"As with almost all data, museum collection catalogues are largely\u0000unstructured, variable in consistency and overwhelmingly composed of\u0000thin records. The form of these catalogues means that the potential for\u0000new forms of research, access and scholarly enquiry that range across\u0000multiple collections and related datasets remains dormant. In the\u0000project Heritage Connector: Transforming text into data to extract\u0000meaning and make connections, we are applying a battery of digital\u0000techniques to connect similar, identical and related items within and\u0000across collections and other publications. In this paper we describe a\u0000framework to create a Linked Open Data knowledge graph (KG) from digital\u0000museum catalogues, connect entities within this graph to Wikidata, and\u0000create new connections in this graph from text. We focus on the use of\u0000machine learning to create these links at scale with a small amount of\u0000labelled data, on a mid-range laptop or a small cloud virtual machine.\u0000We publish open-source software providing tools to perform the tasks of\u0000KG creation, entity matching and named entity recognition under these\u0000constraints.","PeriodicalId":72253,"journal":{"name":"Applied AI letters","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"44868134","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}

引用次数: 7

Corpus processing service: A Knowledge Graph platform to perform deep data exploration on corpora 语料库处理服务:对语料库进行深度数据挖掘的知识图谱平台

Applied AI letters Pub Date : 2020-12-16 DOI: 10.1002/ail2.20

Peter W. J. Staar, Michele Dolfi, Christoph Auer

{"title":"Corpus processing service: A Knowledge Graph platform to perform deep data exploration on corpora","authors":"Peter W. J. Staar, Michele Dolfi, Christoph Auer","doi":"10.1002/ail2.20","DOIUrl":"10.1002/ail2.20","url":null,"abstract":"Knowledge Graphs have been fast emerging as the de facto standard to model and explore knowledge in weakly structured data. Large corpora of documents constitute a source of weakly structured data of particular interest for both the academic and business world. Key examples include scientific publications, technical reports, manuals, patents, regulations, etc. Such corpora embed many facts that are elementary to critical decision making or enabling new discoveries. In this paper, we present a scalable cloud platform to create and serve Knowledge Graphs, which we named corpus processing service (CPS). Its purpose is to process large document corpora, extract the content and embedded facts, and ultimately represent these in a consistent knowledge graph that can be intuitively queried. To accomplish this, we use state-of-the-art natural language understanding models to extract entities and relationships from documents converted with our previously presented corpus conversion service platform. This pipeline is complemented with a newly developed graph engine which ensures extremely performant graph queries and provides powerful graph analytics capabilities. Both components are tightly integrated and can be easily consumed through REST APIs. Additionally, we provide user interfaces to control the data ingestion flow and formulate queries using a visual programming approach. The CPS platform is designed as a modular microservice system operating on Kubernetes clusters. Finally, we validate the quality of queries on our end-to-end knowledge pipeline in a real-world application in the oil and gas industry.","PeriodicalId":72253,"journal":{"name":"Applied AI letters","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-12-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1002/ail2.20","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"98784835","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 6

Evaluating machine learning models for the fast identification of contingency cases 评估用于快速识别突发事件的机器学习模型

Applied AI letters Pub Date : 2020-12-15 DOI: 10.1002/ail2.19

Florian Schäfer, Jan-Hendrik Menke, Martin Braun

{"title":"Evaluating machine learning models for the fast identification of contingency cases","authors":"Florian Schäfer, Jan-Hendrik Menke, Martin Braun","doi":"10.1002/ail2.19","DOIUrl":"https://doi.org/10.1002/ail2.19","url":null,"abstract":"Fast approximations of power flow results are beneficial in power system planning and live operation. In planning, millions of power flow calculations are necessary if multiple years, different control strategies, or contingency policies are to be considered. In live operation, grid operators must assess if grid states comply with contingency requirements in a short time. In this paper, we compare regression and classification methods to either predict multivariable results, for example, bus voltage magnitudes and line loadings, or binary classifications of time steps to identify critical loading situations. We test the methods on three realistic power systems based on time series in 15 and 5 minutes resolution of 1 year. We compare different machine learning models, such as multilayer perceptrons (MLPs), decision trees, k-nearest neighbors, gradient boosting, and evaluate the required training time and prediction times as well as the prediction errors. We additionally determine the amount of training data needed for each method and show results, including the approximation of untrained curtailment of generation. Regarding the compared methods, we identified the MLPs as most suitable for the task. The MLP-based models can predict critical situations with an accuracy of 97% to 98% and a very low number of false negative predictions of 0.0% to 0.64%.","PeriodicalId":72253,"journal":{"name":"Applied AI letters","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1002/ail2.19","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"137685437","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Practical notes on building molecular graph generative models 构建分子图生成模型的实用笔记

Applied AI letters Pub Date : 2020-12-07 DOI: 10.1002/ail2.18

Rocío Mercado, Tobias Rastemo, Edvard Lindelöf, Günter Klambauer, Ola Engkvist, Hongming Chen, Esben Jannik Bjerrum

引用次数: 0

A practical approach for applying Machine Learning in the detection and classification of network devices used in building management 一个实用的方法，应用机器学习在检测和分类的网络设备用于楼宇管理

Applied AI letters Pub Date : 2020-12-02 DOI: 10.22541/au.160689781.19054555/v1

Maroun Touma, Shalisha Witherspoon, S. Witherspoon, Isabelle Crawford-Eng

{"title":"A practical approach for applying Machine Learning in the detection and classification of network devices used in building management","authors":"Maroun Touma, Shalisha Witherspoon, S. Witherspoon, Isabelle Crawford-Eng","doi":"10.22541/au.160689781.19054555/v1","DOIUrl":"https://doi.org/10.22541/au.160689781.19054555/v1","url":null,"abstract":"With the increasing deployment of smart buildings and infrastructure,\u0000Supervisory Control and Data Acquisition (SCADA) devices and the\u0000underlying IT network have become essential elements for the proper\u0000operations of these highly complex systems. Of course, with the increase\u0000in automation and the proliferation of SCADA devices, a corresponding\u0000increase in surface area of attack on critical infrastructure has\u0000increased. Understanding device behaviors in terms of known and\u0000understood or potentially qualified activities versus unknown and\u0000potentially nefarious activities in near-real time is a key component of\u0000any security solution. In this paper, we investigate the challenges with\u0000building robust machine learning models to identify unknowns purely from\u0000network traffic both inside and outside firewalls, starting with missing\u0000or inconsistent labels across sites, feature engineering and learning,\u0000temporal dependencies and analysis, and training data quality (including\u0000small sample sizes) for both shallow and deep learning methods. To\u0000demonstrate these challenges and the capabilities we have developed, we\u0000focus on Building Automation and Control networks (BACnet) from a\u0000private commercial building system. Our results show that ”Model Zoo”\u0000built from binary classifiers based on each device or behavior combined\u0000with an ensemble classifier integrating information from all classifiers\u0000provides a reliable methodology to identify unknown devices as well as\u0000determining specific known devices when the device type is in the\u0000training set. The capability of the Model Zoo framework is shown to be\u0000directly linked to feature engineering and learning, and the dependency\u0000of the feature selection varies depending on both the binary and\u0000ensemble classifiers as well.","PeriodicalId":72253,"journal":{"name":"Applied AI letters","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"44198047","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}

引用次数: 0