Search

The recent progress of deep learning techniques has produced models capable of achieving high scores on traditional Natural Language Inference (NLI) datasets. To understand the generalization limits of these powerful models, an increasing number of adversarial evaluation schemes have appeared. These works use a similar evaluation method: they construct a new NLI test set based on sentences with known logic and semantic properties (the adversarial set), train a model on a benchmark NLI dataset, and evaluate it in the new set. Poor performance on the adversarial set is identified as a model limitation. The problem with this evaluation procedure is that it may only indicate a sampling problem. A machine learning model can perform poorly on a new test set because the text patterns presented in the adversarial set are not well represented in the training sample. To address this problem, we present a new evaluation method, the Invariance under Equivalence test (IE test). The IE test trains a model with sufficient adversarial examples and checks the model's performance on two equivalent datasets. As a case study, we apply the IE test to the state-of-the-art NLI models using synonym substitution as the form of adversarial examples. The experiment shows that, despite their high predictive power, these models usually produce different inference outputs for equivalent inputs, and, more importantly, this deficiency cannot be solved by adding adversarial observations in the training data. [ABSTRACT FROM AUTHOR]

Neural networks are widely used in systems of artificial intelligence, but due to their black box nature, they have so far evaded formal analysis to certify that they satisfy desirable properties, mainly when they perform critical tasks. In this work, we introduce methods for the formal analysis of reachability and robustness of neural networks that are modeled as rational McNaughton functions by, first, stating such properties in the language of Łukasiewicz infinitely-valued logic and, then, using the reasoning techniques of such logical system. We also present a case study where we employ the proposed techniques in an actual neural network that we trained to predict whether it will rain tomorrow in Australia. [ABSTRACT FROM AUTHOR]

This work studies quantitative measures for ranking judicial decisions by the Brazilian Supreme Court [Supremo Tribunal Federal (STF)] and selecting leading cases, which are understood as those with broadness of influence on different legal fields. The measures are based on a network built over decisions whose cases were finalized in the Brazilian Supreme Court between 01/2001 and 12/2019, which were obtained by crawling publicly available STF records. Three ranking measures are proposed; two are adaptations of the PageRank algorithm, and one adapts Kleinberg's algorithm. Such measures are compared with respect to agreement on top 100 rankings; we also analyze each robustness measure based on self-agreement under perturbation. We examine whether the resulting quantitative ranking is congenial to a qualitative intuition of what the legal community usually considers as relevant precedents. We also discuss some possible criteria of relevance in the seek for patterns that suggest how quantitative and qualitative measures would better align. The ranking of leading cases and relevant decisions improved after building decision networks without irrelevant appeals and decisions that overflow the court offers a starting point to discuss the role of STF in the Brazilian judicial system. In our last work, both versions of PageRank and Kleinberg algorithms produced different rankings and all of them were robust with respect to 10% and 20%-perturbation levels, but none of them retrieved leading cases at the top of these rankings. Then, we took a further step in the studies of the STF decision network and we introduced better filtering of network nodes guided by legal expertise on the works of the Supreme Court. We also introduced more fine-grained perturbance levels to understand the impact of such filters in the STF decision network. We concluded that after filtering low-relevance decision types, the STF decision network is still robust under 10%-perturbation, but it presents higher degradation by increasing perturbation levels. The two versions of PageRank still produce different rankings. Kleinberg's algorithm provides a different ranking, with many relevant criminal cases. Although we improved algorithms rankings filtering decisions from the network, which represents an important methodological step, there is still room for improvement. Given that relevant decisions are well ranked after filtering out a large amount of irrelevant decisions, the results set a starting point to discuss the role of STF in the Brazilian judicial system. [ABSTRACT FROM AUTHOR]

This study compares fundamental frequency (fo) and fundamental frequency standard deviation (foSD) of COVID-19 patients with the same parameters in the speech of subjects without COVID-19, and verifies whether there is an effect of age and sex in the patient group. Both groups, subjects with and without COVID-19, are formed by Brazilian Portuguese speakers. Speech samples were obtained from 100 patients with mild to severe symptoms of COVID-19, and 100 healthy subjects. A single 31-syllable Portuguese sentence was used as the elicitation material for all subjects. The recordings were divided into four age groups. The acoustic measures were semi-automatically extracted and analyzed by a series of analyses of variance. Patients with COVID-19 present vocal differences in fo-related parameters when compared to healthy subjects, that is, patient voices presented higher fo and foSD with respect to control voices. In addition, for patient voices, there was an age and sex effect on fo SD values. Vocal parameters of women and elderly subjects showed more marked differences in fo-related parameters, indicating that patient voices are higher-pitched and have a higher variation of fo SD. Consequently, fo-related parameters may be tested as vocal biomarkers in the screening of respiratory insufficiency by voice analysis, in patients with severe symptoms of COVID-19. [ABSTRACT FROM AUTHOR]

This work concerns the representation of a class of continuous functions into Logic, so that one may automatically reason about properties of these functions using logical tools. Rational McNaughton functions may be implicitly represented by logical formulas in Łukasiewicz Infinitely-valued Logic by constraining the set of allowed valuations; such a restriction contemplates only those valuations that satisfy specific formulas. This work investigates two approaches to such depiction, called representation modulo satisfiability. Furthermore, a polynomial-time algorithm that builds this representation is presented, producing a pair of formulas consisting of the representative formula and the constraining one, given as input a rational McNaughton function in a suitable encoding. An implementation of the algorithm is discussed. [ABSTRACT FROM AUTHOR]

In this work, we first deal with the modeling of game situations that reach one of possibly many Nash equilibria. Before an instance of such a game starts, an external observer does not know, a priori, what is the exact profile of actions -- constituting an equilibrium -- that will occur; thus, he assigns subjective probabilities to players' actions. Such scenario is formalized as an observable game, which is a newly introduced structure for that purpose. Then, we study the decision problem of determining if a given set of probabilistic constraints assigned a priori by the observer to a given game is coherent, called the PCE-COHERENCE problem. We show several results concerning algorithms and complexity for PCE-COHERENCE when pure Nash equilibria and specific classes of games, called GNP-classes, are considered. In this context, we also study the computation of maximal and minimal probabilistic constraints on actions that preserves coherence. Finally, we study these problems when mixed Nash equilibria are allowed in GNP-classes of 2-player games. [ABSTRACT FROM AUTHOR]

Pathology reports are a main source of information regarding cancer diagnosis and are commonly written following semi-structured templates that include tumour localisation and behaviour. In this work, we evaluated the efficiency of support vector machines (SVMs) to classify pathology reports written in Portuguese into the International Classification of Diseases for Oncology (ICD-O), a biaxial classification of cancer topography and morphology. A partnership program with the Brazilian hospital A.C. Camargo Cancer Center provided anonymised pathology reports and structured data from 94,980 patients used for training and validation. We employed SVMs with tf-idf weighting scheme in a bagof-words approach and report F1 score of 0.82 for 18 sites and 0.73 for 49 morphology classes. With the largest dataset ever used in such a task, our work provides reliable estimates for the classification of pathology reports in Portuguese and agrees with a few similar studies published in the same kind of data in other languages. [ABSTRACT FROM AUTHOR]

This paper examines two aspects of propositional probabilistic logics: the nesting of probabilistic operators, and the expressivity of probabilistic assessments. We show that nesting can be eliminated when the semantics is based on a single probability measure over valuations; we then introduce a classification for probabilistic assessments, and present novel results on their expressivity. Logics in the literature are categorized using our results on nesting and on probabilistic expressivity. [ABSTRACT FROM AUTHOR]

In traditional abduction methods it is assumed that the goal data has not yet been explained. In first-order logic (FOL) such precondition is, in general, undecidable. To avoid this problem, it was presented in D'Agostino, Finger and Gabbay (2008, Logic J. of the IGPL, 16, 537–560) a first-order cut-based abduction method which has as underlying inference system the KE-tableaux. Such inference system is non-analytical and the method presented in D'Agostino, Finger and Gabbay (2008, Logic J. of the IGPL, 16, 537–560) generalizes the traditional abduction avoiding the undecidable precondition; its correctness was demonstrated. Here we will show the completeness of the method for both propositional and first-order logic. [ABSTRACT FROM PUBLISHER]

In this work we provide a statistical form of empirical analysis of classical propositional logic decision methods called SAT solvers. This work is perceived as an empirical counterpart of a theoretical movement, called the enduring scandal of deduction, that opposes considering Boolean Logic as trivial in any sense. For that, we study the predictability of classical logic, which we take to be the distribution of the runtime of its decision process. We present a series of experiments that determines the run distribution of SAT solvers and discover a varying landscape of distributions, following the known existence of a transition of easy-hard-easy cases of propositional formulas. We find clear distributions for the easy areas and the transitions easy-hard and hard-easy. The hard cases are shown to be hard also for the detection of statistical distributions, indicating that several independent processes may be at play in those cases. [ABSTRACT FROM AUTHOR]

Traditional abduction imposes as a precondition the restriction that the background information may not derive the goal data. In first-order logic such precondition is, in general, undecidable. To avoid such problem, we present a first-order cut-based abduction method, which has KE-tableaux as its underlying inference system. This inference system allows for the automation of non-analytic proofs in a tableau setting, which permits a generalization of traditional abduction that avoids the undecidable precondition problem. After demonstrating the correctness of the method, we show how this method can be dynamically iterated in a process that leads to the construction of non-analytic first-order proofs and, in some terminating cases, to refutations as well. [ABSTRACT FROM PUBLISHER]

This work presents a method for predicting resource availability in opportunistic grids by means of use pattern analysis (UPA), a technique based on non-supervised learning methods. This prediction method is based on the assumption of the existence of several classes of computational resource use patterns, which can be used to predict the resource availability. Trace-driven simulations validate this basic assumptions, which also provide the parameter settings for the accurate learning of resource use patterns. Experiments made with an implementation of the UPA method show the feasibility of its use in the scheduling of grid tasks with very little overhead. The experiments also demonstrate the method's superiority over other predictive and non-predictive methods. An adaptative prediction method is suggested to deal with the lack of training data at initialization. Further adaptative behaviour is motivated by experiments which show that, in some special environments, reliable resource use patterns may not always be detected. Copyright © 2009 John Wiley & Sons, Ltd. [ABSTRACT FROM AUTHOR]

The logic of proofs (lp) was proposed as Gödel's missed link between Intuitionistic and S4-proofs, but so far the tableau-based methods proposed for lp have not explored this closeness with S4 and contain rules whose analycity is not immediately evident. We study possible formulations of analytic tableau proof methods for lp that preserve the subformula property. Two sound and complete tableau decision methods of increasing degree of analycity are proposed, KELP and preKELP. The latter is particularly inspired on S4-proofs. The crucial role of proof constants in the structure of lp-proofs methods is analysed. In particular, a method for the abduction of proof constant specifications in strongly analytic preKELP proofs is presented; abduction heuristics and the complexity of the method are discussed. [ABSTRACT FROM PUBLISHER]

In this paper we explore a generalization of traditional abduction which can simultaneously perform two different tasks: (i) given an unprovable sequent Γ ⊢ G, find a sentence H such that Γ, H ⊢ G is provable (hypothesis generation); (ii) given a provable sequent Γ ⊢ G, find a sentence H such that Γ ⊢ H and the proof of Γ, H ⊢ G is simpler than the proof of Γ ⊢ G (lemma generation). We argue that the two tasks should not be distinguished, and present a general procedure for finding suitable hypotheses or lemmas. When the original sequent is provable, the abduced formula can be seen as a cut formula with respect to Gentzen's sequent calculus, so the abduction method is cut-based. Our method is based on the tableau-like system KE and we argue for its advantages over existing abduction methods based on traditional Smullyan-style Tableaux. [ABSTRACT FROM PUBLISHER]

This work studies the structure of proofs containing non-analytic cuts in the cut-based system, a sequent inference system in which the cut rule is not eliminable and the only branching rule is the cut. Such sequent system is invertible, leading to the KE-tableau decision method. We study the structure of such proofs, proving the existence of a normal form for them in the form of a comb-tree proof. [ABSTRACT FROM PUBLISHER]

In this article we present s1, a family of logics that is useful to disprove propositional formulas by means of an anytime approximation process. The systems follows the paradigm of a parameterized family of logics established by Schaerf''s and Cadoli's system S1. We show that s1 inherits several of the nice properties of S1, while presenting several attractive new properties. The family s1 deals with the full propositional language, has a complete tableau proof system which provides for incremental approximations; furthermore, it constitutes a full approximation of classical logic from above, with an approximation process with better relevance and locality properties than S1 When applied to clausal inference, s1 provides a strong simplification method. An application of s1 to model-based diagnosis is presented, demonstrating how the solution to this problem can benefit from the use of S1 approximations. [ABSTRACT FROM AUTHOR]