1,721,071 research outputs found
Improving the prediction performance of Support Vector Machines
No full textIl miglioramento della capacità predittiva dei metodi di apprendimento è una delle tematiche più rilevanti della Letteratura di Machine Learning e di Artificial intelligence degli ultimi anni. La continua ricerca di tecniche innovative in grado di accrescere la capacità dei modelli ha spinto i ricercatori a proporre nuovi metodi e ad indagare i fattori che ne influenzano le prestazioni. La presente ricerca si indirizza verso la seconda tematica ed in particolare si focalizza sugli attributi che descrivono i dati, sui parametri dei modelli, sui dataset incompleti e sulle relazioni che sussistono tra gli attributi e i parametri degli stessi. Questa ricerca si pone l’obiettivo di migliorare la capacità predittiva dei modelli di apprendimento mediante la selezione di attributi e parametri e la riduzione del contributo delle osservazioni con valori mancanti durante la fase di apprendimento del modello. Il primo tema è sviluppato proponendo due nuovi metodi: il Kernel Matrix Genetic Algorithm (KMGA), studiato per dataset di piccole-medie dimensioni, e il Reduction by Differences (RbD) indicato per dataset di grandi dimensioni. Entrambi i metodi sono basati su algoritmi evolutivi ed eseguono una ricerca simultanea degli attributi del dataset e dei parametri di un modello. Il KMGA esegue la ricerca utilizzando un indicatore approssimato di bontà della matrice kernel in grado di migliorare la predizione di un classificatore di tipo Support Vector Machines e di ridurre lo sforzo computazionale dell’algoritmo evolutivo su cui è basato. Il RbD ha invece carattere più generale e può essere applicato a qualsiasi metodo di apprendimento. Il RbD è sviluppato per dataset di grandi dimensioni e, rispetto ai metodi evolutivi proposti in Letteratura, permette di eseguire la ricerca simultanea di attributi e parametri senza richiedere una riduzione preliminare di migliaia di attributi.
Il secondo tema è sviluppato proponendo il metodo missVal, che è studiato per dataset incompleti in cui i valori mancanti delle osservazioni sono sostituiti da valori ad hoc. Lo scopo è quello di ridurre il contributo di quelle osservazioni durante la fase di apprendimento per migliorare la capacità predittiva.The improvement of prediction performance of Learning methods is one of the leading topics of recent Machine Learning and Artificial Intelligence Literature. The search for innovative techniques has prompted the researchers to explore new areas and to investigate the factors that influence the prediction performance. This research addresses the second issue focusing the attention on features, method parameters, incomplete datasets and relations among features and parameters. The research target is to improve the prediction ability of the Learning methods through the search for the best subset of features and the optimal method parameters and to reduce the contribution of missing values. The former theme has been developed proposing two new methods: the Kernel Matrix Genetic Algorithm (KMGA) and the Reduction by Differences (RbD). Both methods are based on Evolutionary Algorithms and perform a simultaneous search for the features and the parameters. The KMGA is suggested for small-medium size datasets and improves the prediction performance of Support Vector Machines by means of approximated kernel matrix measures. Moreover, it allows reducing the effort of the evolutionary search. The RbD is designed for large datasets that suffer from the Curse of Dimensionality and can be used with every Learning methods. Unlike the evolutionary methods proposed in Literature, the RbD does not require a preliminary reduction of thousands of features. The latter theme has been developed for incomplete datasets proposing the missVal method. The target of the missVal is to improve the prediction of Learning methods reducing the contribution of replaced missing values in the learning phase.DIPARTIMENTO DI INGEGNERIA GESTIONALE23VERCELLIS, CARLOCOLOMBO, MASSIMO GAETAN
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Full text link“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Appropriate Similarity Measures for Author Cocitation Analysis
Full text linkWe provide a number of new insights into the methodological discussion about author cocitation analysis. We first argue that the use of the Pearson correlation for measuring the similarity between authors’ cocitation profiles is not very satisfactory. We then discuss what kind of similarity measures may be used as an alternative to the Pearson correlation. We consider three similarity measures in particular. One is the well-known cosine. The other two similarity measures have not been used before in the bibliometric literature. Finally, we show by means of an example that our findings have a high practical relevance.information science;Pearson correlation;cosine;similarity measure;author cocitation analysis
Learning and Optimization of the Locomotion with an Exoskeleton for Paraplegic People
No full textCette thèse contribue à améliorer la planification de trajectoires et le contrôle des robots bipèdes. Le but concret est de permettre aux paraplégiques de remarcher de façon autonome avec l’exosquelette de membres inférieurs Atalante. Notre approche combine les méthodes issues l’apprentissage automatique et de la robotique traditionnelle. Nous mettons d’abord de côté le contrôle. L’objectif est de permettre la planification de trajectoires en ligne tout en garantissant un fonctionnement sûr. C’est une étape cruciale vers la navigation en milieu incertain et la prise en compte des préférences utilisateur. Nous entraînons ensuite un contrôleur par renforcement afin de généraliser un ensemble prédéfini de mouvements élémentaires. Nous ne cherchons pas la meilleure performance, mais plutôt la transférabilité et la sécurité. Nous proposons une formulation qui apparente à l’apprentissage par imitation mais laisse suffisamment de marge de manœuvre pour affronter des événements inattendus.This thesis contributes to improving the motion planning and control of biped robots. Our concrete goal is restoring natural locomotion for paraplegic people in their daily lives using the medical lower-limb exoskeleton Atalante, notably walkingsafely and autonomously without crutches. The core idea is to combine traditional robotics and state-of-the-art machine learning. We put aside closed-loop control to focus on planning at first. The objective is to enable online trajectory planning while ensuring safe operation. This is a milestone toward realizing versatile navigation in an unstructured environment and accommodating the user preferences. Second, we train a policy using reinforcement learning to generalize a predefined set of primitive motions. We do not seek the best possible performance, but rather transferability and safety. We propose a formulation closely related to imitation learning while giving enough leeway to deal with unexpected events
Dispelling the Myths Behind First-author Citation Counts
Full text linkWe conducted a full-scale evaluative citation analysis study of scholars in the XML research field to explore just how different from each other author rankings resulting from different citation counting methods actually are, and to demonstrate the capability of emerging data and tools on the Web in supporting more realistic citation counting methods. Our results contest some common arguments for the continued
use of first-author citation counts in the evaluation of scholars, such as high correlations between author rankings by first-author citation counts and other citation
counting methods, and high costs of using more realistic citation counting methods that are not well-supported by the ISI databases. It is argued that increasingly available digital full text research papers make it possible for citation analysis studies to go beyond what the ISI databases have directly supported and to employ more
sophisticated methods
The Patrolling Problem: Theoretical and Experimental Results
No full textPour une version française de ce chapitre, voir http://basepub.dauphine.fr/xmlui/handle/123456789/3045ou
Bandits Bilinéaires Graphiques Stochastiques
No full textNous introduisons un nouveau modèle appelé Bandits Bilinéaires Graphiques où un apprenant (ou une entité centrale) alloue des bras aux noeuds d’un graphe et observe pour chaque arête une récompense bilinéaire bruitée représentant l’interaction entre les deux noeuds associés. Dans cette thèse, nous étudions le problème d’identification du meilleur bras et la maximisation des récompenses cumulées. Pour le premier, un apprenant veut trouver l’allocation du graphe maximisant la somme des récompenses bilinéaires obtenues à travers le graphe. Pour le second problème, au cours du processus d’apprentissage, l’apprenant doit faire un compromis entre l’exploration des bras pour acquérir une connaissance précise de l’environnement et l’exploitation des bras qui semblent être les meilleurs pour obtenir la récompense la plus élevée. Quel que soit l’objectif de l’apprenant, le modèle de bandits bilinéaires graphiques révèle un problème combinatoire sous-jacent qui est NP-Dur et qui empêche l’utilisation de tout algorithme existant pour l’identification du meilleur bras (BAI) ou pour la maximisation des récompenses cumulées. Pour cette raison, nous proposons tout d’abord un algorithme d’α-approximation pour le problème NP-Dur sous-jacent, puis nous nous attaquons aux deux problèmes mentionnés ci-dessus. En exploitant efficacement la géométrie du problème du bandit, nous proposons une stratégie d’échantillonnage aléatoire pour le problème BAI avec des garanties théoriques. En particulier, nous caractérisons l’influence de la structure du graphe (par exemple, étoile, complet ou cercle) sur le taux de convergence et proposons des expériences empiriques qui confirment cette dépendance. Pour le problème de la maximisation des récompenses cumulées, nous présentons le premier algorithme basé sur le regret pour les bandits bilinéaires graphiques utilisant le principe d’optimisme face à l’incertitude. L’analyse théorique de la méthode présentée borne l’α-regret par Õ(√T ) et souligne l’impact de la structure du graphe sur le taux de convergence. Enfin, nous démontrons par diverses expériences la validité de nos approches.We introduce a new model called Graphical Bilinear Bandits where a learner (or a central entity) allocates arms to nodes of a graph and observes for each edge a noisy bilinear reward representing the interaction between the two end nodes. In this thesis, we study the best arm identification problem and the maximization of cumulative rewards. For the first problem, a learner wants to find the graph allocation maximizing the sum of the bilinear rewards obtained through the graph. For the second problem, during the learning process, the learner has to make a trade-off between exploring the arms to gain accurate knowledge of the environment and exploiting the arms that appear to be the bests to obtain the highest reward. Regardless of the learner’s goal, the graphical bilinear bandit model reveals an underlying NP-Hard combinatorial problem that precludes the use of any existing best arm identification (BAI) or regret-based algorithms. For this reason, we first propose an α-approximation algorithm for the underlying NP-hard problem, and then tackle the two problems mentioned above. By efficiently exploiting the geometry of the bandit problem, we propose a random sampling strategy for the BAI problem with theoretical guarantees. In particular, we characterize the influence of the graph structure (e.g., star, complete or circle) on the convergence rate and propose empirical experiments that confirm this dependence. For the problem of maximizing the cumulative rewards, we present the first regret-based algorithm for graphical bilinear bandits using the principle of optimism in the face of uncertainty. Theoretical analysis of the presented method gives an upper bound of Õ(√T ) on the α-regret and highlights the impact of the graph structure on the convergence rate. Finally, we demonstrate by various experiments the validity of our approaches
- …
