Searcharxiv⌕ Search

arXiv subjects

Miguel Couceiro

Publications and source records attributed to Miguel Couceiro.

At least 55 records · Page 3Linked to original sources
arXiv2021

A Bayesian Neural Network based on Dropout Regulation

Bayesian Neural Networks (BNN) have recently emerged in the Deep Learning world for dealing with uncertainty estimation in classification tasks, and are used in many application domains such as astrophysics, autonomous driving...BNN assume a prior over the weights of a neural network instead of point estimates, enabling in this way the estimation of both aleatoric and epistemic uncertainty of the model prediction.Moreover, a particular type of BNN, namely MC Dropout, assumes a Bernoulli distribution on the weights by using Dropout.Several attempts to optimize the dropout rate exist, e.g. using a variational approach.In this paper, we present a new method called "Dropout Regulation" (DR), which consists of automatically adjusting the dropout rate during training using a controller as used in automation.DR allows for a precise estimation of the uncertainty which is comparable to the state-of-the-art while remaining simple to implement.

cs.LG↗
arXiv2020

Making ML models fairer through explanations: the case of LimeOut

Algorithmic decisions are now being used on a daily basis, and based on Machine Learning (ML) processes that may be complex and biased. This raises several concerns given the critical impact that biased decisions may have on individuals or on society as a whole. Not only unfair outcomes affect human rights, they also undermine public trust in ML and AI. In this paper we address fairness issues of ML models based on decision outcomes, and we show how the simple idea of "feature dropout" followed by an "ensemble approach" can improve model fairness. To illustrate, we will revisit the case of "LimeOut" that was proposed to tackle "process fairness", which measures a model's reliance on sensitive or discriminatory features. Given a classifier, a dataset and a set of sensitive features, LimeOut first assesses whether the classifier is fair by checking its reliance on sensitive features using "Lime explanations". If deemed unfair, LimeOut then applies feature dropout to obtain a pool of classifiers. These are then combined into an ensemble classifier that was empirically shown to be less dependent on sensitive features without compromising the classifier's accuracy. We present different experiments on multiple datasets and several state of the art classifiers, which show that LimeOut's classifiers improve (or at least maintain) not only process fairness but also other fairness metrics such as individual and group fairness, equal opportunity, and demographic parity.

cs.LG↗
arXiv2020

Tackling scalability issues in mining path patterns from knowledge graphs: a preliminary study

Features mined from knowledge graphs are widely used within multiple knowledge discovery tasks such as classification or fact-checking. Here, we consider a given set of vertices, called seed vertices, and focus on mining their associated neighboring vertices, paths, and, more generally, path patterns that involve classes of ontologies linked with knowledge graphs. Due to the combinatorial nature and the increasing size of real-world knowledge graphs, the task of mining these patterns immediately entails scalability issues. In this paper, we address these issues by proposing a pattern mining approach that relies on a set of constraints (e.g., support or degree thresholds) and the monotonicity property. As our motivation comes from the mining of real-world knowledge graphs, we illustrate our approach with PGxLOD, a biomedical knowledge graph.

cs.DB↗
arXiv2020

LimeOut: An Ensemble Approach To Improve Process Fairness

Artificial Intelligence and Machine Learning are becoming increasingly present in several aspects of human life, especially, those dealing with decision making. Many of these algorithmic decisions are taken without human supervision and through decision making processes that are not transparent. This raises concerns regarding the potential bias of these processes towards certain groups of society, which may entail unfair results and, possibly, violations of human rights. Dealing with such biased models is one of the major concerns to maintain the public trust. In this paper, we address the question of process or procedural fairness. More precisely, we consider the problem of making classifiers fairer by reducing their dependence on sensitive features while increasing (or, at least, maintaining) their accuracy. To achieve both, we draw inspiration from "dropout" techniques in neural based approaches, and propose a framework that relies on "feature drop-out" to tackle process fairness. We make use of "LIME Explanations" to assess a classifier's fairness and to determine the sensitive features to remove. This produces a pool of classifiers (through feature dropout) whose ensemble is shown empirically to be less dependent on sensitive features, and with improved or no impact on accuracy.

cs.LG↗
arXiv2020

Knowledge-Based Matching of $n$-ary Tuples

An increasing number of data and knowledge sources are accessible by human and software agents in the expanding Semantic Web. Sources may differ in granularity or completeness, and thus be complementary. Consequently, they should be reconciled in order to unlock the full potential of their conjoint knowledge. In particular, units should be matched within and across sources, and their level of relatedness should be classified into equivalent, more specific, or similar. This task is challenging since knowledge units can be heterogeneously represented in sources (e.g., in terms of vocabularies). In this paper, we focus on matching n-ary tuples in a knowledge base with a rule-based methodology. To alleviate heterogeneity issues, we rely on domain knowledge expressed by ontologies. We tested our method on the biomedical domain of pharmacogenomics by searching alignments among 50,435 n-ary tuples from four different real-world sources. Results highlight noteworthy agreements and particularities within and across sources.

cs.AI↗
arXiv2019

Every quasitrivial n-ary semigroup is reducible to a semigroup

We show that every quasitrivial n-ary semigroup is reducible to a binary semigroup, and we provide necessary and sufficient conditions for such a reduction to be unique. These results are then refined in the case of symmetric n-ary semigroups. We also explicitly determine the sizes of these classes when the semigroups are defined on finite sets. As a byproduct of these enumerations, we obtain several new integer sequences.

math.RA↗
arXiv2019

Quasitrivial semigroups: characterizations and enumerations

We investigate the class of quasitrivial semigroups and provide various characterizations of the subclass of quasitrivial and commutative semigroups as well as the subclass of quasitrivial and order-preserving semigroups. We also determine explicitly the sizes of these classes when the semigroups are defined on finite sets. As a byproduct of these enumerations, we obtain several new integer sequences.

math.RA↗
arXiv2017

Characterizations of idempotent discrete uninorms

In this paper we provide an axiomatic characterization of the idempotent discrete uninorms by means of three conditions only: conservativeness, symmetry, and nondecreasing monotonicity. We also provide an alternative characterization involving the bisymmetry property. Finally, we provide a graphical characterization of these operations in terms of their contour plots, and we mention a few open questions for further research.

math.RA↗
arXiv2017

Fast full-body reconstruction for a functional human RPC-PET imaging system using list-mode simulated data and its applicability to radiation oncology and radiology

Single-bed whole-body positron emission tomography based on resistive plate chamber detectors (RPC-PET) has been proposed for human studies, as a complementary resource to scintillator-based PET scanners. The purpose of this work is mainly about providing a reconstruction solution to such whole-body single-bed data collection on an event-by-event basis. We demonstrate a fully three-dimensional time-of-flight (TOF)-based reconstruction algorithm that is capable of processing the highly inclined lines of response acquired from a system with a very large axial field of view, such as those used in RPC-PET. Such algorithm must be sufficiently fast that it will not compromise the clinical workflow of an RPC-PET system. We present simulation results from a voxelized version of the anthropomorphic NCAT phantom, with oncological lesions introduced into critical regions within the human body. The list-mode data was reconstructed with a TOF-weighted maximum-likelihood expectation maximization (MLEM). To accelerate the reconstruction time of the algorithm, a multi-threaded approach supported by graphical processing units (GPUs) was developed. Additionally, a TOF-assisted data division method is suggested that allows the data from nine body regions to be reconstructed independently and much more rapidly. The application of a TOF-based scatter rejection method reduces the overall body scatter from 57.1% to 32.9%. The results also show that a 300-ps FWHM RPC-PET scanner allows for the production of a reconstructed image in 3.5 minutes following a 7-minute acquisition upon the injection of 2 mCi of activity (146 M coincidence events). We present for the first time a full realistic reconstruction of a whole body, long axial coverage, RPC-PET scanner. We demonstrate clinically relevant reconstruction times comparable (or lower) to the patient acquisition times on both multi-threaded CPU and GPU.

physics.med-ph↗
arXiv2017

Pivotal decomposition schemes inducing clones of operations

We study pivotal decomposition schemes and investigate classes of pivotally decomposable operations. We provide sufficient conditions on pivotal operations that guarantee that the corresponding classes of pivotally decomposable operations are clones, and show that under certain assumptions these conditions are also necessary. In the latter case, the pivotal operation together with the constant operations generate the corresponding clone.

math.RA↗
arXiv2016

Conservative median algebras and semilattices

We characterize conservative median algebras and semilattices by means of forbidden substructures and by providing their representation as chains. Moreover, using a dual equivalence between median algebras and certain topological structures, we obtain descriptions of the median-preserving mappings between products of finitely many chains.

math.RA↗
arXiv2015

Arrow type impossibility theorems over median algebras

We characterize trees as median algebras and semilattices by relaxing conservativeness. Moreover, we describe median homomorphisms between products of median algebras and show that Arrow type impossibility theorems for mappings from a product $\mathbf{A}_1\times \cdots \times \mathbf{A}_n$ of median algebras to a median algebra $\mathbf{B}$ are possible if and only if $\mathbf{B}$ is a tree, when thought of as an ordered structure.

cs.DM↗
arXiv2015

Relaxations of associativity and preassociativity for variadic functions

In this paper we consider two properties of variadic functions, namely associativity and preassociativity, that are pertaining to several data and language processing tasks. We propose parameterized relaxations of these properties and provide their descriptions in terms of factorization results. We also give an example where these parameterized notions give rise to natural hierarchies of functions and indicate their potential use in measuring the degrees of associativeness and preassociativeness. We illustrate these results by several examples and constructions and discuss some open problems that lead to further directions of research.

math.GR↗
arXiv2015

Galois Connections for Generalized Functions and Relational Constraints

In this paper we focus on functions of the form $A^n\rightarrow \mathcal{P}(B)$, for possibly different arbitrary non-empty sets $A$ and $B$, and where $\mathcal{P}(B)$ denotes the set of all subsets of $B$. These mappings are called \emph{multivalued functions}, and they generalize total and partial functions. We study Galois connections between these generalized functions and ordered pairs $(R,S)$ of relations on $A$ and $B$, respectively, called \emph{constraints}. We describe the Galois closed sets, and decompose the associated Galois operators, by means of necessary and sufficient conditions which specialize, in the total single-valued case, to those given in the author's previous work [M. Couceiro, S. Foldes. On closed sets of relational constraints and classes of functions closed under variable substitutions, Algebra Universalis 54 (2005) 149-165].

math.RA↗
arXiv2015

On Galois Connections between External Operations and Relational Constraints: Arity Restrictions and Operator Decompositions

We study the basic Galois connection induced by the "satisfaction" relation between external operations $A^n\rightarrow B$ defined on a set $A$ and valued in a possibly different set $B$ on the one hand, and ordered pairs $(R,S)$ of relations $R\subseteq A^m$ and $S\subseteq B^m$, called relational constraints, on the other hand. We decompose the closure maps associated with this Galois connection, in terms of closure operators corresponding to simple closure conditions describing the corresponding Galois closed sets of functions and constraints. We consider further Galois correspondences by restricting the sets of primal and dual objects to fixed arities. We describe the restricted Galois closure systems by means of parametrized analogues of the simpler closure conditions, and present factorizations of the corresponding Galois closure maps, similar to those provided in the unrestricted case.

math.RA↗
arXiv2015

Partial clones containing all Boolean monotone self-dual partial functions

The study of partial clones on $\mathbf{2}:=\{0,1\}$ was initiated by R. V. Freivald. In his fundamental paper published in 1966, Freivald showed, among other things, that the set of all monotone partial functions and the set of all self-dual partial functions are both maximal partial clones on $\mathbf{2}$. Several papers dealing with intersections of maximal partial clones on $\mathbf{2}$ have appeared after Freivald work. It is known that there are infinitely many partial clones that contain the set of all monotone self-dual partial functions on $\mathbf{2}$, and the problem of describing them all was posed by some authors. In this paper we show that the set of partial clones that contain all monotone self-dual partial functions is of continuum cardinality on $\mathbf{2}$.

math.CO↗
arXiv2015

Hereditarily rigid relations

An $h$-ary relation $\r$ on a finite set $A$ is said to be \emph{hereditarily rigid} if the unary partial functions on $A$ that preserve $\r$ are the subfunctions of the identity map or of constant maps. A family of relations ${\mathcal F}$ is said to be \emph{hereditarily strongly rigid} if the partial functions on $A$ that preserve every $\r \in {\mathcal F}$ are the subfunctions of projections or constant functions. In this paper we show that hereditarily rigid relations exist and we give a lower bound on their arities. We also prove that no finite hereditarily strongly rigid families of relations exist and we also construct an infinite hereditarily strongly rigid family of relations.

math.CO↗