Elisa Fromont | Université Lyon (original) (raw)

Papers by Elisa Fromont

This paper gives a comparison of two dif- ferent systems that induce cardiac arrhyth- mia rules b... more This paper gives a comparison of two dif- ferent systems that induce cardiac arrhyth- mia rules by symbolic learning: Kardio and Calicot. In particular, it proposes a de- tailed methodology to compare them and gives some results of this comparison.

Dagstuhl Seminars, 2007

We are interested in using parallel universes to learn interpretable models that can be subsequen... more We are interested in using parallel universes to learn interpretable models that can be subsequently used to automatically diagnose cardiac arrhythmias. In our study, parallel universes are heterogeneous sources such as electrocardiograms, blood pressure measurements, phonocardiograms etc. that give relevant information about the cardiac state of a patient. To learn interpretable rules, we use an inductive logic programming (ILP) method on a symbolic version of our data. Aggregating the symbolic data coming from all the sources before learning, increases both the number of possible relations that can be learned and the richness of the language. We propose a two-step strategy to deal with these dimensionality problems when using ILP. First, rules are learned independently in each universe. Second, the learned rules are used to bias a new learning process from the aggregated data. The results show that this method is much more efficient than learning directly from the aggregated data. Furthermore the good accuracy results confirm the benefits of using multiple sources when trying to improve the diagnosis of cardiac arrhythmias.

Computing Research Repository, 2009

This paper formalises the concept of learning symbolic rules from multisource data in a cardiac m... more This paper formalises the concept of learning symbolic rules from multisource data in a cardiac monitoring context. Our sources, electrocardiograms and arterial blood pressure measures, describe cardiac behaviours from different viewpoints. To learn interpretable rules, we use an Inductive Logic Programming (ILP) method. We develop an original strategy to cope with the dimensionality issues caused by using this ILP technique

Lecture Notes in Computer Science, 2010

As an alternative to vector representations, a recent trend in image classification suggests to i... more As an alternative to vector representations, a recent trend in image classification suggests to integrate additional structural information in the description of images in order to enhance classification accuracy. Rather than being represented in a p-dimensional space, images can typically be encoded in the form of strings, trees or graphs and are usually compared either by computing suited metrics such as the (string or tree)-edit distance, or by testing subgraph isomorphism. In this paper, we propose a new way for representing images in the form of strings whose symbols are weighted according to a TF-IDF-based weighting scheme, inspired from information retrieval. To be able to handle such real-valued weights, we first introduce a new weighted string edit distance that keeps the properties of a distance. In particular, we prove that the triangle inequality is preserved which allows the computation of the edit distance in quadratic time by dynamic programming. We show on an image classification task that our new weighted edit distance not only significantly outperforms the standard edit distance but also seems very competitive in comparison with standard histogram distances-based approaches.

Lecture Notes in Computer Science, 2004

This paper proposes an efficient method to learn from multi source data with an Inductive Logic P... more This paper proposes an efficient method to learn from multi source data with an Inductive Logic Programming method. The method is based on two steps. The first one consists in learning rules independently from each source. In the second step the learned rules are used to bias a new learning process from the aggregated data. We validate this method on cardiac data obtained from electrocardiograms or arterial blood pressure measures. Our method is compared to a single step learning on aggregated data.

Lecture Notes in Computer Science, 2012

This paper shows a concrete example of the use of graph mining for tracking objects in videos wit... more This paper shows a concrete example of the use of graph mining for tracking objects in videos with moving cameras and without any contextual information on the objects to track. To make the mining algorithm efficient, we benefit from a video representation based on dynamic (evolving through time) planar graphs. We then define a number of constraints to efficiently find our so-called spatio-temporal graph patterns. Those patterns are linked through an occurrences graph to allow us to tackle occlusion or graph features instability problems in the video. Experiments on synthetic and real videos show that our method is effective and allows us to find relevant patterns for our tracking application.

Proceedings of the 13th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD '07, 2007

Proceedings of the British Machine Vision Conference 2014, 2014

Studies in Computational Intelligence, 2010

Monitoring patients in intensive care units is a critical task. Simple condition detection is gen... more Monitoring patients in intensive care units is a critical task. Simple condition detection is generally insufficient to diagnose a patient and may generate many false alarms to the clinician operator. Deeper knowledge is needed to discriminate among alarms those that necessitate urgent therapeutic action. We propose an intelligent monitoring system that makes use of many artificial intelligence techniques: artificial neural networks for temporal abstraction, temporal reasoning, model based diagnosis, decision rule based system for adaptivity and machine learning for knowledge acquisition. To tackle the difficulty of taking context change into account, we introduce a pilot aiming at adapting the system behavior by reconfiguring or tuning the parameters of the system modules. A prototype has been implemented and is currently experimented and evaluated. Some results, showing the benefits of the approach, are given.

Abstract. This paper proposes an efficient method to learn from multi source data with an inducti... more Abstract. This paper proposes an efficient method to learn from multi source data with an inductive logic programming method. The method is based on two steps. The first one consists in learning rules independently from each source. In the second step the learnt ...

All currently known algorithms for learning decision trees are based on the paradigm of heuristic... more All currently known algorithms for learning decision trees are based on the paradigm of heuristic top-down induction. Although the results of these algorithms are usually good, there is no guarantee that the resulting trees are really as small, accurate or shallow as possible. In this paper, we introduce an al- gorithm for inducing the smallest most accu- rate decision tree

Nous présentons DL8, un algorithme permettant d'apprendre des arbres de décision sous contraintes... more Nous présentons DL8, un algorithme permettant d'apprendre des arbres de décision sous contraintes. Cet algorithme permet d'optimiser des critères de taille, de profondeur et de précision de l'arbre. Un algorithme exact est intéressant du point de vue pratique comme du point de vue purement scientifique. Il peut, par exemple, être utilisé comme référence pour évaluer les performances et comprendre le comportement des systèmes d'apprentissage d'arbres de décision utilisant des heuristiques. Du point de vue applicatif, il peut permettre de découvrir des arbres ne pouvant pas être appris par ces systèmes d'apprentissage. DL8 repose essentiellement sur la relation existant entre les contraintes applicables aux arbres de décision et celles applicables aux itemsets. Nous proposons d'exploiter des treillis d'itemsets pour extraire des arbres de décision optimaux en temps linéaire et donnons différentes stratégies permettant de construire ces treillis efficacement. Nos expériences montrent que la précision en test de DL8 est meilleure que celle de systèmes tel que C4.5 en utilisant les mêmes contraintes, ce qui confirme les résultats stipulant qu'une recherche exhaustive n'entraine pas forcement un sur-apprentissage. Ces expériences prouvent également que DL8 est un outil utile et intéressant pour apprendre des arbres de décision sous contraintes. Mots-clés : Arbres de décision, recherche d'itemsets fréquents, treillis d'itemsets, analyse de concepts formels, fouille de données sous contraintes.