This document summarizes a framework for explaining predictions from probabilistic machine learning models using case-based reasoning. The framework consists of two parts: 1) Defining a similarity metric between cases based on how interchangeable their inclusion is in the probability model. 2) Estimating prediction error for a new case by taking the average error of similar past cases. It then applies this framework to explain predictions from a linear regression model for energy performance of households. The document discusses related work on case-based explanation and introduces statistical concepts like J-divergence used in defining the similarity metric.
REPRESENTATION OF UNCERTAIN DATA USING POSSIBILISTIC NETWORK MODELScscpconf
Uncertainty is a pervasive in real world environment due to vagueness, is associated with the
difficulty of making sharp distinctions and ambiguity, is associated with situations in which the
choices among several precise alternatives cannot be perfectly resolved. Analysis of large
collections of uncertain data is a primary task in the real world applications, because data is
incomplete, inaccurate and inefficient. Representation of uncertain data in various forms such
as Data Stream models, Linkage models, Graphical models and so on, which is the most simple,
natural way to process and produce the optimized results through Query processing. In this
paper, we propose the Uncertain Data model can be represented as Possibilistic data model
and vice versa for the process of uncertain data using various data models such as possibilistic
linkage model, Data streams, Possibilistic Graphs. This paper presents representation and
process of Possiblistic Linkage model through Possible Worlds with the use of product-based
operator.
A SYSTEM OF SERIAL COMPUTATION FOR CLASSIFIED RULES PREDICTION IN NONREGULAR ...ijaia
Objects or structures that are regular take uniform dimensions. Based on the concepts of regular models,
our previous research work has developed a system of a regular ontology that models learning structures
in a multiagent system for uniform pre-assessments in a learning environment. This regular ontology has
led to the modelling of a classified rules learning algorithm that predicts the actual number of rules needed
for inductive learning processes and decision making in a multiagent system. But not all processes or
models are regular. Thus this paper presents a system of polynomial equation that can estimate and predict
the required number of rules of a non-regular ontology model given some defined parameters.
On average case analysis through statistical bounds linking theory to practicecsandit
This summary provides the key points from the document in 3 sentences:
The document discusses the limitations of theoretical analysis of algorithms and argues that empirical analysis through statistical bounds can provide a more robust approach for average case analysis. It explains that statistical bounds allow for mixing of different operations rather than analyzing each separately, and do not require pre-assuming a specific input distribution like uniform. The use of statistical bounds for empirical analysis is proposed as a way to supplement theoretical analysis and provide more insight into how algorithms may perform in practice.
ON AVERAGE CASE ANALYSIS THROUGH STATISTICAL BOUNDS : LINKING THEORY TO PRACTICEcscpconf
Theoretical analysis of algorithms involves counting of operations and a separate bound is provided for a specific operation type . Such a methodology is plagued with its inherent
limitations. In this paper we argue as to why we should prefer weight based statistical bounds,which permit mixing of operations, instead as a robust approach. Empirical analysis is an important idea and should be used to supplement and compliment its existing theoretical counterpart as empirically we can work on weights (e.g. time of an operation can be taken as its weight). Not surprisingly, it should not only be taken as an opportunity so as to amend the mistakes already committed knowingly or unknowingly but also to tell a new story.
This document discusses a new technique called "hedging predictions" that provides quantitative measures of accuracy and reliability for predictions made by machine learning algorithms. Hedging predictions complement predictions from algorithms like support vector machines with measures that are provably valid under the assumption that training data is generated independently from the same probability distribution. Validity is achieved automatically, and the goal of hedged prediction is to produce accurate predictions by leveraging powerful machine learning techniques.
This PhD research proposal discusses using Bayesian inference methods for multi-target tracking in big data settings. The researcher proposes developing new stochastic MCMC algorithms that can scale to billions of data points by using small subsets of data in each iteration. This would make Bayesian methods computationally feasible for big data. The proposal outlines reviewing relevant literature, developing the theoretical foundations, and empirically validating new algorithms like sequential Monte Carlo on real-world problems to analyze text and user preferences at large scale.
A TWO-STAGE HYBRID MODEL BY USING ARTIFICIAL NEURAL NETWORKS AS FEATURE CONST...IJDKP
We propose a two-stage hybrid approach with neural networks as the new feature construction algorithms for bankcard response classifications. The hybrid model uses a very simpleneural network structure as the new feature construction tool in the firststage, thenthe newly created features are used asthe additional input variables in logistic regression in the second stage. The modelis compared with the traditional onestage model in credit customer response classification. It is observed that the proposed two-stage model outperforms the one-stage model in terms of accuracy, the area under ROC curve, andKS statistic. By creating new features with theneural network technique, the underlying nonlinear relationships between variables are identified. Furthermore, by using a verysimple neural network structure, the model could overcome the drawbacks of neural networks interms of its long training time, complex topology, and limited interpretability.
The document proposes a new approach to compare stock market patterns to DNA sequences using compression techniques. Stock market data is converted to binary sequences representing increases and decreases, which are then encoded into DNA nucleotides. These nucleotide sequences are divided and matched against human genome sequences using BLAST. The analysis found certain sub-sequences of the stock market patterns matched 100% to the human genome, suggesting this approach could potentially predict stock market behavior.
REPRESENTATION OF UNCERTAIN DATA USING POSSIBILISTIC NETWORK MODELScscpconf
Uncertainty is a pervasive in real world environment due to vagueness, is associated with the
difficulty of making sharp distinctions and ambiguity, is associated with situations in which the
choices among several precise alternatives cannot be perfectly resolved. Analysis of large
collections of uncertain data is a primary task in the real world applications, because data is
incomplete, inaccurate and inefficient. Representation of uncertain data in various forms such
as Data Stream models, Linkage models, Graphical models and so on, which is the most simple,
natural way to process and produce the optimized results through Query processing. In this
paper, we propose the Uncertain Data model can be represented as Possibilistic data model
and vice versa for the process of uncertain data using various data models such as possibilistic
linkage model, Data streams, Possibilistic Graphs. This paper presents representation and
process of Possiblistic Linkage model through Possible Worlds with the use of product-based
operator.
A SYSTEM OF SERIAL COMPUTATION FOR CLASSIFIED RULES PREDICTION IN NONREGULAR ...ijaia
Objects or structures that are regular take uniform dimensions. Based on the concepts of regular models,
our previous research work has developed a system of a regular ontology that models learning structures
in a multiagent system for uniform pre-assessments in a learning environment. This regular ontology has
led to the modelling of a classified rules learning algorithm that predicts the actual number of rules needed
for inductive learning processes and decision making in a multiagent system. But not all processes or
models are regular. Thus this paper presents a system of polynomial equation that can estimate and predict
the required number of rules of a non-regular ontology model given some defined parameters.
On average case analysis through statistical bounds linking theory to practicecsandit
This summary provides the key points from the document in 3 sentences:
The document discusses the limitations of theoretical analysis of algorithms and argues that empirical analysis through statistical bounds can provide a more robust approach for average case analysis. It explains that statistical bounds allow for mixing of different operations rather than analyzing each separately, and do not require pre-assuming a specific input distribution like uniform. The use of statistical bounds for empirical analysis is proposed as a way to supplement theoretical analysis and provide more insight into how algorithms may perform in practice.
ON AVERAGE CASE ANALYSIS THROUGH STATISTICAL BOUNDS : LINKING THEORY TO PRACTICEcscpconf
Theoretical analysis of algorithms involves counting of operations and a separate bound is provided for a specific operation type . Such a methodology is plagued with its inherent
limitations. In this paper we argue as to why we should prefer weight based statistical bounds,which permit mixing of operations, instead as a robust approach. Empirical analysis is an important idea and should be used to supplement and compliment its existing theoretical counterpart as empirically we can work on weights (e.g. time of an operation can be taken as its weight). Not surprisingly, it should not only be taken as an opportunity so as to amend the mistakes already committed knowingly or unknowingly but also to tell a new story.
This document discusses a new technique called "hedging predictions" that provides quantitative measures of accuracy and reliability for predictions made by machine learning algorithms. Hedging predictions complement predictions from algorithms like support vector machines with measures that are provably valid under the assumption that training data is generated independently from the same probability distribution. Validity is achieved automatically, and the goal of hedged prediction is to produce accurate predictions by leveraging powerful machine learning techniques.
This PhD research proposal discusses using Bayesian inference methods for multi-target tracking in big data settings. The researcher proposes developing new stochastic MCMC algorithms that can scale to billions of data points by using small subsets of data in each iteration. This would make Bayesian methods computationally feasible for big data. The proposal outlines reviewing relevant literature, developing the theoretical foundations, and empirically validating new algorithms like sequential Monte Carlo on real-world problems to analyze text and user preferences at large scale.
A TWO-STAGE HYBRID MODEL BY USING ARTIFICIAL NEURAL NETWORKS AS FEATURE CONST...IJDKP
We propose a two-stage hybrid approach with neural networks as the new feature construction algorithms for bankcard response classifications. The hybrid model uses a very simpleneural network structure as the new feature construction tool in the firststage, thenthe newly created features are used asthe additional input variables in logistic regression in the second stage. The modelis compared with the traditional onestage model in credit customer response classification. It is observed that the proposed two-stage model outperforms the one-stage model in terms of accuracy, the area under ROC curve, andKS statistic. By creating new features with theneural network technique, the underlying nonlinear relationships between variables are identified. Furthermore, by using a verysimple neural network structure, the model could overcome the drawbacks of neural networks interms of its long training time, complex topology, and limited interpretability.
The document proposes a new approach to compare stock market patterns to DNA sequences using compression techniques. Stock market data is converted to binary sequences representing increases and decreases, which are then encoded into DNA nucleotides. These nucleotide sequences are divided and matched against human genome sequences using BLAST. The analysis found certain sub-sequences of the stock market patterns matched 100% to the human genome, suggesting this approach could potentially predict stock market behavior.
Illustration of Medical Image Segmentation based on Clustering Algorithmsrahulmonikasharma
Image segmentation is the most basic and crucial process remembering the true objective to facilitate the characterization and representation of the structure of excitement for medical or basic images. Despite escalated research, segmentation remains a challenging issue because of the differing image content, cluttered objects, occlusion, non-uniform object surface, and different factors. There are numerous calculations and techniques accessible for image segmentation yet at the same time there requirements to build up an efficient, quick technique of medical image segmentation. This paper has focused on K-means and Fuzzy C means clustering algorithm to segment malaria blood samples in more accurate manner.
This document analyzes a single student learning episode using two theoretical lenses: the instrumental genesis perspective and the onto-semiotic approach. The instrumental genesis perspective focuses on how students develop techniques for using tools or artifacts to solve mathematical tasks, and the relationships between thinking and gestures. The onto-semiotic approach views mathematical knowledge and learning as involving systems of practices within social and institutional contexts. Analyzing the same episode from both perspectives provides complementary insights and a richer understanding of the phenomena, while also helping to identify the strengths and limitations of each theoretical approach. Networking the two theories in this way contributes to theoretical development in mathematics education.
The document discusses the differences between machine learning (ML), statistical learning, data mining (DM), and automated learning (AL). It argues that while ML and statistical learning developed similar techniques starting in the 1960s, DM emerged in the 1990s from a merging of database research and automated learning. However, industry was much more enthusiastic about adopting DM techniques compared to AL techniques, even though many DM systems are just friendly interfaces of AL systems. The document aims to explain the key differences between DM and AL that led to DM's greater commercial success.
International Journal of Engineering Research and Development (IJERD)IJERD Editor
journal publishing, how to publish research paper, Call For research paper, international journal, publishing a paper, IJERD, journal of science and technology, how to get a research paper published, publishing a paper, publishing of journal, publishing of research paper, reserach and review articles, IJERD Journal, How to publish your research paper, publish research paper, open access engineering journal, Engineering journal, Mathemetics journal, Physics journal, Chemistry journal, Computer Engineering, Computer Science journal, how to submit your paper, peer reviw journal, indexed journal, reserach and review articles, engineering journal, www.ijerd.com, research journals,
yahoo journals, bing journals, International Journal of Engineering Research and Development, google journals, hard copy of journal
Cost Estimation Predictive Modeling: Regression versus Neural Networkmustafa sarac
This document compares regression and neural network models for cost estimation predictive modeling. It summarizes previous research applying regression and neural networks to cost estimation problems in various fields. The document then describes a study that systematically examines the performance of regression versus neural networks for cost estimation under different data conditions, including varying data set sizes, imperfections from noise, and sampling bias. Both simulated and real-world data sets are used to compare the accuracy, variability, and usability of regression and neural network models for cost estimation.
The document discusses analogical reasoning and case-based reasoning. It provides an overview of research in these areas including structure mapping theory, models of analogical processing like SME and MAC/FAC, and case-based reasoning systems. It proposes an analogy ontology to integrate analogical processing and first-principles reasoning by providing a formal representation of analogy concepts and results.
EXPERT OPINION AND COHERENCE BASED TOPIC MODELINGijnlc
In this paper, we propose a novel algorithm that rearrange the topic assignment results obtained from topic
modeling algorithms, including NMF and LDA. The effectiveness of the algorithm is measured by how much
the results conform to expert opinion, which is a data structure called TDAG that we defined to represent the
probability that a pair of highly correlated words appear together. In order to make sure that the internal
structure does not get changed too much from the rearrangement, coherence, which is a well known metric
for measuring the effectiveness of topic modeling, is used to control the balance of the internal structure.
We developed two ways to systematically obtain the expert opinion from data, depending on whether the
data has relevant expert writing or not. The final algorithm which takes into account both coherence and
expert opinion is presented. Finally we compare amount of adjustments needed to be done for each topic
modeling method, NMF and LDA.
In order to solve the complex decision-making problems, there are many approaches and systems based on the fuzzy theory were proposed. In 1998, Smarandache introduced the concept of single-valued neutrosophic set as a complete development of fuzzy theory. In this paper, we research on the distance measure between single-valued neutrosophic sets based on the H-max measure of Ngan et al. [8]. The proposed measure is also a distance measure between picture fuzzy sets which was introduced by Cuong in 2013 [15]. Based on the proposed measure, an Adaptive Neuro Picture Fuzzy Inference System (ANPFIS) is built and applied to the decision making for the link states in interconnection networks. In experimental evaluation on the real datasets taken from the UPV (Universitat Politècnica de València) university, the performance of the proposed model is better than that of the related fuzzy methods.
UNDERSTANDING NEGATIVE SAMPLING IN KNOWLEDGE GRAPH EMBEDDINGijaia
This document summarizes and categorizes existing approaches for negative sampling in knowledge graph embedding. It divides negative sampling methods into three categories: 1) static distribution-based approaches like uniform and Bernoulli sampling that sample negatives from fixed distributions, 2) dynamic distribution-based approaches that sample from adaptive distributions, and 3) custom cluster-based approaches that group entities for targeted negative sampling. The document analyzes representative approaches within each category and discusses their characteristics and limitations to provide guidance on negative sampling in knowledge graph embedding.
This paper introduces approaches to combining logic, probability, and learning. It discusses past attempts to solve probabilistic logic learning and overviews different formalisms for defining probabilities on logical views. It also describes approaches that combine probabilistic reasoning and logical representation, such as Bayesian logic programs and probabilistic relational models. Learning probabilistic logics involves adapting probabilistic models based on data, including tasks of parameter estimation and structure learning. The paper provides an integrated survey of various concepts in this area.
In context-aware trust evaluation, using ontology tree is a popular approach to represent the relation
between contexts. Usually, similarity between two contexts is computed using these trees. Therefore, the
performance of trust evaluation highly depends on the quality of ontology trees. Fairness or granularity
consistency is one of the major limitations affecting the quality of ontology tree. This limitation refers to
inequality of semantic similarity in the most ontology trees. In other words, semantic similarity of every two
adjacent nodes is unequal in these trees. It deteriorates the performance of contexts similarity computation.
We overcome this limitation by weighting tree edges based on their semantic similarity. Weight of each
edge is computed using Normalized Similarity Score (NSS) method. This method is based on frequencies of
concepts (words) co-occurrences in the pages indexed by search engines. Our experiments represent the
better performance of the proposed approach in comparison with established trust evaluation approaches.
The suggested approach can enhance efficiency of any solution which models semantic relations by
ontology tree.
A NEW METHOD OF CENTRAL DIFFERENCE INTERPOLATIONmathsjournal
In Numerical analysis, interpolation is a manner of calculating the unknown values of a function for any conferred value of argument within the limit of the arguments. It provides basically a concept of estimating unknown data with the aid of relating acquainted data. The main goal of this research is to constitute a central difference interpolation method which is derived from the combination of Gauss’s third formula, Gauss’s Backward formula and Gauss’s forward formula. We have also demonstrated the graphical presentations as well as comparison through all the existing interpolation formulas with our propound method of central difference interpolation. By the comparison and graphical presentation, the new method gives the best result with the lowest error from another existing interpolationformula.
INCREMENTAL SEMI-SUPERVISED CLUSTERING METHOD USING NEIGHBOURHOOD ASSIGNMENTIJCSEA Journal
Semi-supervised considering so as to cluster expects to enhance clustering execution client supervision as
pair wise imperatives. In this paper, we contemplate the dynamic learning issue of selecting pair wise
must-connect and can't interface imperatives for semi supervised clustering. We consider dynamic learning
in an iterative way where in every emphasis questions are chosen in light of the current clustering
arrangement and the current requirement set. We apply a general system that expands on the idea of
Neighbourhood, where Neighbourhoods contain "named samples" of distinctive bunches as indicated by
the pair wise imperatives. Our dynamic learning strategy extends the areas by selecting educational
focuses and questioning their association with the areas. Under this system, we expand on the fantastic
vulnerability based rule and present a novel methodology for figuring the instability related with every
information point. We further present a determination foundation that exchanges off the measure of
vulnerability of every information point with the expected number of inquiries (the expense) needed to
determine this instability. This permits us to choose questions that have the most astounding data rate. We
assess the proposed strategy on the benchmark information sets and the outcomes show predictable and
significant upgrades over the current cutting edge.
Fundamentals of the fuzzy logic based generalized theory of decisionsSpringer
The document discusses decision making under imperfect information when information is represented by geometric primitives rather than numbers or fuzzy sets. It introduces the concept of "fuzzy geometry" or "F-geometry" where visual perceptions that cannot be precisely represented are modeled using geometric shapes. Five cases of imperfect information are considered, from numerical information to purely visual perceptions. F-geometry uses primitives like points, lines, and shapes to represent perceptions and enables reasoning about decisions despite imprecise information. Basic primitives, operations, and axioms of F-geometry are defined to formally represent and reason with perceptions.
This document provides tips and sample answers for common janitor interview questions. It discusses how to answer questions about yourself, your strengths, career goals, reasons for leaving previous jobs, weaknesses, knowledge of the organization, and ways you've improved your skills in the last year. For each question, it offers steps and guidelines to provide well-prepared, positive responses that highlight your relevant qualifications and experience. Sample answers are given as examples of effective ways to respond to help land the janitor position.
The document discusses how personalization and dynamic content are becoming increasingly important on websites. It notes that 52% of marketers see content personalization as critical and 75% of consumers like it when brands personalize their content. However, personalization can create issues for search engine optimization as dynamic URLs and content are more difficult for search engines to index than static pages. The document provides tips for SEOs to help address these personalization and SEO challenges, such as using static URLs when possible and submitting accurate sitemaps.
Lightning Talk #9: How UX and Data Storytelling Can Shape Policy by Mika Aldabaux singapore
How can we take UX and Data Storytelling out of the tech context and use them to change the way government behaves?
Showcasing the truth is the highest goal of data storytelling. Because the design of a chart can affect the interpretation of data in a major way, one must wield visual tools with care and deliberation. Using quantitative facts to evoke an emotional response is best achieved with the combination of UX and data storytelling.
This document summarizes a study of CEO succession events among the largest 100 U.S. corporations between 2005-2015. The study analyzed executives who were passed over for the CEO role ("succession losers") and their subsequent careers. It found that 74% of passed over executives left their companies, with 30% eventually becoming CEOs elsewhere. However, companies led by succession losers saw average stock price declines of 13% over 3 years, compared to gains for companies whose CEO selections remained unchanged. The findings suggest that boards generally identify the most qualified CEO candidates, though differences between internal and external hires complicate comparisons.
Illustration of Medical Image Segmentation based on Clustering Algorithmsrahulmonikasharma
Image segmentation is the most basic and crucial process remembering the true objective to facilitate the characterization and representation of the structure of excitement for medical or basic images. Despite escalated research, segmentation remains a challenging issue because of the differing image content, cluttered objects, occlusion, non-uniform object surface, and different factors. There are numerous calculations and techniques accessible for image segmentation yet at the same time there requirements to build up an efficient, quick technique of medical image segmentation. This paper has focused on K-means and Fuzzy C means clustering algorithm to segment malaria blood samples in more accurate manner.
This document analyzes a single student learning episode using two theoretical lenses: the instrumental genesis perspective and the onto-semiotic approach. The instrumental genesis perspective focuses on how students develop techniques for using tools or artifacts to solve mathematical tasks, and the relationships between thinking and gestures. The onto-semiotic approach views mathematical knowledge and learning as involving systems of practices within social and institutional contexts. Analyzing the same episode from both perspectives provides complementary insights and a richer understanding of the phenomena, while also helping to identify the strengths and limitations of each theoretical approach. Networking the two theories in this way contributes to theoretical development in mathematics education.
The document discusses the differences between machine learning (ML), statistical learning, data mining (DM), and automated learning (AL). It argues that while ML and statistical learning developed similar techniques starting in the 1960s, DM emerged in the 1990s from a merging of database research and automated learning. However, industry was much more enthusiastic about adopting DM techniques compared to AL techniques, even though many DM systems are just friendly interfaces of AL systems. The document aims to explain the key differences between DM and AL that led to DM's greater commercial success.
International Journal of Engineering Research and Development (IJERD)IJERD Editor
journal publishing, how to publish research paper, Call For research paper, international journal, publishing a paper, IJERD, journal of science and technology, how to get a research paper published, publishing a paper, publishing of journal, publishing of research paper, reserach and review articles, IJERD Journal, How to publish your research paper, publish research paper, open access engineering journal, Engineering journal, Mathemetics journal, Physics journal, Chemistry journal, Computer Engineering, Computer Science journal, how to submit your paper, peer reviw journal, indexed journal, reserach and review articles, engineering journal, www.ijerd.com, research journals,
yahoo journals, bing journals, International Journal of Engineering Research and Development, google journals, hard copy of journal
Cost Estimation Predictive Modeling: Regression versus Neural Networkmustafa sarac
This document compares regression and neural network models for cost estimation predictive modeling. It summarizes previous research applying regression and neural networks to cost estimation problems in various fields. The document then describes a study that systematically examines the performance of regression versus neural networks for cost estimation under different data conditions, including varying data set sizes, imperfections from noise, and sampling bias. Both simulated and real-world data sets are used to compare the accuracy, variability, and usability of regression and neural network models for cost estimation.
The document discusses analogical reasoning and case-based reasoning. It provides an overview of research in these areas including structure mapping theory, models of analogical processing like SME and MAC/FAC, and case-based reasoning systems. It proposes an analogy ontology to integrate analogical processing and first-principles reasoning by providing a formal representation of analogy concepts and results.
EXPERT OPINION AND COHERENCE BASED TOPIC MODELINGijnlc
In this paper, we propose a novel algorithm that rearrange the topic assignment results obtained from topic
modeling algorithms, including NMF and LDA. The effectiveness of the algorithm is measured by how much
the results conform to expert opinion, which is a data structure called TDAG that we defined to represent the
probability that a pair of highly correlated words appear together. In order to make sure that the internal
structure does not get changed too much from the rearrangement, coherence, which is a well known metric
for measuring the effectiveness of topic modeling, is used to control the balance of the internal structure.
We developed two ways to systematically obtain the expert opinion from data, depending on whether the
data has relevant expert writing or not. The final algorithm which takes into account both coherence and
expert opinion is presented. Finally we compare amount of adjustments needed to be done for each topic
modeling method, NMF and LDA.
In order to solve the complex decision-making problems, there are many approaches and systems based on the fuzzy theory were proposed. In 1998, Smarandache introduced the concept of single-valued neutrosophic set as a complete development of fuzzy theory. In this paper, we research on the distance measure between single-valued neutrosophic sets based on the H-max measure of Ngan et al. [8]. The proposed measure is also a distance measure between picture fuzzy sets which was introduced by Cuong in 2013 [15]. Based on the proposed measure, an Adaptive Neuro Picture Fuzzy Inference System (ANPFIS) is built and applied to the decision making for the link states in interconnection networks. In experimental evaluation on the real datasets taken from the UPV (Universitat Politècnica de València) university, the performance of the proposed model is better than that of the related fuzzy methods.
UNDERSTANDING NEGATIVE SAMPLING IN KNOWLEDGE GRAPH EMBEDDINGijaia
This document summarizes and categorizes existing approaches for negative sampling in knowledge graph embedding. It divides negative sampling methods into three categories: 1) static distribution-based approaches like uniform and Bernoulli sampling that sample negatives from fixed distributions, 2) dynamic distribution-based approaches that sample from adaptive distributions, and 3) custom cluster-based approaches that group entities for targeted negative sampling. The document analyzes representative approaches within each category and discusses their characteristics and limitations to provide guidance on negative sampling in knowledge graph embedding.
This paper introduces approaches to combining logic, probability, and learning. It discusses past attempts to solve probabilistic logic learning and overviews different formalisms for defining probabilities on logical views. It also describes approaches that combine probabilistic reasoning and logical representation, such as Bayesian logic programs and probabilistic relational models. Learning probabilistic logics involves adapting probabilistic models based on data, including tasks of parameter estimation and structure learning. The paper provides an integrated survey of various concepts in this area.
In context-aware trust evaluation, using ontology tree is a popular approach to represent the relation
between contexts. Usually, similarity between two contexts is computed using these trees. Therefore, the
performance of trust evaluation highly depends on the quality of ontology trees. Fairness or granularity
consistency is one of the major limitations affecting the quality of ontology tree. This limitation refers to
inequality of semantic similarity in the most ontology trees. In other words, semantic similarity of every two
adjacent nodes is unequal in these trees. It deteriorates the performance of contexts similarity computation.
We overcome this limitation by weighting tree edges based on their semantic similarity. Weight of each
edge is computed using Normalized Similarity Score (NSS) method. This method is based on frequencies of
concepts (words) co-occurrences in the pages indexed by search engines. Our experiments represent the
better performance of the proposed approach in comparison with established trust evaluation approaches.
The suggested approach can enhance efficiency of any solution which models semantic relations by
ontology tree.
A NEW METHOD OF CENTRAL DIFFERENCE INTERPOLATIONmathsjournal
In Numerical analysis, interpolation is a manner of calculating the unknown values of a function for any conferred value of argument within the limit of the arguments. It provides basically a concept of estimating unknown data with the aid of relating acquainted data. The main goal of this research is to constitute a central difference interpolation method which is derived from the combination of Gauss’s third formula, Gauss’s Backward formula and Gauss’s forward formula. We have also demonstrated the graphical presentations as well as comparison through all the existing interpolation formulas with our propound method of central difference interpolation. By the comparison and graphical presentation, the new method gives the best result with the lowest error from another existing interpolationformula.
INCREMENTAL SEMI-SUPERVISED CLUSTERING METHOD USING NEIGHBOURHOOD ASSIGNMENTIJCSEA Journal
Semi-supervised considering so as to cluster expects to enhance clustering execution client supervision as
pair wise imperatives. In this paper, we contemplate the dynamic learning issue of selecting pair wise
must-connect and can't interface imperatives for semi supervised clustering. We consider dynamic learning
in an iterative way where in every emphasis questions are chosen in light of the current clustering
arrangement and the current requirement set. We apply a general system that expands on the idea of
Neighbourhood, where Neighbourhoods contain "named samples" of distinctive bunches as indicated by
the pair wise imperatives. Our dynamic learning strategy extends the areas by selecting educational
focuses and questioning their association with the areas. Under this system, we expand on the fantastic
vulnerability based rule and present a novel methodology for figuring the instability related with every
information point. We further present a determination foundation that exchanges off the measure of
vulnerability of every information point with the expected number of inquiries (the expense) needed to
determine this instability. This permits us to choose questions that have the most astounding data rate. We
assess the proposed strategy on the benchmark information sets and the outcomes show predictable and
significant upgrades over the current cutting edge.
Fundamentals of the fuzzy logic based generalized theory of decisionsSpringer
The document discusses decision making under imperfect information when information is represented by geometric primitives rather than numbers or fuzzy sets. It introduces the concept of "fuzzy geometry" or "F-geometry" where visual perceptions that cannot be precisely represented are modeled using geometric shapes. Five cases of imperfect information are considered, from numerical information to purely visual perceptions. F-geometry uses primitives like points, lines, and shapes to represent perceptions and enables reasoning about decisions despite imprecise information. Basic primitives, operations, and axioms of F-geometry are defined to formally represent and reason with perceptions.
This document provides tips and sample answers for common janitor interview questions. It discusses how to answer questions about yourself, your strengths, career goals, reasons for leaving previous jobs, weaknesses, knowledge of the organization, and ways you've improved your skills in the last year. For each question, it offers steps and guidelines to provide well-prepared, positive responses that highlight your relevant qualifications and experience. Sample answers are given as examples of effective ways to respond to help land the janitor position.
The document discusses how personalization and dynamic content are becoming increasingly important on websites. It notes that 52% of marketers see content personalization as critical and 75% of consumers like it when brands personalize their content. However, personalization can create issues for search engine optimization as dynamic URLs and content are more difficult for search engines to index than static pages. The document provides tips for SEOs to help address these personalization and SEO challenges, such as using static URLs when possible and submitting accurate sitemaps.
Lightning Talk #9: How UX and Data Storytelling Can Shape Policy by Mika Aldabaux singapore
How can we take UX and Data Storytelling out of the tech context and use them to change the way government behaves?
Showcasing the truth is the highest goal of data storytelling. Because the design of a chart can affect the interpretation of data in a major way, one must wield visual tools with care and deliberation. Using quantitative facts to evoke an emotional response is best achieved with the combination of UX and data storytelling.
This document summarizes a study of CEO succession events among the largest 100 U.S. corporations between 2005-2015. The study analyzed executives who were passed over for the CEO role ("succession losers") and their subsequent careers. It found that 74% of passed over executives left their companies, with 30% eventually becoming CEOs elsewhere. However, companies led by succession losers saw average stock price declines of 13% over 3 years, compared to gains for companies whose CEO selections remained unchanged. The findings suggest that boards generally identify the most qualified CEO candidates, though differences between internal and external hires complicate comparisons.
A Formal Machine Learning or Multi Objective Decision Making System for Deter...Editor IJCATR
Decision-making typically needs the mechanisms to compromise among opposing norms. Once multiple objectives square measure is concerned of machine learning, a vital step is to check the weights of individual objectives to the system-level performance. Determinant, the weights of multi-objectives is associate in analysis method, associated it's been typically treated as a drawback. However, our preliminary investigation has shown that existing methodologies in managing the weights of multi-objectives have some obvious limitations like the determination of weights is treated as one drawback, a result supporting such associate improvement is limited, if associated it will even be unreliable, once knowledge concerning multiple objectives is incomplete like an integrity caused by poor data. The constraints of weights are also mentioned. Variable weights square measure is natural in decision-making processes. Here, we'd like to develop a scientific methodology in determinant variable weights of multi-objectives. The roles of weights in a creative multi-objective decision-making or machine-learning of square measure analyzed, and therefore the weights square measure determined with the help of a standard neural network.
A Magnified Application of Deficient Data Using Bolzano Classifierjournal ijrtem
Abstract: Deficient and Inconsistent knowledge are a pervasive and lasting problem in heavy information sets. However, basic accusation is attractive often used to impute missing data, whereas multiple imputation generates right value to replace. Consequently, variety of machine learning (ML) techniques are developed to reprocess the incomplete information. It is estimated multiple imputation of missing information in large datasets and the performance of MI focuses on several unsupervised ML algorithms like mean, median, standard deviation and Supervised ML techniques for probabilistic algorithm like NBI classifier. Such research is carried out employing a comprehensive range of databases, for which missing cases are first filled in by various sets of plausible values to create multiple completed datasets, then standard complete- data operations are applied to each completed dataset, and finally the multiple sets of results combine to generate a single inference. The main aim of this report is to offer general guidelines for selection of suitable data imputation algorithms based on characteristics of the data. Implementing Bolzano Weierstrass theorem in machine learning techniques to assess the functioning of every sequence of rational and irrational number has a monotonic subsequence and specify every sequence always has a finite boundary. For estimate imputation of missing data, the standard machine learning repository dataset has been applied. Tentative effects manifest the proposed approach to have good accuracy and the accuracy measured in terms of percent. Keywords: Bolzano Classifier, Maximum Likelihood, NBI classifier, posterior probability, predictor probability, prior probability, Supervised ML, Unsupervised ML.
rule refinement in inductive knowledge based systemsarteimi
This document summarizes a paper about refining inductive knowledge-based systems through empirical methods. It describes a system called ARIS that interleaves learning and performance evaluation to allow accurate classifications on real-world datasets. ARIS employs an ordering of rules according to their learned weights, which are calculated using Bayes' theorem. The system focuses rule analysis and refinement by heuristics. It has been used to refine knowledge bases from C4.5 and RIPPER, showing improved accuracy and the ability to gradually enhance the knowledge base during refinement.
A Novel Clustering Method for Similarity Measuring in Text DocumentsIJMER
International Journal of Modern Engineering Research (IJMER) is Peer reviewed, online Journal. It serves as an international archival forum of scholarly research related to engineering and science education.
This document summarizes an article that proposes a novel cost-free learning (CFL) approach called ABC-SVM to address the class imbalance problem. The approach aims to maximize the normalized mutual information of the predicted and actual classes to balance errors and rejects without requiring cost information. It optimizes misclassification costs, SVM parameters, and feature selection simultaneously using an artificial bee colony algorithm. Experimental results on several datasets show the method performs effectively compared to sampling techniques for class imbalance.
On average case analysis through statistical bounds linking theory to practicecsandit
Theoretical analysis of algorithms involves counting of operations and a separate bound is
provided for a specific operation type. Such a methodology is plagued with its inherent
limitations. In this paper we argue as to why we should prefer weight based statistical bounds,
which permit mixing of operations, instead as a robust approach. Empirical analysis is an
important idea and should be used to supplement and compliment its existing theoretical
counterpart as empirically we can work on weights (e.g. time of an operation can be taken as its
weight). Not surprisingly, it should not only be taken as an opportunity so as to amend the
mistakes already committed knowingly or unknowingly but also to tell a new story.
DEVELOPING A CASE-BASED RETRIEVAL SYSTEM FOR SUPPORTING IT CUSTOMERSIJCSEA Journal
The case-based reasoning (CBR) approach is a modern approach. It is adopted for designing knowledgebased expertise systems. The aforementioned approach depends much on the stored experiences. These experiences serve as cases that can be employed for solving new problems. That is done through retrieving similar cases from the system and utilizing their solutions. The latter approach aims at solving problems through reviewing, processing and applying their experiences. The present study aimed to shed a light on a CBR application. That is done to develop a new system for assisting Internet users and solving the problems faced by those users. In addition, the present study focuses on the cases retrieval stage. It aimed at designing and building an experienced inquiry system for solving any problem that internet users might face when using a case-based reasoning (CBR) dialogue. That shall enable internet users to solve the problems faced when using the Internet. The system that was developed by the researcher operates through displaying the similar cases through a dialogue. That is done through using a well-developed algorithm and reviewing the relevant previous studies. It was found that the success rate of the proposed system is high.
DEVELOPING A CASE-BASED RETRIEVAL SYSTEM FOR SUPPORTING IT CUSTOMERSIJCSEA Journal
This document describes the development of a case-based retrieval system to help IT customers solve problems. It discusses using a case-based reasoning approach where prior solutions and experiences are stored as cases. A conversational case-based reasoning system is developed that allows users to describe problems and receive potential solutions through a dialogue. The system was tested on sample problems and achieved a high success rate of 90% with an average of 7.7 steps to retrieve solutions.
The document summarizes five papers that address challenges in context-aware recommendation systems using factorization methods. Three key challenges are high dimensionality, data sparsity, and cold starts. The papers propose various algorithms using matrix factorization and tensor factorization to address these challenges. COT models each context as an operation on user-item pairs to reduce dimensionality. Another approach extracts latent contexts from sensor data using deep learning and matrix factorization. CSLIM extends the SLIM algorithm to incorporate contextual ratings. TAPER uses tensor factorization to integrate various contexts for expert recommendations. Finally, GFF provides a generalized factorization framework to handle different recommendation models. The document analyzes how well each paper meets the challenges.
Textual Data Partitioning with Relationship and Discriminative AnalysisEditor IJMTER
Data partitioning methods are used to partition the data values with similarity. Similarity
measures are used to estimate transaction relationships. Hierarchical clustering model produces tree
structured results. Partitioned clustering produces results in grid format. Text documents are
unstructured data values with high dimensional attributes. Document clustering group ups unlabeled text
documents into meaningful clusters. Traditional clustering methods require cluster count (K) for the
document grouping process. Clustering accuracy degrades drastically with reference to the unsuitable
cluster count.
Textual data elements are divided into two types’ discriminative words and nondiscriminative
words. Only discriminative words are useful for grouping documents. The involvement of
nondiscriminative words confuses the clustering process and leads to poor clustering solution in return.
A variation inference algorithm is used to infer the document collection structure and partition of
document words at the same time. Dirichlet Process Mixture (DPM) model is used to partition
documents. DPM clustering model uses both the data likelihood and the clustering property of the
Dirichlet Process (DP). Dirichlet Process Mixture Model for Feature Partition (DPMFP) is used to
discover the latent cluster structure based on the DPM model. DPMFP clustering is performed without
requiring the number of clusters as input.
Document labels are used to estimate the discriminative word identification process. Concept
relationships are analyzed with Ontology support. Semantic weight model is used for the document
similarity analysis. The system improves the scalability with the support of labels and concept relations
for dimensionality reduction process.
The document proposes a framework for evaluating information retrieval systems that accounts for variance in system effectiveness across query variations for the same topic, in addition to average effectiveness. It argues current approaches only consider average effectiveness across topics and do not differentiate variance due to queries versus topics. The framework applies concepts from mean-variance analysis to rank systems based on their average performance and stability over query variations, weighted by topic importance and user tolerance for effectiveness fluctuations. It aims to provide a more nuanced and user-centered approach to comparing system effectiveness.
On the benefit of logic-based machine learning to learn pairwise comparisonsjournalBEEI
This document describes a study that used a logic-based machine learning approach called APARELL to learn user preferences from pairwise comparisons in a recommender system. APARELL learns description logic rules from examples of users' pairwise item preferences and background knowledge about item properties. It was implemented in a used car recommender system. A user study evaluated the pairwise preference elicitation method compared to a standard list interface. Results showed the pairwise interface was significantly better in recommending items that matched users' preferences.
IJERD (www.ijerd.com) International Journal of Engineering Research and Devel...IJERD Editor
This document discusses distance similarity measures that can be used for data mining classification and clustering techniques. It proposes a novel distance similarity measure called "Supervised & Unsupervised learning" that uses Euclidean distance similarity to partition training data into clusters. It then builds decision trees on each cluster to improve classification performance. The document also discusses using these measures for other applications like image processing, where k-means clustering can be used to segment images into clusters of similar pixel intensities. In conclusion, it states these similarity measures can help analyze complex datasets for business analysis purposes.
Classifier Model using Artificial Neural NetworkAI Publications
This document summarizes a research paper that investigates using supervised instance selection (SIS) as a preprocessing step to improve the performance of artificial neural networks (ANNs) for classification tasks. SIS aims to select a subset of examples from the original dataset to enhance the accuracy of future classifications. The goal of applying SIS before ANNs is to provide a cleaner input dataset that handles noisy or redundant data better. The paper presents the architecture of feedforward neural networks and the backpropagation algorithm for training networks. It also discusses using mutual information-based feature selection as part of the SIS preprocessing approach.
An Introspective Component-Based Approach For Meta-LevelLaurie Smith
This document presents an introspective component-based architecture for clinical decision support systems. The architecture allows integrating rule-based, case-based, and probabilistic reasoning. It is being developed and tested in the domain of palliative cancer care to identify and utilize clinical guidelines. The architecture uses a meta-level reasoner to modify and combine different reasoning components to improve the system's performance over time based on its experiences.
DALL-E 2 - OpenAI imagery automation first developed by Vishal Coodye in 2021...MITAILibrary
The document provides a review of machine learning interpretability methods. It begins with an introduction to explainable artificial intelligence and a discussion of key concepts like interpretability and explainability. It then presents a taxonomy of interpretability methods that are divided into four main categories: methods for explaining black-box models, creating white-box models, promoting fairness, and analyzing model sensitivity. Specific machine learning interpretability techniques are summarized within each category.
The document describes a machine learning toolbox developed using Python that implements and compares several supervised machine learning algorithms, including Naive Bayes, K-nearest neighbors, decision trees, SVM, and neural networks. The toolbox allows users to test algorithms on various datasets, including Iris and diabetes data, and compare the accuracy results. Testing on these datasets showed Naive Bayes and K-nearest neighbors had the highest average accuracy rates, while neural networks and decision trees showed more variable performance depending on parameters and dataset splits. The toolbox is intended to help users evaluate which algorithms best fit their datasets.
The document discusses clustering documents using a multi-viewpoint similarity measure. It begins with an introduction to document clustering and common similarity measures like cosine similarity. It then proposes a new multi-viewpoint similarity measure that calculates similarity between documents based on multiple reference points, rather than just the origin. This allows a more accurate assessment of similarity. The document outlines an optimization algorithm used to cluster documents by maximizing the new similarity measure. It compares the new approach to existing document clustering methods and similarity measures.
WITH SEMANTICS AND HIDDEN MARKOV MODELS TO AN ADAPTIVE LOG FILE PARSERkevig
We aim to model an adaptive log file parser. As the content of log files often evolves over time, we
established a dynamic statistical model which learns and adapts processing and parsing rules. First, we
limit the amount of unstructured text by clustering based on semantics of log file lines. Next, we only take
the most relevant cluster into account and focus only on those frequent patterns which lead to the desired
output table similar to Vaarandi [10]. Furthermore, we transform the found frequent patterns and the
output stating the parsed table into a Hidden Markov Model (HMM). We use this HMM as a specific,
however, flexible representation of a pattern for log file parsing to maintain high quality output. After
training our model on one system type and applying it to a different system with slightly different log file
patterns, we achieve an accuracy over 99.99%.
Similar to Case-Based Reasoning for Explaining Probabilistic Machine Learning (20)
Climate Impact of Software Testing at Nordic Testing DaysKari Kakkonen
My slides at Nordic Testing Days 6.6.2024
Climate impact / sustainability of software testing discussed on the talk. ICT and testing must carry their part of global responsibility to help with the climat warming. We can minimize the carbon footprint but we can also have a carbon handprint, a positive impact on the climate. Quality characteristics can be added with sustainability, and then measured continuously. Test environments can be used less, and in smaller scale and on demand. Test techniques can be used in optimizing or minimizing number of tests. Test automation can be used to speed up testing.
Let's Integrate MuleSoft RPA, COMPOSER, APM with AWS IDP along with Slackshyamraj55
Discover the seamless integration of RPA (Robotic Process Automation), COMPOSER, and APM with AWS IDP enhanced with Slack notifications. Explore how these technologies converge to streamline workflows, optimize performance, and ensure secure access, all while leveraging the power of AWS IDP and real-time communication via Slack notifications.
Building RAG with self-deployed Milvus vector database and Snowpark Container...Zilliz
This talk will give hands-on advice on building RAG applications with an open-source Milvus database deployed as a docker container. We will also introduce the integration of Milvus with Snowpark Container Services.
Enchancing adoption of Open Source Libraries. A case study on Albumentations.AIVladimir Iglovikov, Ph.D.
Presented by Vladimir Iglovikov:
- https://www.linkedin.com/in/iglovikov/
- https://x.com/viglovikov
- https://www.instagram.com/ternaus/
This presentation delves into the journey of Albumentations.ai, a highly successful open-source library for data augmentation.
Created out of a necessity for superior performance in Kaggle competitions, Albumentations has grown to become a widely used tool among data scientists and machine learning practitioners.
This case study covers various aspects, including:
People: The contributors and community that have supported Albumentations.
Metrics: The success indicators such as downloads, daily active users, GitHub stars, and financial contributions.
Challenges: The hurdles in monetizing open-source projects and measuring user engagement.
Development Practices: Best practices for creating, maintaining, and scaling open-source libraries, including code hygiene, CI/CD, and fast iteration.
Community Building: Strategies for making adoption easy, iterating quickly, and fostering a vibrant, engaged community.
Marketing: Both online and offline marketing tactics, focusing on real, impactful interactions and collaborations.
Mental Health: Maintaining balance and not feeling pressured by user demands.
Key insights include the importance of automation, making the adoption process seamless, and leveraging offline interactions for marketing. The presentation also emphasizes the need for continuous small improvements and building a friendly, inclusive community that contributes to the project's growth.
Vladimir Iglovikov brings his extensive experience as a Kaggle Grandmaster, ex-Staff ML Engineer at Lyft, sharing valuable lessons and practical advice for anyone looking to enhance the adoption of their open-source projects.
Explore more about Albumentations and join the community at:
GitHub: https://github.com/albumentations-team/albumentations
Website: https://albumentations.ai/
LinkedIn: https://www.linkedin.com/company/100504475
Twitter: https://x.com/albumentations
Maruthi Prithivirajan, Head of ASEAN & IN Solution Architecture, Neo4j
Get an inside look at the latest Neo4j innovations that enable relationship-driven intelligence at scale. Learn more about the newest cloud integrations and product enhancements that make Neo4j an essential choice for developers building apps with interconnected data and generative AI.
Encryption in Microsoft 365 - ExpertsLive Netherlands 2024Albert Hoitingh
In this session I delve into the encryption technology used in Microsoft 365 and Microsoft Purview. Including the concepts of Customer Key and Double Key Encryption.
Cosa hanno in comune un mattoncino Lego e la backdoor XZ?Speck&Tech
ABSTRACT: A prima vista, un mattoncino Lego e la backdoor XZ potrebbero avere in comune il fatto di essere entrambi blocchi di costruzione, o dipendenze di progetti creativi e software. La realtà è che un mattoncino Lego e il caso della backdoor XZ hanno molto di più di tutto ciò in comune.
Partecipate alla presentazione per immergervi in una storia di interoperabilità, standard e formati aperti, per poi discutere del ruolo importante che i contributori hanno in una comunità open source sostenibile.
BIO: Sostenitrice del software libero e dei formati standard e aperti. È stata un membro attivo dei progetti Fedora e openSUSE e ha co-fondato l'Associazione LibreItalia dove è stata coinvolta in diversi eventi, migrazioni e formazione relativi a LibreOffice. In precedenza ha lavorato a migrazioni e corsi di formazione su LibreOffice per diverse amministrazioni pubbliche e privati. Da gennaio 2020 lavora in SUSE come Software Release Engineer per Uyuni e SUSE Manager e quando non segue la sua passione per i computer e per Geeko coltiva la sua curiosità per l'astronomia (da cui deriva il suo nickname deneb_alpha).
Unlock the Future of Search with MongoDB Atlas_ Vector Search Unleashed.pdfMalak Abu Hammad
Discover how MongoDB Atlas and vector search technology can revolutionize your application's search capabilities. This comprehensive presentation covers:
* What is Vector Search?
* Importance and benefits of vector search
* Practical use cases across various industries
* Step-by-step implementation guide
* Live demos with code snippets
* Enhancing LLM capabilities with vector search
* Best practices and optimization strategies
Perfect for developers, AI enthusiasts, and tech leaders. Learn how to leverage MongoDB Atlas to deliver highly relevant, context-aware search results, transforming your data retrieval process. Stay ahead in tech innovation and maximize the potential of your applications.
#MongoDB #VectorSearch #AI #SemanticSearch #TechInnovation #DataScience #LLM #MachineLearning #SearchTechnology
Full-RAG: A modern architecture for hyper-personalizationZilliz
Mike Del Balso, CEO & Co-Founder at Tecton, presents "Full RAG," a novel approach to AI recommendation systems, aiming to push beyond the limitations of traditional models through a deep integration of contextual insights and real-time data, leveraging the Retrieval-Augmented Generation architecture. This talk will outline Full RAG's potential to significantly enhance personalization, address engineering challenges such as data management and model training, and introduce data enrichment with reranking as a key solution. Attendees will gain crucial insights into the importance of hyperpersonalization in AI, the capabilities of Full RAG for advanced personalization, and strategies for managing complex data integrations for deploying cutting-edge AI solutions.
TrustArc Webinar - 2024 Global Privacy SurveyTrustArc
How does your privacy program stack up against your peers? What challenges are privacy teams tackling and prioritizing in 2024?
In the fifth annual Global Privacy Benchmarks Survey, we asked over 1,800 global privacy professionals and business executives to share their perspectives on the current state of privacy inside and outside of their organizations. This year’s report focused on emerging areas of importance for privacy and compliance professionals, including considerations and implications of Artificial Intelligence (AI) technologies, building brand trust, and different approaches for achieving higher privacy competence scores.
See how organizational priorities and strategic approaches to data security and privacy are evolving around the globe.
This webinar will review:
- The top 10 privacy insights from the fifth annual Global Privacy Benchmarks Survey
- The top challenges for privacy leaders, practitioners, and organizations in 2024
- Key themes to consider in developing and maintaining your privacy program
UiPath Test Automation using UiPath Test Suite series, part 6DianaGray10
Welcome to UiPath Test Automation using UiPath Test Suite series part 6. In this session, we will cover Test Automation with generative AI and Open AI.
UiPath Test Automation with generative AI and Open AI webinar offers an in-depth exploration of leveraging cutting-edge technologies for test automation within the UiPath platform. Attendees will delve into the integration of generative AI, a test automation solution, with Open AI advanced natural language processing capabilities.
Throughout the session, participants will discover how this synergy empowers testers to automate repetitive tasks, enhance testing accuracy, and expedite the software testing life cycle. Topics covered include the seamless integration process, practical use cases, and the benefits of harnessing AI-driven automation for UiPath testing initiatives. By attending this webinar, testers, and automation professionals can gain valuable insights into harnessing the power of AI to optimize their test automation workflows within the UiPath ecosystem, ultimately driving efficiency and quality in software development processes.
What will you get from this session?
1. Insights into integrating generative AI.
2. Understanding how this integration enhances test automation within the UiPath platform
3. Practical demonstrations
4. Exploration of real-world use cases illustrating the benefits of AI-driven test automation for UiPath
Topics covered:
What is generative AI
Test Automation with generative AI and Open AI.
UiPath integration with generative AI
Speaker:
Deepak Rai, Automation Practice Lead, Boundaryless Group and UiPath MVP
Unlocking Productivity: Leveraging the Potential of Copilot in Microsoft 365, a presentation by Christoforos Vlachos, Senior Solutions Manager – Modern Workplace, Uni Systems
How to Get CNIC Information System with Paksim Ga.pptxdanishmna97
Pakdata Cf is a groundbreaking system designed to streamline and facilitate access to CNIC information. This innovative platform leverages advanced technology to provide users with efficient and secure access to their CNIC details.
Introducing Milvus Lite: Easy-to-Install, Easy-to-Use vector database for you...Zilliz
Join us to introduce Milvus Lite, a vector database that can run on notebooks and laptops, share the same API with Milvus, and integrate with every popular GenAI framework. This webinar is perfect for developers seeking easy-to-use, well-integrated vector databases for their GenAI apps.
In the rapidly evolving landscape of technologies, XML continues to play a vital role in structuring, storing, and transporting data across diverse systems. The recent advancements in artificial intelligence (AI) present new methodologies for enhancing XML development workflows, introducing efficiency, automation, and intelligent capabilities. This presentation will outline the scope and perspective of utilizing AI in XML development. The potential benefits and the possible pitfalls will be highlighted, providing a balanced view of the subject.
We will explore the capabilities of AI in understanding XML markup languages and autonomously creating structured XML content. Additionally, we will examine the capacity of AI to enrich plain text with appropriate XML markup. Practical examples and methodological guidelines will be provided to elucidate how AI can be effectively prompted to interpret and generate accurate XML markup.
Further emphasis will be placed on the role of AI in developing XSLT, or schemas such as XSD and Schematron. We will address the techniques and strategies adopted to create prompts for generating code, explaining code, or refactoring the code, and the results achieved.
The discussion will extend to how AI can be used to transform XML content. In particular, the focus will be on the use of AI XPath extension functions in XSLT, Schematron, Schematron Quick Fixes, or for XML content refactoring.
The presentation aims to deliver a comprehensive overview of AI usage in XML development, providing attendees with the necessary knowledge to make informed decisions. Whether you’re at the early stages of adopting AI or considering integrating it in advanced XML development, this presentation will cover all levels of expertise.
By highlighting the potential advantages and challenges of integrating AI with XML development tools and languages, the presentation seeks to inspire thoughtful conversation around the future of XML development. We’ll not only delve into the technical aspects of AI-powered XML development but also discuss practical implications and possible future directions.
Case-Based Reasoning for Explaining Probabilistic Machine Learning
1. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
DOI:10.5121/ijcsit.2014.6206 87
Case-Based Reasoning for Explaining Probabilistic
Machine Learning
Tomas Olsson1,2
, Daniel Gillblad2
, Peter Funk1
, and Ning Xiong1
1
School of Innovation, Design, and Engineering, Mälardalen University, Västerås, Sweden
2
SICS Swedish ICT, Isafjordsgatan 22, Box 1263, SE-164 29 Kista, Sweden
ABSTRACT
This paper describes a generic framework for explaining the prediction of probabilistic machine learning
algorithms using cases. The framework consists of two components: a similarity metric between cases that
is defined relative to a probability model and an novel case-based approach to justifying the probabilistic
prediction by estimating the prediction error using case-based reasoning. As basis for deriving similarity
metrics, we define similarity in terms of the principle of interchangeability that two cases are considered
similar or identical if two probability distributions, derived from excluding either one or the other case in the
case base, are identical. Lastly, we show the applicability of the proposed approach by deriving a metric for
linear regression, and apply the proposed approach for explaining predictions of the energy performance of
households.
KEYWORDS
Case-based Reasoning, Case-based Explanation, Artificial Intelligence, Decision Support, Machine
Learning
1. INTRODUCTION
Explanation has been identified as a key factor for user acceptance of an intelligent system [1–
6]. If a user does not understand or trust a decision support system, it will be less likely that the
system will be accepted. For instance, in the medical domain it is well-known that a good
prediction performance will not automatically mean user acceptance unless the physicians
understand the reasoning behind [7]. In this paper, we propose a novel approach to using case-
based reasoning (CBR) as an intuitive approach for justifying (explaining) the predictions of a
probabilistic model as a complement to traditional statistical measures of uncertainty such as the
mean value and the variance.
CBR is a conceptually simple and intuitive, but yet powerful approach for knowledge
management and learning [8, 9]. In contrast to model-based approaches in machine learning and
statistics, inference in CBR is done directly from a set of cases without generalizing to a model.
The fundamental idea in CBR is that similar problems have similar solutions and therefore, new
solutions can be created from previous solutions. Traditionally, CBR is not used if there is a
sufficiently good model-based solution to a problem. Yet, CBR has some advantages that
complement model-based learning approaches. For instance, a probabilistic machine learning
model can be hard to understand for non-experts while CBR is conceptually much more intuitive
and easy to explain. Therefore, by complementing a probabilistic model with a CBR-based
explanation facility, we can make the system more understandable.
2. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
88
Explanation using preceding cases has some advantages compared to other approaches. For
instance, it has been shown in a user experiment that users in some domains prefer case-based
over rule-based explanation [10]. In [11], Nugent et al. list three advantages of case-based
explanation. Firstly, it is a natural form of explanation in many domains where explanation by
analogy is common. Second, it uses real evidence in form of a set of cases relevant to the task at
hand. This, we argue in this paper, is the key strength of combining probabilistic methods with
CBR for explanation. Lastly, it is a fixed and simple form of explanation that is directly related
to the problem at hand. Thus, regardless of the complexity of the problem at hand, the content of
the explanation can be kept rather constant.
The purpose of explaining a system can vary depending on which type of user the explanation is
addressing. For instance, Sørmo et al. list five common types of explanations: transparency,
learning, conceptualization, relevance, and justification [12]. The goal of transparency is to
explain how the system computed the prediction. This goal is more relevant for expert users that
are able to assess the reasoning process by themselves. The goal of learning is to help novice
users to learn the application domain. The goal of conceptualization is to help novice users to
understand the meaning of concepts used in the system. The explanation goal relevance is about
explaining why the system does something such as asking a question for more information. In
this paper, we only consider the last type of explanation – justification – where the goal is to
support a non-expert user in assessing the reliability of the system predictions using CBR.
The work in the current paper presents developments in line with previous work in knowledge
light CBE that uses cases to explain model-based machine learning algorithms [13–15].
However, while previous work define similarity metrics ad-hoc and by intuitive means, we differ
by defining similarity metrics with a good theoretical foundation and with a clear meaning. This
can be achieved by restricting the application to learning algorithms that result in probability
models. In probabilistic machine learning, the inference is visible in terms of probability
distributions. So by analyzing the probability distributions from two different probability models,
we can draw some conclusions of the relation between them. This probabilistic assumption makes
the proposed approach generically applicable to any probabilistic machine learning algorithm.
In addition, we also differ with respect to previous work in how a prediction is explained. Since
the probability model is most likely trained on all cases, it would not be sufficient to justify the
system performance by only showing a list of the most similar cases. A list of similar cases would
lead the user into checking each case and compare its difference or similarity to the new case.
This would be very time consuming and would not say much of the system performance at a
larger scale. As justification of the system performance, we instead propose using the average
prediction error for all similar preceding cases. This is a straightforward application of CBR
applied to predicting the probabilistic prediction error.
The rest of the paper is organized as follows. Sect. 2 presents related work in case-based
explanation. In Sect. 3, we give background to similarity metrics and statistical metrics. Sect. 4
presents the overall framework for explanation by cases. We present a definition of similarity in
terms of probability distributions and our approach to case-based explanation. We also derive a
similarity metric for linear regression. Sect. 5 describes the application of the proposed approach
to explaining the prediction of the energy performance of a household. In Sect. 6, we make
concluding remarks and describe future work.
2. CASE-BASED EXPLANATION
Case-based explanation (CBE) is a research field within CBR that investigates the use of cases for
explaining systems [16–18, 5]. CBE can, similarly to CBR, be divided into knowledge intensive
3. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
89
and knowledge light CBE where the former makes use of explicit domain knowledge while the
latter uses mainly knowledge already contained in the similarity metric and the case base [10].
The main work in knowledge light CBE has been in explaining classification while less work
has been invested in explaining regression. The current work is an instance of knowledge light
CBE where no explicit explanation mechanisms are modeled and the considered learning task is
regression.
Furthermore, CBE research differs in how cases are explained. One type of knowledge light CBE
research investigates the use of other types of learning methods for explanation. The ProCon
system described in [19, 20] uses a naive Bayes classifier trained on all cases to find which
features of a case support or oppose the classification. The author argues that even the most
similar case can some times contain information that contradicts the prediction and that must be
made explicit in order to keep the confidence of the user. The system presented in [21] by the
same author generates rules from the nearest neighbors in order to explain the retrieved cases.
Similarly to the previous system, it is ensured that the learned rules subsumes both cases that
supports and opposes the classification. Case-based explanation of a lazy learning approach in the
same vein for classifying chemical compounds was presented in [23]. The approach lets the user
compare the molecular structures using the similarities between cases with respect to cases from
different classes. Thus, by showing the user similarities only common to cases of each class the
user is also able to understand the difference between classifications. In [22], the authors
describes a system that applies logistic regression to a set of retrieved cases and uses the logistic
model to explain the importance of features and assign a probabilistic confidence measure of the
system’s prediction.
A second type of research considers which cases to present to a user for explaining classification
of new cases. This research was started when it was noticed that the set of cases used for making
the classification is not necessary also the best cases for explaining the classification. Instead of
presenting the most similar case, it might be better to show a case close to the decision border
between two classes. In [24], the authors compare similarity metrics optimized for explanation
with similarity metrics optimized for classification, while in [25], the authors use the same
similarity metric as for classification but explore different rules for selecting which case to use as
explanation. In [22], logistic regression is used to find cases close to the classification border. A
more recent work describing all these three approaches is presented in [11].
A third type of knowledge light CBE research has addressed the problem of explaining model-
based machine learning methods using cases, but so far, mainly neural networks have been
investigated [13, 14, 26, 15, 27, 28]. The first knowledge light CBE for model-based learning
algorithms was presented in [13, 14]. In the first paper, the author sketches ideas on how to use the
model of a neural network or a decision tree as a similarity metric. In case of neural networks the
activation difference between two cases was proposed as a metric while the leaves in the decision
tree naturally contain similar cases. In [26] the authors explain the prediction of an ensemble of
neural networks using rules extracted from the network. Then, for a new case, only rules relevant
for explaining that case are used. The rules are filtered using heuristic criteria. The work presented
in [27, 28] explains the output of an ensemble of neural networks using the most important feature
values relative to a case. In [15], a generic knowledge light CBE framework for black-box
machine learning algorithms is presented. The authors trained a locally weighted linear model to
approximate a neural network using artificial cases generated from the neural network. Then they
used the coefficients of the linear model both as feature weights of a similarity metric to retrieve
relevant cases and for explaining which features are most import for a prediction.
4. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
90
3. PRELIMINARIES
In this section, we define the notion of a true metric that is important in order to index cases for
fast retrieval. In addition, we present the J-divergence that is a statistical measure of similarity
between probability distribution that we use in our definition of similarity between cases.
3.1. True Metrics
In order to make fast retrieval of cases similarity metrics should adhere to the axioms of a true
metric. Given a true metric, the search space can be partitioned into smaller regions and
organized so that there is no need to search through all regions.
In this paper, we use the term metric informally as any function that makes a comparison
between two cases, while a true metric is a metric in a mathematical sense. This means that a true
metric is a function d that satisfy the following three axioms where X denotes the case base with
the set of all cases:
1. d(x, y) ≥ 0 (non-negative and identity) with d(x,y) = 0 if and only if x = y, x,y ∈ X
2. d(x, y) = d(y, x) (symmetric) for all x,y ∈ X
3. d(x, z) ≤ d(x, y) + d(y,z) (triangle inequality) for all x, y, z ∈ X
There is a discussion in the CBR literature whether all of the above axioms are required for useful
similarity and distance metrics [29, 30]. A common true metric that we will use in this paper is
the Manhattan distance. The definition of the Manhattan distance is
d(x, y) = ∑|xk
− yk
|
k
where | ...| denotes the absolute value function.
3.2. Statistical Metrics
A commonly used statistical metric for comparing two probability distributions is the well-
known Kullback-Leibler divergence (KL) [31]. KL is also sometimes called the relative entropy
or the information gain, since it is closely related to the entropy concept introduced by Shannon
[32, 33]. The KL for the two probability distributions pi, pj, with parameter θ, is
In case of discrete variables, the integral is replaced with a sum.
KL is not symmetric but by computing the KL divergence in both directions and then add them
together we get a symmetric metric. This is an important characteristic if we desire a true metric
as described in Sect. 3.1. The symmetric KL is often called Jeffreys divergence (J-divergence). The
J-divergence will then be:
5. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
91
In this paper, we use the J-divergence as basis for the similarity metrics, because it is a commonly
used measure and it has a clear information theoretical interpretation. Other statistical metrics for
comparing distributions are also available such as the total variation distance, the Euclidean
distance and the Jensen-Shannon divergence [34, 31,35–37].
4. THE CASE-BASED EXPLANATION FRAMEWORK
In this paper, we propose a generic case-based explanation framework that justifies the predictions
of an intelligent system by estimating the prediction reliabilitycase by case. We have restricted the
approach to probabilistic machine learning methods, since that gives the framework a good
theoretical foundation. However, the explanation part can in principle also be used for any
learning algorithm.
Assume that we have trained a probability model for predicting an unknown variable, and that we
have derived a relevant similarity metric. Then, the framework creates a case-based explanation
according to the following pseudo-algorithm:
1. Make prediction for a new case using the probability model
2. Retrieve most similar previous cases
3. For each previous case
a. Make prediction for the case using the probability model
b. Compute the absolute prediction error given the ground truth
4. Estimate the prediction error for the new case as the average of previous prediction errors
5. Present predicted value and estimated prediction error to the user
We will apply this framework to a real example in Sect. 5 where we explain the predictions from
a linear regression model of the energy performance of a household. However, before that, we
will describe the approach in more detail in the rest of this section. First, in Sect. 4.1, we describe
a generic approach to defining a similarity metric using methods from statistics and machine
learning. Then, Sect 4.2 shows how a similarity metric based on the linear regression model can
be derived. Last, Sect. 4.3 describes a CBR approach to estimating the prediction error.
4.1. A Statistical Measure of Similarity
In this section, we present a principled approach to defining similarity metrics using methods
from statistics and machine learning.
As a means to relate the similarity between two cases to probability models, we have formulated
the principle of interchangeability as a general definition. We define the principle of
interchangeability as follows:
Definition 1. Two cases xi , x j in case base X are similar if they can be interchanged such that
the two probability distributions Pi , Pj inferred from excluding xi and x j respectively from the
case base – X xi and X xj – are identical with respect to some parameter(s) of interest.
6. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
92
i
We have chosen to use the J-divergence presented in Sect. 3.2 for comparing two probability
distributions. The J-divergence distance between two cases xi , xj in case base X is
where pi (θ) and p j (θ) are probability distributions derived from the case base when excluding
the cases xi and xj respectively and θ are the parameters that define in what aspects cases are
similar.
The resulting measure between two cases can then be interpreted information theoretically as the
sum of the information gain from including one case in the case base over the other case and the
information gain from including the other case in the case base over the first case. However, the
resulting J-divergence distance is not a true metric, so an additional step might be needed that
turns it into a final distance that fulfills the axioms of a true metric.
4.2. Derivation of a Similarity Metric for Linear Regression
In this section, we derive a similarity metric for linear regression showing that this simple
statistical model leads to a simple metric based on the Manhattan distance. The derived distance
metric is the ratio between the distribution parameters plus the square of the sum of the weighted
differences between case features.
In linear regression, we assume that each case xi in case base X can be modeled as a weighted
sum of a feature vector:
K
yi = ω0 + ∑ ωkxk
+ εi (4)
k=1
where εi are the residual error and the ω is a weight vector, and an unknown value y for a new case
x is estimated by:
K
yˆ = ω0 + ∑ ωkxk
(5)
k=1
Let ε be normally distributed with mean 0 and standard deviation σ and then, the predictive
distribution, conditioned on the weights and the standard deviation from a point estimation will
be denoted as:
Theorem 1. Let pi (y|x) and p j (y|x) be the predictive probability distributions for linear
regression derived from excluding xi and x j respectively. Let ωi , σi , ω j, σ j be the corresponding
point estimations for the parameters of the distributions. Then, it can be shown that the J-
divergence distance for two cases xi, xj is:
Remark 1. Notice that a special case of Eq. 7 is when σi ≈ σj ≈ σ and ωi ≈ ωj ≈ ω , in which case
the distance metric becomes:
7. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
93
which is approximatelytrue with a large number of cases.
If we ignore the difference between the standard deviations, the resulting metric is square of the
difference between the predicted value of yˆi and yˆj for case xi and xj respectively. The square
root of this is a true metric, with respect to the predicted values, but not with respect to the cases.
As a consequence, two cases that lead to the same predicted values would be considered similar.
If a true metric is desirable, we have to ensure that different features do not cancel out each other.
One way of doing this is by rewriting the distance metric as follows:
where P and N are all features k where the contribution ω
k
(xk − xk ) to the sum is positive and
negative respectively. Then, by then taking the square root of the last metric, we get a true metric
that is a weighted version of the Manhattan distance:
So, the absolute weights are indicating the importance of each feature, which is similar to how
the weights of a locally weighted linear regression model are used in [15]. Thereby, we can
theoretically justify the use of the regression weights in a distance metric.
4.3. Estimating the Prediction Error
In this paper, we have formulated case-based explanation as a regression problem that we solve
using CBR. Thus, by retrieving a set of similar cases, we can estimate the error of the prediction
for a new case. Below, we describe the simple average prediction estimation that we use for
estimating the prediction error.
The average prediction error approach is a simple application of the k nearest neighbor algorithm
[9]. Thus, given a new case xi , we can retrieve the set of k most similar cases using the metric in
Eq. 10. Thereafter, we can estimate the prediction error ei as the average prediction error of the
retrieved cases:
8. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
94
where ej = |yˆj − y j| is the actual prediction error, yˆj is the predicted value for case xj from Eq. 12
and yj is the true value.
5. EXPLAINING PREDICTED ENERGY PERFORMANCE OF A HOUSEHOLD
In this section, we apply the proposed framework to explaining the prediction of energy
performance of households. The energy performance of a household is measured in kilowatt
hours per square meter and year, and corresponds to the energy need in a building for space and
hot water heating, cooling, ventilation, and lighting based on standard occupancy. The energy
performance has been computed by a certified energy and it is intended to make it easy to
compare houses when selling and buying a house.
The rest of this section is organized as follows. Sect. 5.1 describes the used data set. Sect. 5.2
presents which features we have used and the result from fitting a linear regression model to the
data. Sect. 5.3 describes the implementation and evaluation of the proposed case-based
explanation framework for energy performance prediction. Finally, Sect. 5.4 ends this section by
showing examples of case-based explanation for two households.
5.1. Energy Performance Data
The data that we have used comes from around 1800 energy reports collected in the ME3Gas
project [38,41]. The energy report consists of four attributes related to the building and location of
a house, which are shown in the upper part of Table 1, and energy measurements for 12 different
heating system types, which are shown in Table 2. Each used energy system type is measured in
kWh. In addition, there are also the time period when measuring was done and the energy
performance measurement that are shown in the second part of Table 1.
Table 1. The building and location attributes with summary of the data followed by measuring
period start and energy performance. Climate zone indicate location in Sweden: zone 1 is in the
most northern part and 4 is in the most southern part of Sweden.
House attribute Summary
Year of construction
Climate zone 1-4
House type
Size of heated area (m2)
mean: 1956, min: 1083, max: 2013
90, 248, 934, and 515 in each zone 1-4
#Detached: 1727, #Terraced: 57
mean: 177, min: 38, max: 925
Start dates of measuring period (year)
Energy performance
mean: 2011, min: 2007, max: 2013
mean: 111, min: 24, max: 388
9. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
95
Table 2. Heating systems and how many households with each.
Heating System
District heating
Heating oil
Natural gas
Firewood
Wood chips/pellets
Electric water-borne
200
75
19
512
133
316
Electric direct Acting
Electric airborne
Geothermal heat pump electrical
Exhaust air heat pump
Heat pump Air Air
Heat pump Air Water
738
50
317
147
417
137
5.2. Log-Linear Regression Model Fitting
In this section, we first select which features to use and then fit a linear regression model to the
energy performance data. Our first observation is that the energy performance is not normally
distributed, but log-normally distributed. This means that the logarithm of the energy
performance is normally distributed. In addition, all relations between features and the log of the
energy are not linear. Thus, the following set of new features were added that capture non-linear
relations: age of the house when the measuring was started, log of age that is the natural
logarithm of age, log of climate zone, and log of heated area. For the household heating systems,
we assume that it is only known which types of heating systems a new household uses, not how
much energy is used by each heating system. Each heating system is therefore represented as 1 if
present or 0 if not.
A linear regression model was then fitted to the data using the ridge regularization
implementation of the scikit-learn project [39]. This resulted in a log-linear regression model
with the weights listed in Table 3. The energy performance can then be predicted as below, using
the exponential power of the result from Eq. 5 plus an extra term (s2
/2) that is an adjustment for the
bias of the log-normal model:
where s2
is the estimated standard error of the predicted value yˆ. The distance metric would still
be the same as in Eq. 10, but with the extra features added to the cases, and that the distance
metric is defined with respect to the distribution of log(y) and not y.
10. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
96
Table 3. The regression weight of each feature.
House Characteristics Weight (ω) Heating Systems Weight (ω)
Year of construction
Age
Log of Age
Climate zone
Log of Climate zone
Detached house
Terraced house
Heated area
Log of Heated Area
0.002
0.002
0.107
-0.021
-0.064
0.068
0.064
0.000
-0.221
District heating
Heating oil
Natural gas
Firewood
Wood chips/pellets
Electric water-borne
Electric direct Acting
Electric airborne
Geothermal heat pump electrical
Exhaust air heat pump
Heat pump Air Air
Heat pump Air Water
0.142
0.190
0.040
0.150
0.263
0.100
0.008
0.036
-0.441
-0.046
-0.159
-0.197
5.3. Evaluation of the Estimation of the Prediction Error
In this section, we evaluate the estimation approach proposed in Sect 4.3 for estimating the
prediction error of the log-linear regression model. We compare the estimated error to the true
error.
In the experiment, we split the data set 10 times into three sets: a 60% training set, a 20% validation
set and a 20% test set. Each time the log-linear model is trained on the training set. Then, the k
nearest neighbor algorithm was trained on the training set and configured to use the distance
metric from Sect. 5.2 together with average prediction error approaches from Sect. 4.3.
Thereafter, the performance of the k nearest neighbor algorithm was measured on both the
validation set and the test set. The results from all data splits were then averaged. The validation
set is used for selecting the algorithm parameters k. The performance is measured using root mean
square error (RMSE).
In Figure 1, we show the average RMSE of the estimated prediction error on a validation set and
a test set. The figure shows that the performance of both the test and the validation set is quite
consistent. The minimum error of the validation set is located somewhere between k = 25 and k =
45, so we can select k = 35 as the size of the case set that will be used for explaining the
prediction. However, we should also consider the experience of the users. If the users distrust the
explanation because we have selected a too small set of previous cases, we should consider
selecting a larger k. For instance, the difference in performance between k = 35 and k = 60 is
quite small, so the latter could be preferred in some situations if it is considered more convincing.
11. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
97
Figure 1. Number of k nearest neighbors (x-axis) versus the root mean square error (y-axis) of the
estimated prediction error.
5.4. Case-based Explanation Examples
This section presents two examples of case-based explanations for two different households using
the average prediction error as estimation with k = 60 neighbors. The two examples are shown in
Table 4 and Table 5 respectively. Beginning from top of the tables. First, the characteristics of the
house are listed, and then the used heating systems. Thereafter, the predicted value is shown,
together with the true value that is assumed to be unknown but is here shown for comparison.
Last, we explain the prediction in words, where we classify a estimated prediction error to be
low if less than 10% of the predicted value, medium high if less than 20%, quite high if less
than 30%, high if less than 40% and very high otherwise. In classifying the estimated
prediction error, we use background knowledge in that low values are better than large and that
the energy performance should be larger than zero. Thus, a relative value is an intuitive means of
assessing the severity of the error. As can be seen from the examples, the explanations show that
the prediction errors are medium high or high and that the prediction are not completely reliable,
which can be confirmed by looking at the value of the true energy performance compared to the
predicted value (especially the second household). However, the reason that the prediction of the
second household is so bad is that there are very few houses with that combination of heating
systems, especially wood chips/pellets, which can be easily seen by in addition listing the most
similar cases. In Table 6 are the three most similar households listed, and as can be seen, none of
them are very similar to example 2.
12. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
98
Table 4. Household example 1
Feature Value
Year of construction
Climate zone 1-4
House type
Size of heated area
1977
2
Detached house
215 m2
Electric direct Acting
Heat pump Air Air
Yes
Yes
Predicted energy performance
True energy performance
84.0 kWh/m2
90 kWh/m2 (Unknown)
Explanation: The predicted energy performance of this house is 84.0 kWh/m2.
The average prediction error for the 60 most similar houses is 14.5 kWh/m2. That
is about 17.3% of the predicted energy performance, which is medium high.
Table 5. Household example 2
Year of construction
Climate zone 1-4
House type
Size of heated area
1961
1
Detached house
100
Wood chips/pellets
Electric water-borne
Heat pump Air Air
Yes
Yes
Yes
Predicted energy performance
True energy performance
185.6 kWh/m2
81 kWh/m2 (Unknown)
Explanation: The predicted energy performance of this house is 185.6 kWh/m2.
The average prediction error for the 60 most similar houses is 57.5 kWh/m2. That
is about 31.0% of the predicted energy performance, which is high.
13. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
99
Table 6. The three most similar cases to household example 2.
Feature Case 1 Case 2 Case 3
Year of construction 1968 1909 1987
Heated area 105 107 112
Detached Yes Yes Yes
Climate zone 1 2 3
Heating system Firewood, Electric
direct Acting
Electric water-
borne, Heat
pump Air Air
District heating
Predicted energy
performance
186.0 128.7 137.3
True energy
performance
172 132 128
6. CONCLUSIONS AND FUTURE WORK
In this paper, we have proposed a framework for knowledge light case-based explanation of
probabilistic machine learning. The first contribution of this work is a principled and theoretically
well-founded approach to defining a similarity metric for retrieving cases relative a probability
model.
As a second contribution, we have proposed a novel approach to justifying a prediction by
estimating the prediction error as the average prediction error of the most similar cases. Since the
justification is based on real cases and not merely on the correctness of the probability model, we
argue that this is a more intuitive justification of the reliability than traditional statistical measures.
However, it should be regarded as a complement to traditional measures rather than a replacement.
The work in this paper can be developed further in many directions. Clearly, we could develop
case-based explanation approaches for other types of probability models. Especially, we would
like to extend this approach to classification tasks. In addition, in order to evaluate the proposed
explanation any further we would need to conduct user studies where we let real users use different
versions of explanations and assess the effectiveness of each approaches.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the funding from the Swedish Knowledge Foundation (KK-
stiftelsen) [40] through ITS-EASY Research School, the European ARTEMIS project ME3Gas
(JU Grant Agreement number 100266), the ITEA 2 (ITEA 2 Call 5, 10020) and Swedish
Governmental Agency for Innovation Systems (VINNOVA) grant no 10020 and JU grant no
100266 making this research possible. The energy performance data is used by courtesy of CNet
Svenska AB [41].
14. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
100
REFERENCES
[1] Wick, M.R., Thompson, W.B, (1992) “Reconstructive expert system explanation”, Artificial
Intelligence 54(1), pp33–70
[2] Ye, L.R., Johnson, P.E. (1995) “The impact of explanation facilities on user acceptance of expert
systems advice”, MIS Quarterly, pp157–172
[3] Gregor, S., Benbasat, I. (1999) “Explanations from intelligent systems: Theoretical foundations and
implications for practice”, MIS quarterly, pp497–530
[4] Lacave, C., Diez, F.J. ) (2004) “A review of explanation methods for heuristic expert systems”, The
Knowledge Engineering Review 19(02, pp133–146
[5] Leake, D., McSherry, D. (2005) “Introduction to the special issue on explanation in case-based
reasoning”, Artificial Intelligence Review 24(2), pp103–108
[6] Darlington, K. (2013) “Aspects of intelligent systems explanation”, Universal Journal of Control and
Automation 1, pp40 – 51
[7] Langlotz, C.P., Shortliffe, E.H. (1983) “Adapting a consultation system to critique user plans”,
International Journal of Man-Machine Studies 19(5), pp479–496
[8] Aamodt, A., Plaza, E. (1994) “Case-based reasoning: Foundational issues, methodological variations,
and system approaches”, AI communications 7(1), pp39–59
[9] Aha, D., Kibler, D., Albert, M. (1991) “Instance-based learning algorithms”, Machine learning 6(1),
pp37–66
[10] Cunningham, P., Doyle, D., Loughrey, J. (2003) “An evaluation of the usefulness of case-based
explanation”, In: Case-Based Reasoning Research and Development. Springer, pp122–130
[11] Nugent, C., Doyle, D., Cunningham, P. (2009) “Gaining insight through case-based explanation”,
Journal of Intelligent Information Systems 32(3), pp267–295
[12] Sørmo, F., Cassens, J., Aamodt, A. (2005) “Explanation in case-based reasoning–perspectives and
goals”, Artificial Intelligence Review 24(2), 109–143
[13] Caruana, R., Kangarloo, H., Dionisio, J., Sinha, U., Johnson, D. (1999) “Case-based explanation of
non-case-based learning methods”, In: Proceedings of the AMIA Symposium, American Medical
Informatics Association, pp212
[14] Caruana, R. (2000) “Case-based explanation for artificial neural nets”, In: Artificial Neural Networks
in Medicine and Biology. Springer, pp303–308
[15] Nugent, C., Cunningham, P. (2005) “A case-based explanation system for black-box systems”,
Artificial Intelligence Review 24(2), pp163–178
[16] Schank, R.C., Leake, D.B. (1989) “Creativity and learning in a case-based explainer”, Artificial
Intelligence 40(1), pp353–385
[17] Aamodt, A. (1994) “Explanation-driven case-based reasoning”, In: Topics in case-based reasoning.
Springer, pp274–288
[18] Doyle, D., Tsymbal, A., Cunningham, P. (2003) “A review of explanation and explanation in case-
based reasoning”, Dublin, Trinity college https://www.cs.tcd.ie/publications/techreports/reports 3
[19] McSherry, D. (2003) “Explanation in case-based reasoning: an evidential approach”, In: Proceedings
of the 8th UK Workshop on Case-Based Reasoning, pp47–55
[20] McSherry, D. (2004) “Explaining the pros and cons of conclusions in CBR”, In: Advances in Case-
Based Reasoning. Springer, pp317–330
[21] McSherry, D. (2012) “A lazy learning approach to explaining case-based reasoning solutions”, In:
Case-Based Reasoning Research and Development. Springer, pp241–254
[22] Nugent, C., Cunningham, P., Doyle, D. (2005) “The best way to instil confidence is by being right”,
In: Case-Based Reasoning Research and Development. Springer, pp368–381
[23] Armengol, E. (2007) “Discovering plausible explanations of carcinogenecity in chemical
compounds”, In: Machine Learning and Data Mining in Pattern Recognition. Springer, pp756–769
[24] Doyle, D., Cunningham, P., Bridge, D., Rahman, Y. (2004) “Explanation oriented retrieval”, In:
Advances in Case-Based Reasoning. Springer, pp157–168
[25] Cummins, L., Bridge, D.: Kleor (2006) A knowledge lite approach to explanation oriented retrieval.
Computing and Informatics 25(2-3), pp173–193
[26] Wall, R., Cunningham, P., Walsh, P. (2002) “Explaining predictions from a neural network ensemble
one at a time”, In: Principles of Data Mining and Knowledge Discovery. Springer, pp449–460
[27] Green, M., Ekelund, U., Edenbrandt, L., Bjo¨rk, J., Hansen, J., Ohlsson, M. (2008) “Explaining
artificial neural network ensembles: A case study with electrocardiograms from chest pain patients”,
15. International Journal of Computer Science & Information Technology (IJCSIT) Vol 6, No 2, April 2014
101
In: Proceedings of the ICML/UAI/COLT 2008 Workshop on Machine Learning for Health-Care
Applications.
[28] Green, M., Ekelund, U., Edenbrandt, L., Bjo¨rk, J., Forberg, J.L., Ohlsson, M. (2009) “Exploring new
possibilities for case-based explanation of artificial neural network ensembles”, Neural Networks
22(1), pp75–81
[29] Burkhard, H.D., Richter, M.M. (2001) “On the notion of similarity in case based reasoning and fuzzy
theory”, In: Soft computing in case based reasoning. Springer , pp29–45
[30] Burkhard, H.D. (2001) “Similarity and distance in case based reasoning”, Fundamenta Informaticae
47(3), pp201 – 215
[31] Kullback, S., Leibler, R.A (1951) “On information and sufficiency”, The Annals of Mathematical
Statistics 22(1), pp79–86
[32] Ihara, S. (1993) “Information theory for continuous systems”, Volume 2. World Scientific
[33] Shannon, C.E. (2001) “A mathematical theory of communication”, ACM SIGMOBILE Mobile
Computing and Communications Review 5(1), pp3–55
[34] Rachev, S. T., Stoyanov, S. V., & Fabozzi, F. J. (2011) “A Probability Metrics Approach to Financial
Risk Measures”, John Wiley & Sons.
[35] Lin, J. (1991) “Divergence measures based on the Shannon entropy”, Information Theory, IEEE
Transactions on 37(1), pp145–151
[36] Cha, S. H. (2007) “Comprehensive survey on distance/similarity measures between probability
density functions”, City, 1(2), 1.
[37] Dragomir, S.C. (2008) “Some properties for the exponential of the Kullback-Leibler divergence”,
Tamsui Oxford Journal of Mathematical Sciences 24(2), pp141–151
[38] Me3gas – smart gas meters & middleware for energy efficient embedded services. url:
http://www.me3gas.eu
[39] Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M.,
Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M.,
Perrot, M., Duchesnay, E. (2011) “Scikit-learn: Machine learning in Python”, Journal of Machine
Learning Research 12, pp2825–2830
[40] KK-Stiftelse: Swedish Knowledge Foundation. http://www.kks.se (Last Accessed: April 2014)
[41] CNet Svenska AB. http://www.cnet.se (Last Accessed: April 2014)