Gérard Dray

Applications of artificial neural networks

In this article we consider some theoretical aspects of neural networks and some of their varied applications. The theoretical aspects are presented from the point of view of a system, basically input/state/output. For the applications, we consider large systems: from production systems, through biological and chemical systems and on to environmental systems.

Web opinion mining: how to extract opinions from blogs?

The growing popularity of Web 2.0 provides with increasing numbers of documents expressing opinions on different topics. Recently, new research approaches have been defined in order to automatically extract such opinions from the Internet. They usually consider opinions to be expressed through adjectives, and make extensive use of either general dictionaries or experts to provide the relevant adjectives. Unfortunately, these approaches suffer from the following drawback: in a specific domain, a given adjective may either not exist or have a different meaning from another domain. In this paper, we propose a new approach focusing on two steps. First, we automatically extract a learning dataset for a specific domain from the Internet. Secondly, from this learning set we extract the set of positive and negative adjectives relevant to the domain. The usefulness of our approach was demonstrated by experiments performed on real data.

NIRS-measured prefrontal cortex activity in neuroergonomics: strengths and weaknesses

Contemporary daily life is more and more characterized by ubiquitous interaction with computational devices and systems. For example, it is commonplace for a person walking a busy street, to be engaged in conversation with a distant person using telephony, while simultaneously receiving directions via a GPS-enabled web application on their mobile device. This overwhelming increase in human-computer interactions has prompted the need for a better understanding of how brain activity is shaped by performing sensorimotor actions in the physical world. In this context, neuroergonomics aims at bridging the gap between the abundant flow of information contained within a persons technological environment and related brain activity in order to adapt machine settings and facilitate optimal human-computer interactions (Parasuraman, 2013). One way to achieve this goal consists in developing adaptive systems. In neuroergonomics, adaptive automation relies on passive brain-computer interfaces (BCI) capable of spotting brain signatures linked to the operators cognitive state in order to adjust in real-time the operators technological environment. With the growing area of interest in this topic, the need for neuroimaging methods properly suited to ecological experimental settings has risen. In this vein, near-infrared spectroscopy (NIRS) presents some advantages as compared to other neuroimaging methods. In this opinion article, we first concentrate on the benefits of utilizing NIRS for investigation in neuroergonomics. Recent neuroergonomics investigations have used NIRS recordings in a number of laboratories (e.g., Ayaz et al., 2012; Mandrick et al., 2013a,b). It is particularly worth noting that most of these investigations have reported NIRS data from the prefrontal cortex (PFC). We provide a brief review of these recent studies and their impact in the field by presenting a detailed analysis of the applicability of NIRS-measured PFC activity to discriminate cognitive states in real life environments. In this paper, we will address two main questions: are NIRS-derived hemodynamic variables sufficiently sensitive to changes in sustained attention when measured over the PFC area? Are these measures useful for delineating different levels of mental workload?

Prefrontal cortex activity during motor tasks with additional mental load requiring attentional demand: a near-infrared spectroscopy study.

Functional near-infrared spectroscopy (fNIRS) is suitable for investigating cerebral oxygenation changes during motor and/or mental tasks. In the present study, we investigated how an additional mental load during a motor task at two submaximal loadings affects the fNIRS-measured brain activation over the right prefrontal cortex (PFC). Fifteen healthy males performed isometric grasping contractions at 15% and 30% of the maximal voluntary contraction (MVC) with or without an additional mental (i.e., arithmetic) task. Mental performance, force variability, fNIRS and subjective perception responses were measured in each condition. The performance of the mental task decreased significantly while the force variability increased significantly at 30% MVC as compared to 15% MVC, suggesting that performance of dual-task required more attentional resources. PFC activity increased significantly as the effort increased from 15% to 30% MVC (p<.001). Although a larger change in the deoxyhemoglobin was observed in dual-task conditions (p=.051), PFC activity did not change significantly as compared to the motor tasks alone. In summary, participants were unable to invest more attention and effort in performing the more difficult levels in order to maintain adequate mental performance.

Towards a near infrared spectroscopy-based estimation of operator attentional state.

Given the critical risks to public health and safety that can involve lapses in attention (e.g., through implication in workplace accidents), researchers have sought to develop cognitive-state tracking technologies, capable of alerting individuals engaged in cognitively demanding tasks of potentially dangerous decrements in their levels of attention. The purpose of the present study was to address this issue through an investigation of the reliability of optical measures of cortical correlates of attention in conjunction with machine learning techniques to distinguish between states of full attention and states characterized by reduced attention capacity during a sustained attention task. Seven subjects engaged in a 30 minutes duration sustained attention reaction time task with near infrared spectroscopy (NIRS) monitoring over the prefrontal and the right parietal areas. NIRS signals from the first 10 minutes of the task were considered as characterizing the ‘full attention’ class, while the NIRS signals from the last 10 minutes of the task were considered as characterizing the ‘attention decrement’ class. A two-class support vector machine algorithm was exploited to distinguish between the two levels of attention using appropriate NIRS-derived signal features. Attention decrement occurred during the task as revealed by the significant increase in reaction time in the last 10 compared to the first 10 minutes of the task (p<.05). The results demonstrate relatively good classification accuracy, ranging from 65 to 90%. The highest classification accuracy results were obtained when exploiting the oxyhemoglobin signals (i.e., from 77 to 89%, depending on the cortical area considered) rather than the deoxyhemoglobin signals (i.e., from 65 to 66%). Moreover, the classification accuracy increased to 90% when using signals from the right parietal area rather than from the prefrontal cortex. The results support the feasibility of developing cognitive tracking technologies using NIRS and machine learning techniques.

Is a Voting Approach Accurate for Opinion Mining

In this paper, we focus on classifying documents according to opinion and value judgment they contain. The main originality of our approach is to combine linguistic pre-processing, classification and a voting system using several classification methods. In this context, the relevant representation of the documents allows to determine the features for storing textual data in data warehouses. The conducted experiments on very large corpora from a French challenge on text mining (DEFT) show the efficiency of our approach.

Towards an automatic characterization of criteria

The number of documents is growing exponentially with the rapid expansion of the Web. The new challenge for Internet users is now to rapidly find appropriate data to their requests. Thus information retrieval, automatic classification and detection of opinions appear as major issues in our information society. Many efficient tools have already been proposed to Internet users to ease their search over the web and support them in their choices. Nowadays, users would like genuine decision tools that would efficiently support them when focusing on relevant information according to specific criteria in their area of interest. In this paper, we propose a new approach for automatic characterization of such criteria. We bring out that this approach is able to automatically build a relevant lexicon for each criterion. We then show how this lexicon can be useful for documents classification or segmentation tasks. Experiments have been carried out with real datasets and show the efficiency of our proposal.

Knowledge Transfer for Reducing Calibration Time in Brain-Computer Interfacing

Reducing calibration time while maintaining good classification accuracy has been one of the most challenging problems in electroencephalography (EEG) -based brain-computer interfaces (BCIs) research during the last years. Most of machine learning approaches that have been attempted to address this issue are based on knowledge transfer between different BCIs users. Assuming that there is a common underlying data generating process, they try to learn a subject-independent classification model from multiple users in order to classify data of future users. In this paper, we propose a novel approach that allows inter-subjects classification of EEG signals without relying on the strong assumptions considered in previous work. It consists of learning a prediction model of a new BCI user through an ensemble of classifiers where base classifiers are trained on data from other users separately and weighted according to the performance of the ensemble on few labeled data of the new user. Evaluation on real EEG data showed that our approach allows achieving good classification accuracy when the size of calibration set is small.

Web Information Retrieval Based on User Profile

With the growing popularity of the World Wide Web, the amount of available information is so great that finding the right and useful information becomes a very hard task for an end user. In this paper, we propose a new approach for personal Web information retrieval. The originality of our approach is a choice of indexing terms depending on the user request but also on his profile. The general idea is to consider that the need of a user depends on his request but also on his knowledge acquired through time on the thematic of his request.

Opinion Extraction Applied to Criteria

The success of Information technologies and associated services (e.g., blogs, forums,...) eases the way to express massive opinion on various topics. Recently new techniques known as opinion mining have emerged. One of their main goals is to automatically extract a global trend from expressed opinions. While it is quite easy to get this overall assessment, a more detailed analysis will highlight that opinions are expressed on more specific topics: one will acclaim a movie for its soundtrack and another will criticize it for its scenario. Opinion mining approaches have little explored this multicriteria aspect. In this paper we propose an automatic extraction of text segments related to a set of criteria. The opinion expressed in each text segment is then automatically extracted. From a small set of opinion keywords, our approach automatically builds a training set of texts from the web. A lexicon reflecting the polarity of words is then extracted from this training corpus. This lexicon is then used to compute the polarity of extracted text segments. Experiments show the efficiency of our approach.