Computer Sciences Commons^™

A Quantitative Evaluation Of Global, Rule-Based Explanations Of Post-Hoc, Model Agnostic Methods, Giulia Vilone, Luca Longo

Articles

Understanding the inferences of data-driven, machine-learned models can be seen as a process that discloses the relationships between their input and output. These relationships consist and can be represented as a set of inference rules. However, the models usually do not explicit these rules to their end-users who, subsequently, perceive them as black-boxes and might not trust their predictions. Therefore, scholars have proposed several methods for extracting rules from data-driven machine-learned models to explain their logic. However, limited work exists on the evaluation and comparison of these methods. This study proposes a novel comparative approach to evaluate and compare the …

Predicting The Adoption Of Password Managers: A Tale Of Two Samples, Shelia Kennison, D. Eric Chan-Tin

Computer Science: Faculty Publications and Other Works

Using weak passwords and re-using passwords can make one vulnerable to cybersecurity breaches. Cybersecurity experts recommend the adoption of password managers (PMs), as they generate and store strong passwords for all accounts. Prior research has shown that few people adopt PMs. Our research examined PM adoption in a sample of 221 undergraduates from psychology courses and a sample of 278 MTurk workers. We hypothesized that PM adoption could be predicted using a small set of user characteristics (i.e., gender, age, Big Five personality traits, number of devices used, frequency of using social media, and cybersecurity knowledge). The results showed that …

From Mdp To Alphazero, David Robert Sewell

Dissertations and Theses

In this paper I will explain the AlphaGo family of algorithms starting from first principles and requiring little previous knowledge from the reader. The focus will be upon one of the more recent versions AlphaZero but I hope to explain the core principles that allowed these algorithms to be so successful. I will generally refer to AlphaZero as theses [sic] core set of principles and will make it clear when I am referring to a specific algorithm of the AlphaGo family. AlphaZero in short combines Monte Carlo Tree Search (MCTS) with Deep learning and self-play. We will see how these …

Improving Accurate Candidates For Missing Data Using Benefit Performance Of (Ml-Som), Abeer Abdullah Al-Mohdar, Mohamed Abdullah Bamatraf

Hadhramout University Journal of Natural & Applied Sciences

Missing data is one of the major challenges in extracting and analyzing knowledge from datasets. The performance of training quality was affected by the appearance of missing data in a dataset. For this reason, there is a need for a quick and reliable method to find possible solutions in order to provide an accurate system. Therefore, the previous studies provided robust ability of Self Organizing Map (SOM) algorithm to deal with the missing values [6, 20]. However, it has a drawback such as an error rate(ERR) in the missing values that increase huge dataset. This study is mainly based on …

Why Moments (And Generalized Moments) Are Used In Statistics And Why Expected Utility Is Used In Decision Making: A Possible Explanation, R. Noah Padilla, Vladik Kreinovich

Departmental Technical Reports (CS)

Among the most efficient characteristics of a probability distribution are its moments and, more generally, generalized moments. One of the most adequate numerical characteristics describing human behavior is expected utility. In both cases, the corresponding characteristic is the sum of results of applying appropriate nonlinear functions applied to individual inputs. In this paper, we provide a possible theoretical explanation of why such functions are efficient.

Notmad: Estimating Bayesian Networks With Sample-Specific Structures And Parameters, Benjamin Lengerich, Caleb Ellington, Bryon Aragam, Eric P. Xing, Manolis Kellis

Machine Learning Faculty Publications

Context-specific Bayesian networks (i.e. directed acyclic graphs, DAGs) identify context-dependent relationships between variables, but the non-convexity induced by the acyclicity requirement makes it difficult to share information between context-specific estimators (e.g. with graph generator functions). For this reason, existing methods for inferring context-specific Bayesian networks have favored breaking datasets into subsamples, limiting statistical power and resolution, and preventing the use of multidimensional and latent contexts. To overcome this challenge, we propose NOTEARS-optimized Mixtures of Archetypal DAGs (NOTMAD). NOTMAD models context-specific Bayesian networks as the output of a function which learns to mix archetypal networks according to sample context. The archetypal …

How Multi-View Techniques Can Help In Processing Uncertainty, Olga Kosheleva, Vladik Kreinovich

Departmental Technical Reports (CS)

Multi-view techniques help us reconstruct a 3-D object and its properties from its 2-D (or even 1-D) projections. It turns out that similar techniques can be used in processing uncertainty -- where many problems can reduced to a similar task of reconstructing properties of a multi-D object from its 1-D projections. In this chapter, we provide an overview of these techniques.

Why Do People Become Addicted: Towards A Theoretical Explanation For Eyal's Experiment-Based Hook Model, Christopher Reyes, Vladik Kreinovich

Departmental Technical Reports (CS)

Why do people become addicted, e.g., to gambling? Experiments have shown that simple lotteries, in which we can win a small prize with a certain probability, and not addictive. However, if we add a second possibility -- of having a large prize with a small probability -- the lottery becomes highly addictive to many participants. In this paper, we provide a possible theoretical explanation for this empirical phenomenon.

Why Ovals In Eliciting Intervals?, Joshua Zamora, Vladik Kreinovich

Departmental Technical Reports (CS)

To elicit people's opinions, we usually ask them to mark their degree of satisfaction on a scale -- e.g., from 0 to 5 or from 0 to 10. Often, people are unsure about the exact degree: 7 or 8? To cover such situations, it is desirable to elicit not a single value but an interval of possible values. However, it turns out that most people are not comfortable with marking an interval. Empirically, it turned out that the best way to elicit an interval is to ask them to draw an oval whose intersection with the 0-to-10 line is the …

Decision Making Under Uncertainty: Cases When We Only Know An Upper Bound Or A Lower Bound, Toshiki Kamio, Gavin Baechle, Vladik Kreinovich

Departmental Technical Reports (CS)

In situations when we have a perfect knowledge about the outcomes of several situations, a natural idea is to select the best of these situations. For example, among different investments, we should select the one with the largest gain. In practice, however, we rarely know the exact consequences of each action. In some cases, we know the lower and upper bounds on the corresponding gain. It has been proven that in such cases, an appropriate decision is to use Hurwicz optimism-pessimism criterion. In this paper, we extend the corresponding results to the cases when we only know an upper bound …

Commonsense "And"-Operations, Javier Tellez, Wenbo Xie, Vladik Kreinovich

Departmental Technical Reports (CS)

In many practical situations, we need to estimate our degree of belief in a statement "A and B" when the only thing we know are the degrees of belief a and b in combined statements A and B. An algorithm for this estimation is known as an "and"-operation, or, for historical reasons, a t-norm. Usually, "and"-operations are selected in such a way that if one of the statements A or B is false, our degree of belief in "A and B" is 0. However, in practice, this is sometimes not the case: for example, an ideal faculty candidate must satisfy …

Fourier Transform And Other Quadratic Problems Under Interval Uncertainty, Oscar Galindo, Christopher Ibarra, Vladik Kreinovich

Departmental Technical Reports (CS)

In general, computing the range of a quadratic function on given intervals is NP-hard. Recently, a feasible algorithm was proposed for computing the range of a specific quadratic function -- square of the modulus of a Fourier coefficient. For this function, the rank of the quadratic form -- i.e., the number of nonzero eigenvalues -- is 2. In this paper, we show that this algorithm can be extended to all the cases when the rank of the quadratic form is bounded by a constant.

Why Model Order Reduction, Salvador Robles, Martine Ceberio, Vladik Kreinovich

Departmental Technical Reports (CS)

Reasonably recently, a new efficient method appeared for solving complex non-linear differential equations (and systems of differential equations). In this method -- known as Model Order Reduction (MOR) -- we select several solutions, and approximate a general solution by a linear combination of the selected solutions. In this paper, we use the known explanation for efficiency of neural networks to explain the efficiency of MOR techniques.

Why Residual Neural Networks, Sofia Holguin, Vladik Kreinovich

Departmental Technical Reports (CS)

In the traditional neural networks, the outputs of each layer serve as inputs to the next layer. It is known that in many cases, it is beneficial to also allow outputs from pre-previous etc. layers as inputs. Such networks are known as residual. In this paper, we provide a possible theoretical explanation for the empirical success of residual neural networks.

How To Gauge The Quality Of A Multi-Class Classification When Ground Truth Is Known With Uncertainty, Ricardo Mendez, Osagumwenro Osaretin, Vladik Kreinovich

Departmental Technical Reports (CS)

The usual formulas for gauging the quality of a classification method assume that we know the ground truth, i.e., that for several objects, we know for sure to which class they belong. In practice, we often only know this with some degree of certainty. In this paper, we explain how to take this uncertainty into account when gauging the quality of a classification method.

Kinematic Metric Spaces Under Interval Uncertainty: Towards An Adequate Definition, Vladik Kreinovich, Olga Kosheleva, Victor Selivanov

Departmental Technical Reports (CS)

In the physical space, we define distance between the two points as the length of the shortest path connecting these points. Similarly, in space-time, for every pair of events for which the event a can causally effect the event b, we can define the longest proper time t(a,b) over all causal trajectories leading from a to b. The resulting function is known as kinematic metric. In practice, our information about all physical quantities -- including time -- comes from measurement, and measurements are never absolutely precise: the measurement result V is, in general, different from the actual (unknown) value v …

Fuzzy Logic Beyond Traditional "And"- And "Or"-Operations, Vladik Kreinovich, Olga Kosheleva

Departmental Technical Reports (CS)

In the traditional fuzzy logic, we can use "and"-operations (also known as t-norms) to estimate the expert's degree of confidence in a composite statement A&B based on his/her degrees of confidence d(A) and d(B) in the corresponding basic statements A and B. But what if we want to estimate the degree of confidence in A&B&C in situations when, in addition to the degrees of estimate d(A), d(B), and d(C) of the basic statements, we also know the expert's degrees of confidence in the pairs d(A&B), d(A&C), and d(B&C)? Traditional "and"-operations can provide such an estimate -- but only by ignoring …

The R Journal (December 2021) 13(2): Complete Issue, The R Foundation

The R Journal

On behalf of the R Foundation and the Editorial board, I am pleased to present Volume 13 Issue 2 of the R Journal. This is the biggest issue ever!

First, some news from the Editorial board. A big thank you to Mike Kane, who has finished his term. As Editor-in-Chief in 2020, Mike expanded operations to include Associate Editors in the reviewing process. The R Journal now has a team of 20 Associate Editors. This has helped to manage the increasing number of submissions. We welcome new Associate Editors, Przemek Biecek, Chris Brunsdon, Mine Çetinkaya-Rundel, Kieran Healy, Adam Loy, Priyanga …

Different Concepts, Similar Computational Complexity: Nguyen's Results About Fuzzy And Interval Computations 35 Years Later, Hung T. Nguyen, Vladik Kreinovich

Departmental Technical Reports (CS)

When we know for sure which values are possible and which are not, we have crisp uncertainty -- of which interval uncertainty is a usual case. In practice, we are often not 100% sure about our knowledge, i.e., we have fuzzy uncertainty -- i.e., we have fuzzy knowledge, of which crisp is a particular case. Usually, general problems are more difficult to solve that most of their particular cases. It was therefore expected that processing fuzzy data is, in general, more computationally difficult than processing interval data -- and indeed, Zadeh's extension principle -- a natural formula for fuzzy computations …

Comparing The Popularity Of Testing Careers Among Canadian, Indian, Chinese, And Malaysian Students, Luiz Fernando Capretz, Pradeep Waychal, Jingdong Jia, Shuib Basri

Electrical and Computer Engineering Publications

This study attempts to understand motivators and de-motivators that influence the decisions of software students to take up and sustain software testing careers across four different countries, Canada, India, China, and Malaysia. Towards that end, we have developed a cross-sectional, but simple, survey-based instrument. In this study we investigated how software engineering and computer science students perceive and value what they do and their environmental settings. This study found that very few students are keen to take up software testing careers - why is this happening with such an important task in the software life cycle? The common advantages of …

Fault Detection In A Smart Electric Grid: Geometric Analysis, Hector Reyes, Dillon Trinh, Vladik Kreinovich

Departmental Technical Reports (CS)

The main idea behind a smart grid is to equip the grid with a dense lattice of sensors monitoring the state of the grid. If there is a fault, the sensors closer to the fault will detect larger deviations from the normal readings that sensors that are farther away. In this paper, we show that this fact can be used to locate the fault with high accuracy.

Why Geological Regions?, Daniela Flores, Olga Kosheleva, Vladik Kreinovich

Departmental Technical Reports (CS)

In most practical applications, we approximate the spatial dependence by smooth functions. The main exception is geosciences, where, to describe, e.g., how the density depends on depth and/or on spatial location, geophysicists divide the area into regions on each of which the corresponding quantity is approximately constant. In this paper, we provide a possible explanation for this difference.

Why People Overestimate Small Probabilities?, David Amparan, Vladik Kreinovich

Departmental Technical Reports (CS)

It is a known empirical fact that people overestimate small probabilities. This fact seems to be inconsistent with the fact that we humans are the product of billions years of improving evolution -- and that we therefore perceive the world as accurately as possible. In this paper, we provide a possible explanation for this seeming contradiction.

Why Rectified Linear Neurons: A Possible Interval-Based Explanation, Jonathan Contreras, Martine Ceberio, Vladik Kreinovich

Departmental Technical Reports (CS)

At present, the most efficient machine learning techniques are deep neural networks. In these networks, a signal repeatedly undergoes two types of transformations: linear combination of inputs, and a non-linear transformation of each value v -> s(v). Empirically, the function s(v) = max(v,0) -- known as the rectified linear function -- works the best. There are some partial explanations for this empirical success; however, none of these explanations is fully convincing. In this paper, we analyze this why-question from the viewpoint of uncertainty propagation. We show that reasonable uncertainty-related arguments lead to another possible explanation of why rectified linear functions …

How Probabilistic Methods For Data Fitting Deal With Interval Uncertainty: A More Realistic Analysis, Vladik Kreinovich, Sergey P. Shary

Departmental Technical Reports (CS)

In our previous paper, we showed that a simplified probabilistic approach to interval uncertainty leads to the known notion of a united solution set. In this paper, we show that a more realistic probabilistic analysis of data fitting under interval uncertainty leads to another known notion -- the notion of a tolerable solution set. Thus, the notion of a tolerance solution set also has a clear probabilistic interpretation. Good news is that, in contrast to the united solution set whose computation is, in general, NP-hard, the tolerable solution set can be computed by a feasible algorithm.

Developing An International Framework For Addressing Non-State Actors In Cyberspace, Joanna C. Di Scipio

Cybersecurity Undergraduate Research Showcase

On May 7, 2021, Colonial Pipeline shut down its operations following a ransomware attack by the criminal group DarkSide (Bordoff, 2021). It took five days to resume normal operations, but this short period led to panic buying, rising prices, and significant gas shortages. The attack underscores an emerging threat in the landscape of cybersecurity: critical infrastructure attacks carried out by non-state actors.

Self-Supervised Perceptual Ad-Blocker, Killian Robinson

Cybersecurity Undergraduate Research Showcase

This project proposes a new self-supervised ad-blocker to minimize the amount of human effort required to effectively combat pushed advertisements. Current ad-blocker models are expensive to develop and not always effective in identifying advertisements. We investigated the possibility of solving these problems with the introduction of a deep learning, self-supervised ad-blocker model. More specifically, the proposed ad-blocker will be trained in a self-supervised fashion to tackle the issue of lacking labelled training data. The proposed solution was prototyped using Pytorch and achieved a detection accuracy of 81% on a diverse selection of popular websites.

Protection Of Patient Privacy On Mobile Device Machine Learning, Matthew Nguyen

Cybersecurity Undergraduate Research Showcase

An existing StudentLife Study mobile dataset was evaluated and organized to be applied to different machine learning methods. Different variables like user activity, exercise, sleep, study space, social, and stress levels are optimized to train a model that could predict user stress level. The different machine learning methods would test if both patient data privacy and training efficiency can be ensured.

Gdpr, Pipl & Lgpd: Privacy Regulations & Policies Across The Globe, Raymond H. Geistel

Cybersecurity Undergraduate Research Showcase

Several privacy laws around the world are adopting similar regulations to the GPDR; this has effects on privacy policies of companies providing services in across multiple countries & continents. While these regulations share many attributes, their differing requirements can make things difficult for companies regarding said policies. Automation could be a potential solution to both analyze and compare regulations from different nations & international organizations, analyze and monitor privacy policy adherence to said regulations.

How Secure Are Android And Apple’S Operating Systems And Based Applications Against Cyber Attacks And Cyber Crime, Marlowe Cosby Jr.

Cybersecurity Undergraduate Research Showcase

Smartphone has become an important part of our everyday life. Android and apple are the two most used operating system (OS) for smart phones. We usually store important information in our smart phone, e.g.: credit card, bank account, driving ID, SSN. As a result, Android and Apple operating systems and applications have both been subject to a wide number of vulnerabilities and attacks. This directly effects many people being that they are the global leaders of users within their platforms reaching billions of people daily. It is important that smartphones receive better defense and security. In this paper, we aim …