J. Ignacio Serrano

An EEG-based design for the online detection of movement intention

The development of EEG-based wearable technologies for real-life environments has experienced an increasing interest over the last years. During activities of daily living, these systems need to be able to distinguish predefined mental states from the ongoing EEG signal, and these states of interest can be given after long periods of inactivity. A detector of the intention to move that is conceived to be used in real-time is proposed and offline validated with an experimental protocol with long intervals of inactivity that are also used for the detectors validation.

Text representation by a computational model of reading

Traditional document indexing methods, although useful, do not take into account some important aspects of language, such as syntax and semantics. Unlikely, semantic hyperspaces are mathematical and statistical-based techniques that do it. However, although they are an improvement on traditional methods, the output representation is still vector like. This paper proposes a computational model of text reading, called Cognitive Reading Indexing (CRIM), inspired by some aspects of human reading cognition, such as sequential perception, temporality, memory, forgetting and inferences. The model produces not vectors but nets of activated concepts. This paper is focused on indexing or representing documents that way so that they can be labeled or retrieved, presenting promising results. The system was applied to model human subjects as well, and some interesting results were obtained.

Highly accurate error-driven method for noun phrase detection

We present a new model for detection of noun phrases in unrestricted text, whose most outstanding feature is its flexibility: the system is able to recognize noun phrases similar enough to the ones given by the inferred noun phrase grammar. The system provides a probabilistic finite-state automaton able to recognize the part-of-speech tag sequences which define a noun phrase. The recognition flexibility is possible by using a very accurate set of rankings for the FSA transitions. These accurate rankings are obtained by means of an evolutionary algorithm, which works with both, positive and negative examples of the language, thus improving the system coverage while maintaining its precision. We have tested the system on different corpora and evaluated different aspects of the system performance. We have also investigated other ways of improving the performance such as the application of certain filters in the training sets. The comparison of our results with other systems has revealed a considerable performance improvement.

An interactive hybrid system for identifying and filtering unsolicited e-mail

This paper presents a system for automatically detecting and filtering unsolicited electronic messages. The underlying hybrid filtering method is based on e-mail origin and content. The system classifies each of the three parts of e-mails separately by using a sinole Bayesian filter together with a heuristic knowledge base. The system extracts heuristic knowledge from a set of labelled words as the basis on which to begin filtering instead of conducting a training stage using a historic body of pre-classified e-mails. The classification resulting from each part is then integrated to achieve optimum effectiveness. The heuristic knowledge base allows the system to carry out intelligent management of the increase in filter vocabularies and thus ensures efficient classification. The system is dynamic and interactive and the role of the user is essential to keep the evolution of the system up to date by incremental machine learning with the evolution of spam. The user can interact with the system over a customized, friendly interface, in real time or at intervals of the user’s choosing.

Detection of the Onset of Voluntary Movements Based on the Combination of ERD and BP Cortical Patterns

The electroencephalographic activity allows the characterization of movement-related cortical processes. This information may lead to novel rehabilitation technologies with the patients’ cortical activity taking an active role during the intervention. For such applications, the reliability of the estimations based on the electroencephalographic activity is critical both in terms of specificity and temporal accuracy. In this study, a detector of the onset of voluntary upper-limb reaching movements based on cortical rhythms and slow cortical potentials is proposed. To that end, upper-limb movements and cortical activity were recorded while participants performed self-paced movements. A logistic regression combined the output of two classifiers: a) a naive Bayes trained to detect the event-related desynchronization at the movement onset, and b) a matched filter detecting the bereitschaftspotential. On average, 74.5±10.8 % of the movements were detected and 1.32 ± 0.87 false detections were generated per minute. The detections were performed with an average latency of -89.9 ± 349.2 ms with respect to the actual movements. Therefore, the combination of two different sources of information (event-related desynchronization and bereitschaftspotential) is proposed as a way to boost the performance of this kind of systems.

Dealing with written language semantics by a connectionist model of cognitive reading

Although machines perform much better than human beings in most of the tasks, it is not the case of natural language processing. Computational linguistic systems usually rely on mathematical and statistical formalisms, which are efficient and useful but far from human procedures and therefore not so skilled. This paper proposes a computational model of natural language reading, called Cognitive Reading Indexing Model (CRIM), inspired by some aspects of human cognition, that tries to become as more psychologically plausible as possible. The model relies on a semantic neural network and it produces not vectors but nets of activated concepts as text representations. Based on these representations, measures of semantic similarity are also defined. Human comparison results show that the system is suitable to model human reading. Additional results also point out that the system could be used in real applications concerning natural language processing tasks.

Detecting phishing e-mails by heterogeneous classification

This paper presents a system for classifying e-mails into two categories, legitimate and fraudulent. This classifier system is based on the serial application of three filters: a Bayesian filter that classifies the textual content of e-mails, a rule- based filter that classifies the non grammatical content of e-mails and, finally, a filter based on an emulator of fictitious accesses which classifies the responses from websites referenced by links contained in e-mails. This system is based on an approach that is hybrid, because it uses different classification methods, and also integrated, because it takes into account all kind of data and information contained in e-mails. This approach aims to provide an effective and efficient classification. The system first applies fast and reliable classification methods, and only when the resulting classification decision is imprecise does the system apply more complex analysis and classification methods.

A data mining approach using cortical thickness for diagnosis and characterization of essential tremor.

Essential tremor (ET) is one of the most prevalent movement disorders. Being that it is a common disorder, its diagnosis is considered routine. However, misdiagnoses may occur regularly. Over the past decade, several studies have identified brain morphometric changes in ET, but these changes remain poorly understood. Here, we tested the informativeness of measuring cortical thickness for the purposes of ET diagnosis, applying feature selection and machine learning methods to a study sample of 18 patients with ET and 18 age- and sex-matched healthy control subjects. We found that cortical thickness features alone distinguished the two, ET from controls, with 81% diagnostic accuracy. More specifically, roughness (i.e., the standard deviation of cortical thickness) of the right inferior parietal and right fusiform areas was shown to play a key role in ET characterization. Moreover, these features allowed us to identify subgroups of ET patients as well as healthy subjects at risk for ET. Since treatment of tremors is disease specific, accurate and early diagnosis plays an important role in tremor management. Supporting the clinical diagnosis with novel computer approaches based on the objective evaluation of neuroimage data, like the one presented here, may represent a significant step in this direction.

An integrated approach to filtering phishing e-mails

This paper presents a system for classifying e-mails into two categories, legitimate and fraudulent. This classifier system is based on the serial application of three filters: a Bayesian filter that classifies the textual content of e-mails, a rule based filter that classifies the nongrammatical content of e-mails and, finally, a filter based on an emulator of fictitious accesses which classifies the responses from websites referenced by links contained in e-mails. The approach of this system is hybrid, because it uses different classification methods, and also integrated, because it takes into account all kind of data and information contained in e-mails.

Cross-Linguistic Cognitive Modeling of Verbal Morphology Acquisition

How children acquire and process inflectional morphology is still an open question. Despite the fact that English past tense acquisition has been studied and modeled in depth, the current approaches do not account for many of the errors made by humans. Moreover, not much work has been done with highly inflected languages, like Spanish. However, the modeling of any linguistic phenomenon in different languages is very important in order to understand the general cognitive processes underlying each particular phenomenon. This paper presents an ACT-R dual-mechanism model that accomplishes the task of acquiring verbal morphology systems from one of the simplest systems (the English one) to one of the most complex systems (the Spanish one), by using a double analogy process of stem and suffix. The model proposed was able to match all types of errors that developing children make (from a sample of them), both in English and Spanish. The models for both languages used very similar parameters. The introduced approach not only shows how children could acquire a highly inflected morphology system in terms of dual-mechanism theories but, given its cross-linguistic character, also sheds light on the possible general processes involved in the acquisition and processing of inflectional morphology.