The aim of this study was to identify specific cognitive patterns related to long-term vocational training outcome. Records of twenty-eight patients who had benefited from a professional rehabilitation program were retrospectively processed. Screening through machine learning algorithms of patients' neuropsychological scores identified cognitive patterns related to both vocational training outcomes: succeeded or failed. These patterns were based on cognitive performance intervals even if the cognitive ability was not impaired. The cognitive pattern related to a successful vocational training included performance intervals on measures underlying verbal memory consolidation, visual memory incidental recall, problem solving and planning abilities. The cognitive pattern explaining failure of vocational training included performance intervals on tasks involving planning and problem solving abilities. From a comprehensive neuropsychological battery, memory and executive measures appeared to be the best attributes related to the vocational training outcome in patients with brain injury. Even with a cognitive functioning above the pathological cutoff, patients suffering from a brain injury could fail a vocational training. The attributes related to the vocational training outcome would be more a specific level of cognitive functioning rather than an interpretation of neuropsychological scores only based on a normal versus pathological distinction.

Seventy-five percent of people with an acquired brain injury are of working age, most of them are young adults who want to return-to-work or need to resume a vocational training. However, a failure in returning to work/vocational training is associated with depression and isolation. Our aim was to identify objective measures to allow clinicians to anticipate failure in resuming vocational training following a brain injury. Neuropsychological data from 343 patients with brain injury, who benefited from a vocational rehabilitation program (UEROS Lille, France) between 2002 and 2017, were retrospectively analyzed. After completing this program, 28 patients resumed a vocational training and we identified, trough machine learning algorithms, the neuropsychological measures predicting vocational outcome. Success in resuming vocational training after brain injury was predicted by mnemonic scores (verbal forgetting and visual retrieval). The scores obtained on the Tower of London test predicted both success and failure. Our study highlighted that neuropsychological measures underlying planning abilities could help clinicians to anticipate a failure in resuming vocational training in patients with brain injury. Planning and memory abilities would predict an effective vocational outcome. These initial results, that need to be confirmed with larger samples of patients with brain injury who resumed vocational training, may have a relevant implication for neuropsychological practice, allowing a better vocational guidance of these patients.

We address the problem of multi-task learning with no label correspondence among tasks. Learning multiple related tasks simultaneously, by exploiting their shared knowledge can improve the predictive performance on every task. We develop the multi-task Adaboost environment with Multi-Task Decision Trees as weak classifiers. We first adapt the well known decision tree learning to the multi-task setting. We revise the information gain rule for learning decision trees in the multi-task setting. We use this feature to develop a novel criterion for learning Multi-Task Decision Trees. The criterion guides the tree construction by learning the decision rules from data of different tasks, and representing different degrees of task relatedness. We then modify MT-Adaboost to combine Multi-task Decision Trees as weak learners. We experimentally validate the advantage of the new technique; we report results of experiments conducted on several multi-task datasets, including the Enron email set and Spam Filtering collection.

Dans cet article, nous presentons un cadre d'apprentissage general pour la classification supervisee. Ce cadre ne necessite que la definition d'un operateur de generalisation et fournit en particulier des methodes d'ensemble. Pour les tâches de classification de sequences, nous montrons que l'inference grammaticale, avec des objectifs differents, a deja defini de tels apprenants pour certaines familles d'automates comme les reversibles ou les k-TSS. Nous proposons ensuite un operateur de generalisation original pour la famille des boules de mots. Enfin, nous montrons au travers de differentes experimentations que notre approche permet effectivement de resoudre des tâches de classification de sequences.

In this paper, we present a general framework for supervised classification. This framework provides methods like boosting and only needs the definition of a generalisation operator called lgg. For sequence classification tasks, lgg is a learner that only uses positive examples. We show that grammatical inference has already defined such learners for automata classes like reversible automata or k-TSS automata. Then we propose a generalisation algorithm for the class of balls of words. Finally, we show through experiments that our method efficiently resolves sequence classification tasks.

Learning multiple related tasks from data simultaneously can improve predictive performance relative to learning these tasks independently. In this paper we propose a novel multi-task learning algorithm called MT-Adaboost: it extends Adaboost algorithm Freund1999Short to the multi-task setting, it uses as multi-task weak classifier a multi-task decision stump. This allows to learn different dependencies between tasks for different regions of the learning space. Thus, we relax the conventional hypothesis that tasks behave similarly in the whole learning space. Moreover, MT-Adaboost can learn multiple tasks without imposing the constraint of sharing the same label set and/or examples between tasks. A theoretical analysis is derived from the analysis of the original Adaboost. Experiments for multiple tasks over large scale textual data sets with social context (Enron and Tobacco) give rise to very promising results.

Document decrivant la demonstration du systeme Volata effectuee lors de la conference CAp - Conference Francophone sur l'Apprentissage Automatique - 2010.

Résumé : Nous présentons un cadre général pour la classification supervisée basé sur la notion de moindre généralisé (généralisation minimale d'un ensemble d'exemples). Nous montrons que le fait de définir cette opération dans l'espace d'hypothèses choisi met à notre disposition, sans effort supplémentaire, plusieurs algorithmes d'apprentissage supervisé à plusieurs classes, en particulier des mé- thodes d'ensemble (de boosting par exemple). Après avoir décrit cette architec- ture générique, nous expliquons son utilisation pour l'Inférence Grammaticale : le calcul du moindre généralisé y est un apprentissage à partir d'exemples positifs seuls comme ceux effectués par les algorithmes TSSI et ZR. Des classifications efficaces de séquences sont alors possibles par vote de plusieurs automates élé- mentaires. Des expérimentations valident notre approche et nous discutons des intérêts et implications de ces idées. Mots-clés : Inférence Grammaticale, méthodes d'ensemble, moindres générali- sés, classification supervisée.

Nous définissons les familles de langages rationnels à résiduels k-disjoints et nous présentons les possibilités d'identification de ces familles. Chaque famille de langages rationnels à résiduels k-disjoints contient la famille correspondante de langages k-réversibles. L'union des familles, pour k entier naturel, couvre tous les langages rationnels. Nous montrons que chacune de ces familles est identifiable à la limite, en temps et données polynômiaux à partir d'exemples positifs, en les représentant par des AFD.

This paper is about the evaluation of the results of clustering algorithms, and the comparison of such algorithms. We propose a new method based on the enrichment of a set of independent labeled datasets by the results of clustering, and the use of a supervised method to evaluate the interest of adding such new information to the datasets. We thus adapt the cascade generalization [1] paradigm in the case where we combine an unsupervised and a supervised learner. We also consider the case where independent supervised learnings are performed on the different groups of data objects created by the clustering [2]. We then conduct experiments using different supervised algorithms to compare various clustering algorithms. And we thus show that our proposed method exhibits a coherent behavior, pointing out, for example, that the algorithms based on the use of complex probabilistic models outperform algorithms based on the use of simpler models.

Subspace clustering is an extension of traditional clustering that seeks to find clusters embedded in different subspaces within a dataset. This is a particularly important challenge with high dimensional data where the curse of dimensionality occurs. It also has the benefit of providing smaller descriptions of the clusters found. In this field, we show that using probabilistic models provides many advantages over other existing methods. In particular, we show that the difficult problem of the parameter settings of subspace clustering algorithms can be seen as a model selection problem in the framework of probabilistic models. It thus allows us to design a method that does not require any input parameter from the user. We also point out the interest in allowing the clusters to overlap. And finally, we show that it is well suited for detecting the noise that may exist in the data, and that this helps to provide a more understandable representation of the clusters found.