The present dissertation on autism spectrum disorders (ASD) addressed several questions. First, the behavioral symptoms of speaking and mute individuals with ASD were compared on the Autism Behavior Checklist (ABC), a commonly used diagnostic instrument. In addition, questions about the construction of the ABC were investigated. In Study I it was found that although the two groups did not significantly differ on the total ABC score, the mute group demonstrated significantly more pathology on 21 of 57 items and 3 of 5 subscales. The speaking group obtained significantly higher scores on only 8 items and 1 subscale (Language). The appropriateness of providing greater pathology scores to speaking, rather than mute, individuals with ASD was called into question. In addition, it was speculated that the expressive language items are weighted too heavily, on both the Language subscale and the total score. Second, because the 57 items of the ABC were categorized into 5 subscales on a subjective basis, an empirical investigation of the factor structure of the ABC was performed. In Study II, a 5-factor model that contained 39 of the items accounted for 80% of the total variance in the checklist. No support was found for classifying the 57 items into the existing 5 subscales. Finally, the two most common forms of instruction to teach children with ASD, discrete-trial teaching and incidental teaching, were assessed. In Study III it was ascertained that discrete-trial teaching was more efficient and produced faster acquisition and initially, greater generalization. However, by follow-up, the incidental teaching methods resulted in equal retention, greater generalization, and equal or greater spontaneous usage. The findings indicate that although it takes a longer time for children with autism to learn with incidental teaching procedures, once they have acquired an ability, it may be more permanent.

Many prey species lack innate recognition of their potential predators. Hence, learning is required for them to recognize and respond to predation threats. When wild-caught, these same species may show amazing sophistication in their responses to predator cues. They are able to adjust the intensity of their antipredator responses to a particular predator according to the degree of threat posed by that predator. This ability is therefore acquired through learning. While many studies have shown that prey can learn to respond to predator cues through different learning modes, little is known about what the prey are actually learning. The results presented in this thesis show that learned predator recognition goes beyond the simple labelling of predators as dangerous. Using fathead minnows (Pimephales promelas), woodfrog (Rana sylvatica) tadpoles and boreal chorus frog (Pseudacris maculata) tadpoles, I demonstrated that a one time learning event, either through pairing with alarm cues or through social learning, was enough for prey to learn the level of threat associated with the novel predator cues. I showed that the level of danger associated with the predator cues was determined by the concentration of alarm cues when learning through pairing of alarm cues, or by the intensity of antipredator response displayed by the tutors and by the tutor-to-observer ratio when learning occurred through cultural transmission. Moreover, when subsequently exposed to predator cues, prey adjusted their antipredator responses according to the change in concentration of predator cues between the learning event and the subsequent exposure. Prey displayed stronger antipredator responses when exposed to higher concentrations of predator cues and vice versa. When minnows were provided with conflicting information about the danger level associated with a predator, they displayed a safety strategy and used the most recent information available to respond to predation threats. On a longer time scale, the data also suggest that woodfrog tadpoles are able to learn to respond to predation threats according to the risk posed by the predator at different times of day. Finally, I showed that prey learn to recognize particular characteristics of predators and can generalize their antipredator responses to novel species sharing those characteristics. However, generalization of predator recognition is dependent on the level of risk associated with the predator. Threat-sensitive learning is an extremely complex process shaped by the millions of years of selection imposed by predators on prey.

Virtual 3D city and landscape models are the main subject investigated in this thesis. They digitally represent urban space and have many applications in different domains, e.g., simulation, cadastral management, and city planning. Visualization is an elementary component of these applications. Photo-realistic visualization with an increasingly high degree of detail leads to fundamental problems for comprehensible visualization. A large number of highly detailed and textured objects within a virtual 3D city model may create visual noise and overload the users with information. Objects are subject to perspective foreshortening and may be occluded or not displayed in a meaningful way, as they are too small. In this thesis we present abstraction techniques that automatically process virtual 3D city and landscape models to derive abstracted representations. These have a reduced degree of detail, while essential characteristics are preserved. After introducing definitions for model, scale, and multi-scale representations, we discuss the fundamentals of map generalization as well as techniques for 3D generalization. The first presented technique is a cell-based generalization of virtual 3D city models. It creates abstract representations that have a highly reduced level of detail while maintaining essential structures, e.g., the infrastructure network, landmark buildings, and free spaces. The technique automatically partitions the input virtual 3D city model into cells based on the infrastructure network. The single building models contained in each cell are aggregated to abstracted cell blocks. Using weighted infrastructure elements, cell blocks can be computed on different hierarchical levels, storing the hierarchy relation between the cell blocks. Furthermore, we identify initial landmark buildings within a cell by comparing the properties of individual buildings with the aggregated properties of the cell. For each block, the identified landmark building models are subtracted using Boolean operations and integrated in a photo-realistic way. Finally, for the interactive 3D visualization we discuss the creation of the virtual 3D geometry and their appearance styling through colors, labeling, and transparency. We demonstrate the technique with example data sets. Additionally, we discuss applications of generalization lenses and transitions between abstract representations. The second technique is a real-time-rendering technique for geometric enhancement of landmark objects within a virtual 3D city model. Depending on the virtual camera distance, landmark objects are scaled to ensure their visibility within a specific distance interval while deforming their environment. First, in a preprocessing step a landmark hierarchy is computed, this is then used to derive distance intervals for the interactive rendering. At runtime, using the virtual camera distance, a scaling factor is computed and applied to each landmark. The scaling factor is interpolated smoothly at the interval boundaries using cubic Bézier splines. Non-landmark geometry that is near landmark objects is deformed with respect to a limited number of landmarks. We demonstrate the technique by applying it to a highly detailed virtual 3D city model and a generalized 3D city model. In addition we discuss an adaptation of the technique for non-linear projections and mobile devices. The third technique is a real-time rendering technique to create abstract 3D isocontour visualization of virtual 3D terrain models. The virtual 3D terrain model is visualized as a layered or stepped relief. The technique works without preprocessing and, as it is implemented using programmable graphics hardware, can be integrated with minimal changes into common terrain rendering techniques. Consequently, the computation is done in the rendering pipeline for each vertex, primitive, i.e., triangle, and fragment. For each vertex, the height is quantized to the nearest isovalue. For each triangle, the vertex configuration with respect to their isovalues is determined first. Using the configuration, the triangle is then subdivided. The subdivision forms a partial step geometry aligned with the triangle. For each fragment, the surface appearance is determined, e.g., depending on the surface texture, shading, and height-color-mapping. Flexible usage of the technique is demonstrated with applications from focus+context visualization, out-of-core terrain rendering, and information visualization. This thesis presents components for the creation of abstract representations of virtual 3D city and landscape models. Re-using visual language from cartography, the techniques enable users to build on their experience with maps when interpreting these representations. Simultaneously, characteristics of 3D geovirtual environments are taken into account by addressing and discussing, e.g., continuous scale, interaction, and perspective. / Gegenstand der Arbeit sind virtuelle 3D-Stadt- und Landschaftsmodelle, die den städtischen Raum in digitalen Repräsentationen abbilden. Sie werden in vielfältigen Anwendungen und zu unterschiedlichen Zwecken eingesetzt. Dabei ist die Visualisierung ein elementarer Bestandteil dieser Anwendungen. Durch realitätsnahe Darstellung und hohen Detailgrad entstehen jedoch zunehmend fundamentale Probleme für eine verständliche Visualisierung. So führt beispielsweise die hohe Anzahl von detailliert ausmodellierten und texturierten Objekten eines virtuellen 3D-Stadtmodells zu Informationsüberflutung beim Betrachter. In dieser Arbeit werden Abstraktionsverfahren vorgestellt, die diese Probleme behandeln. Ziel der Verfahren ist die automatische Transformation virtueller 3D-Stadt- und Landschaftsmodelle in abstrakte Repräsentationen, die bei reduziertem Detailgrad wichtige Charakteristika erhalten. Nach der Einführung von Grundbegriffen zu Modell, Maßstab und Mehrfachrepräsentationen werden theoretische Grundlagen zur Generalisierung von Karten sowie Verfahren zur 3D-Generalisierung betrachtet. Das erste vorgestellte Verfahren beschreibt die zellbasierte Generalisierung von virtuellen 3DStadtmodellen. Es erzeugt abstrakte Repräsentationen, die drastisch im Detailgrad reduziert sind, erhält dabei jedoch die wichtigsten Strukturen, z.B. das Infrastrukturnetz, Landmarkengebäude und Freiflächen. Dazu wird in einem vollautomatischen Verfahren das Eingabestadtmodell mithilfe des Infrastrukturnetzes in Zellen zerlegt. Pro Zelle wird abstrakte Gebäudegeometrie erzeugt, indem die enthaltenen Einzelgebäude mit ihren Eigenschaften aggregiert werden. Durch Berücksichtigung gewichteter Elemente des Infrastrukturnetzes können Zellblöcke auf verschiedenen Hierarchieebenen berechnet werden. Weiterhin werden Landmarken gesondert berücksichtigt: Anhand statistischer Abweichungen der Eigenschaften der Einzelgebäudes von den aggregierten Eigenschaften der Zelle werden Gebäude gegebenenfalls als initiale Landmarken identifiziert. Schließlich werden die Landmarkengebäude aus den generalisierten Blöcken mit Booleschen Operationen ausgeschnitten und realitätsnah dargestellt. Die Ergebnisse des Verfahrens lassen sich in interaktiver 3D-Darstellung einsetzen. Das Verfahren wird beispielhaft an verschiedenen Datensätzen demonstriert und bezüglich der Erweiterbarkeit diskutiert. Das zweite vorgestellte Verfahren ist ein Echtzeit-Rendering-Verfahren für geometrische Hervorhebung von Landmarken innerhalb eines virtuellen 3D-Stadtmodells: Landmarkenmodelle werden abhängig von der virtuellen Kameradistanz vergrößert, so dass sie innerhalb eines spezifischen Entfernungsintervalls sichtbar bleiben; dabei wird ihre Umgebung deformiert. In einem Vorverarbeitungsschritt wird eine Landmarkenhierarchie bestimmt, aus der die Entfernungsintervalle für die interaktive Darstellung abgeleitet werden. Zur Laufzeit wird anhand der virtuellen Kameraentfernung je Landmarke ein dynamischer Skalierungsfaktor bestimmt, der das Landmarkenmodell auf eine sichtbare Größe skaliert. Dabei wird der Skalierungsfaktor an den Intervallgrenzen durch kubisch interpoliert. Für Nicht-Landmarkengeometrie in der Umgebung wird die Deformation bezüglich einer begrenzten Menge von Landmarken berechnet. Die Eignung des Verfahrens wird beispielhaft anhand verschiedener Datensätze demonstriert und bezüglich der Erweiterbarkeit diskutiert. Das dritte vorgestellte Verfahren ist ein Echtzeit-Rendering-Verfahren, das eine abstrakte 3D-Isokonturen-Darstellung von virtuellen 3D-Geländemodellen erzeugt. Für das Geländemodell wird eine Stufenreliefdarstellung für eine Menge von nutzergewählten Höhenwerten erzeugt. Das Verfahren arbeitet ohne Vorverarbeitung auf Basis programmierbarer Grafikkarten-Hardware. Entsprechend erfolgt die Verarbeitung in der Prozesskette pro Geometrieknoten, pro Dreieck, und pro Bildfragment. Pro Geometrieknoten wird zunächst die Höhe auf den nächstliegenden Isowert quantisiert. Pro Dreieck wird dann die Konfiguration bezüglich der Isowerte der drei Geometrieknoten bestimmt. Anhand der Konfiguration wird eine geometrische Unterteilung vorgenommen, so dass ein Stufenausschnitt entsteht, der dem aktuellen Dreieck entspricht. Pro Bildfragment wird schließlich die finale Erscheinung definiert, z.B. anhand von Oberflächentextur, durch Schattierung und Höheneinfärbung. Die vielfältigen Einsatzmöglichkeiten werden mit verschiedenen Anwendungen demonstriert. Die Arbeit stellt Bausteine für die Erzeugung abstrakter Darstellungen von virtuellen 3D-Stadt und Landschaftsmodellen vor. Durch die Orientierung an kartographischer Bildsprache können die Nutzer auf bestehende Erfahrungen bei der Interpretation zurückgreifen. Dabei werden die charakteristischen Eigenschaften 3D geovirtueller Umgebungen berücksichtigt, indem z.B. kontinuierlicher Maßstab, Interaktion und Perspektive behandelt und diskutiert werden.

Many prey species lack innate recognition of their potential predators. Hence, learning is required for them to recognize and respond to predation threats. When wild-caught, these same species may show amazing sophistication in their responses to predator cues. They are able to adjust the intensity of their antipredator responses to a particular predator according to the degree of threat posed by that predator. This ability is therefore acquired through learning. While many studies have shown that prey can learn to respond to predator cues through different learning modes, little is known about what the prey are actually learning. The results presented in this thesis show that learned predator recognition goes beyond the simple labelling of predators as dangerous. Using fathead minnows (Pimephales promelas), woodfrog (Rana sylvatica) tadpoles and boreal chorus frog (Pseudacris maculata) tadpoles, I demonstrated that a one time learning event, either through pairing with alarm cues or through social learning, was enough for prey to learn the level of threat associated with the novel predator cues. I showed that the level of danger associated with the predator cues was determined by the concentration of alarm cues when learning through pairing of alarm cues, or by the intensity of antipredator response displayed by the tutors and by the tutor-to-observer ratio when learning occurred through cultural transmission. Moreover, when subsequently exposed to predator cues, prey adjusted their antipredator responses according to the change in concentration of predator cues between the learning event and the subsequent exposure. Prey displayed stronger antipredator responses when exposed to higher concentrations of predator cues and vice versa. When minnows were provided with conflicting information about the danger level associated with a predator, they displayed a safety strategy and used the most recent information available to respond to predation threats. On a longer time scale, the data also suggest that woodfrog tadpoles are able to learn to respond to predation threats according to the risk posed by the predator at different times of day. Finally, I showed that prey learn to recognize particular characteristics of predators and can generalize their antipredator responses to novel species sharing those characteristics. However, generalization of predator recognition is dependent on the level of risk associated with the predator. Threat-sensitive learning is an extremely complex process shaped by the millions of years of selection imposed by predators on prey.

Previous research has demonstrated the efficacy of a behavioral parent training program for increasing the accuracy of trained skills; however, few studies have examined the extent to which those skills generalize to the natural environment (i.e., the home) and are used with the target individual (i.e., the child). In addition, little is known about the direct effect that caregiver implementation of the skills has on child behavior. A multiple baseline across participants design was used to (a) assess caregiver accuracy with implementation of three parenting skills, and (b) assess subsequent effects of the parenting skills on child behavior. Results demonstrated that three caregiver participants successfully generalized parenting skills taught during behavioral skills training (BST) to naturally occurring routines by recognizing appropriate and inappropriate child behaviors as opportunities to implement the trained skills. In addition, the behavior of each caregiver's child improved following BST, suggesting that the parenting skills were effective in addressing challenging child behavior. All caregivers rated the training and skills to be highly socially valid. Limitations and suggestions for future research are discussed.

In recent years, the crucial importance of metrics in machine learningalgorithms has led to an increasing interest in optimizing distanceand similarity functions using knowledge from training data to make them suitable for the problem at hand.This area of research is known as metric learning. Existing methods typically aim at optimizing the parameters of a given metric with respect to some local constraints over the training sample. The learned metrics are generally used in nearest-neighbor and clustering algorithms.When data consist of feature vectors, a large body of work has focused on learning a Mahalanobis distance, which is parameterized by a positive semi-definite matrix. Recent methods offer good scalability to large datasets.Less work has been devoted to metric learning from structured objects (such as strings or trees), because it often involves complex procedures. Most of the work has focused on optimizing a notion of edit distance, which measures (in terms of number of operations) the cost of turning an object into another.We identify two important limitations of current supervised metric learning approaches. First, they allow to improve the performance of local algorithms such as k-nearest neighbors, but metric learning for global algorithms (such as linear classifiers) has not really been studied so far. Second, and perhaps more importantly, the question of the generalization ability of metric learning methods has been largely ignored.In this thesis, we propose theoretical and algorithmic contributions that address these limitations. Our first contribution is the derivation of a new kernel function built from learned edit probabilities. Unlike other string kernels, it is guaranteed to be valid and parameter-free. Our second contribution is a novel framework for learning string and tree edit similarities inspired by the recent theory of (epsilon,gamma,tau)-good similarity functions and formulated as a convex optimization problem. Using uniform stability arguments, we establish theoretical guarantees for the learned similarity that give a bound on the generalization error of a linear classifier built from that similarity. In our third contribution, we extend the same ideas to metric learning from feature vectors by proposing a bilinear similarity learning method that efficiently optimizes the (epsilon,gamma,tau)-goodness. The similarity is learned based on global constraints that are more appropriate to linear classification. Generalization guarantees are derived for our approach, highlighting that our method minimizes a tighter bound on the generalization error of the classifier. Our last contribution is a framework for establishing generalization bounds for a large class of existing metric learning algorithms. It is based on a simple adaptation of the notion of algorithmic robustness and allows the derivation of bounds for various loss functions and regularizers.

Children with phonological disorder are at risk of developing reading and writing impairments. Phonological difficulties may also influence a child’s social interaction and it is for this reason that effective treatment of phonological disorder is important. One factor that may influence treatment efficacy, is treatment intensity. There has been little research done in Sweden, on the effects of intensity of phonological intervention. Further research into phonological intervention is therefore needed to determine which treatment intensity is the most effective. The aim of this study is to evaluate the possible effects of intensive treatment of two preschool children with phonological disorder. The research questions focus on the nature of the possible effects of phonological treatment and if they lead to generalization, and whether caregivers notice any changes in their children’s speech. A single-subject design was used with three baseline measurements, followed by treatment and two follow up measurements. The treatment was given three times a week over the course of three weeks, and every treatment session lasted for approximately 40 minutes. The results of the baseline and follow ups were calculated with Percentage Consonants Correct (PCC), Percentage Words Correct (PWC), and the percentage of manner of articulation correct. Both participants displayed improvements between the baseline and the second follow up, in all the measurements. Changes could be seen in the target phonemes both in new positions and in new words, but also in untrained phonemes. The caregivers of both participants experienced an improvement in their children's speech. How the estimated intensity affected the results is unclear. However, the results indicated that intensive treatment led to generalization and that it showed effect. Thus, the results of the present study could have clinical relevance. / Förskolebarn med fonologiska svårigheter riskerar att få läs- och skrivsvårigheter i skolåldern och påverkan på det sociala samspelet. Det är därför viktigt med effektiv fonologisk intervention, som kan påverkas av intensiteten. Det finns sparsamt med svenska studier om intensitetens påverkan vid fonologisk intervention. Föreliggande studie syftar till att utvärdera eventuell effekt av intensiv behandling hos två förskolebarn med fonologiska svårigheter. Frågeställningarna innefattar hur den eventuella effekten av fonologisk behandling ser ut, om den leder till generalisering, samt om vårdnadshavarna upplever någon förändring.  Studien hade en single-subject design med tre baslinjemätningar och två uppföljande mätningar. Fonemtestet LINköpingsUnderSökningen (LINUS) användes vid samtliga mätningar. Behandlingen genomfördes vid nio tillfällen, tre gånger i veckan i tre veckor. Varje behandlingstillfälle var cirka 40 minuter långt. Resultatet av LINUS-testningen beräknades med Percentage Consonants Correct (PCC), Percentage Words Correct (PWC) och andel korrekt artikulationssätt och analyserades även kvalitativt. Resultaten jämfördes mellan de olika mätningarna. Båda deltagarna uppvisade förbättringar för samtliga mått mellan baslinjemätningarna och den sista uppföljande mätningen. Förändringar kunde ses av tränade ljud till nya ord och positioner samt av ljud som inte tränades i behandling. Vårdnadshavarna till de deltagande barnen upplevde att deras barns tal hade förbättrats efter behandlingen. Det är svårt att uttala sig om vilken inverkan den uppskattade intensiteten hade på behandlingsutgången. Resultatet indikerar dock att intensiv behandling hade effekt och att den ledde till generalisering, och skulle således kunna vara till nytta i klinisk verksamhet.

Une carte multi-échelle est un ensemble de cartes à différentes échelles, dans lequel l’utilisateur peut naviguer via un géoportail. Chacune de ces cartes est préalablement construite par généralisation cartographique, processus qui adapte la représentation cartographique à une échelle donnée. Les changements de représentations qu’implique la généralisation entre deux cartes à différentes échelles sont susceptibles de perturber l’utilisateur, rendant sa navigation plus difficile. Nous proposons dans cette thèse d’ajouter des représentations intermédiaires dans une carte multi-échelle existante, pour créer une évolution plus fluide du contenu cartographique au fil des échelles. Alors que de solides connaissances théoriques existent pour la conception cartographique traditionnelle, on ne sait pas encore comment concevoir une carte multi-échelle efficace. Pour formaliser des connaissances à ce sujet, nous avons étudié un panel de seize cartes multi-échelles existantes. Nous avons analysé les systèmes de zoom utilisés ainsi que l’évolution des représentations cartographiques au fil des échelles, en particulier les changements de niveaux d’abstraction pour les objets bâtis et routiers. Nous avons aussi évalué la variation de complexité visuelle du contenu cartographique au fil des échelles, en utilisant des mesures de clutter visuel. Nous avons ainsi identifié les tendances générales en termes de représentations multi-échelles (comme l’application du standard WMTS), certains facteurs que nous considérons comme ayant une influence négative sur la navigation de l’utilisateur (comme l’utilisation d’une même carte à différentes échelles), ainsi que des pratiques intéressantes visant à la faciliter (comme les représentations mixtes). A partir de ces constats, nous avons formulé des hypothèses sur l’influence des variables de construction des représentations intermédiaires sur la fluidité de navigation. Nous avons construit un matériel de test à partir d’un extrait de la carte multi-échelle Scan Express de l’IGN, entre les cartes existant au 1 : 25k et au 1 : 100k. Nous avons ainsi produit quatre versions différentes de représentations intermédiaires entre ces deux cartes, implémentant nos différentes hypothèses. Cet exercice nous a permis de mieux cerner les verrous techniques que soulève la production de représentations intermédiaires. Nous avons enfin conduit un test utilisateurs contrôlé, en demandant à 15 participants de réaliser une tâche cartographique sur ces différentes cartes multi-échelles, pour évaluer la pertinence de nos hypothèses / A multi-scale map is a set of maps at different scales, displayed on mapping applications, in which users may navigate by zooming in or out. Each of these maps is produced beforehand by cartographic generalization, which aims to adapt the cartographic representation for a target scale. Due to generalization, the representation changes between maps at different scales may disturb the user during its navigation. We assume that adding intermediate representations in an existing multi-scale map may enable a smooth evolution of cartographic content across scales. While theoretical knowledge exists for traditional cartography, we still do not know how to design efficient multi-scale maps. To formalize knowledge on that subject, we studied sixteen existing multi-scale maps. We focused on the used zooming system (zoom levels and display scales) and on the evolution of cartographic representations across scales, in particular for building and road entities. We also analyzed the variation of visual complexity of the map content across scales, using visual clutter measures. We thus identified general trends in terms of multi-scale representation (i.e. use of WMTS standard), some potential disturbing factors (i.e. use of a same map at different scales), but also good practices which may ease the user navigation (i.e. mixed representations). Based on these findings, we made assumptions on the influence of intermediate representations design on user navigation. We built test material from an extract of the Scan Express multi-scale map of the French IGN, between the existing maps at 1:25k and 1:100k scales. We thus produced four different versions of intermediate representations between these two maps, implementing our different hypotheses. This way, we highlighted the technical issues that we faced when producing intermediate representations. Finally, we conducted a controlled user study, asking 15 participants to perform a cartographic task on these different multi-scale maps, to evaluate our hypotheses

Title: Artificial Neural Networks and Their Usage For Knowledge Extraction Author: RNDr. Zuzana Petříčková Department: Department of Theoretical Computer Science and Mathema- tical Logic Supervisor: doc. RNDr. Iveta Mrázová, CSc., Department of Theoretical Computer Science and Mathematical Logic Abstract: The model of multi/layered feed/forward neural networks is well known for its ability to generalize well and to find complex non/linear dependencies in the data. On the other hand, it tends to create complex internal structures, especially for large data sets. Efficient solutions to demanding tasks currently dealt with require fast training, adequate generalization and a transparent and simple network structure. In this thesis, we propose a general framework for training of BP/networks. It is based on the fast and robust scaled conjugate gradient technique. This classical training algorithm is enhanced with analytical or approximative sensitivity inhibition during training and enforcement of a transparent in- ternal knowledge representation. Redundant hidden and input neurons are pruned based on internal representation and sensitivity analysis. The performance of the developed framework has been tested on various types of data with promising results. The framework provides a fast training algorithm,...

Chez le rongeur, lorsque l’expression d’une peur conditionnée au contexte s’appuie sur un souvenir ancien, des réponses de peur importantes sont souvent rapportées lors de l’exposition à un contexte différent de celui dans lequel les chocs électriques ont été administrés. Cette généralisation des réponses de peur serait liée à une réorganisation concomitante des réseaux sous-tendant l’expression de la peur, en particulier à un « désengagement » de l’hippocampe dans le rappel des informations contextuelles. Cependant, d’autres études suggèrent que des modifications dans les traitements amenant à l’expression de comportements défensifs, liés à la peur ou l’anxiété, puissent également se mettre en place avec le temps suite à un conditionnement. Ainsi, une incubation des réponses de peur, correspondant à une augmentation globale des niveaux de peur avec le temps, pourrait contribuer à la généralisation de ces réponses. Nos travaux, par une approche d’imagerie de gènes précoces immédiats ainsi que d’évaluation comportementale, ont visé à démêler les mécanismes impliqués dans la généralisation de la peur conditionnée au contexte accompagnant sa consolidation à long terme. Nos résultats suggèrent que la généralisation soit moins liée à une altération du souvenir du contexte qu’à des modifications portant sur le traitement et / ou l’expression des émotions. La nature de ces dernières modifications semble dépendre du statut prédictif initial du contexte (en avant- ou en arrière-plan). De manière cohérente avec l’observation d’une absence de dégradation de la trace du contexte avec le temps, le rôle de l’hippocampe dans le rappel de l’information contextuelle nous est apparu maintenu au cours du temps. Les réorganisations observées dans les réseaux sous-tendant l’expression des réponses de peur concerneraient plutôt des modifications liées au traitement associatif et / ou émotionnel des informations contextuelles. Dans leur ensemble, nos résultats soulèvent la nécessaire prise en compte du haut degré de complexité des traitements amenant à l’expression d’une peur conditionnée, pour qui souhaite évaluer la qualité de la représentation contextuelle sur la base de réponses de peur discriminantes. / At remote delays following contextual fear conditioning in rodents, generalization of fear responses is usually described, as fear responses are elicited by exposure to a context different from the one in which footshocks were delivered. This generalization has been proposed to rely on the degradation and/or transformation of the memory trace due to systemic consolidation. The latter corresponds to the time-dependent reorganization of structures implicated in contextual fear expression, from networks involving the hippocampus to mainly cortical networks. However, other studies suggest that changes in defensive behaviors’ expression tied to fear and anxiety might take place in the time period following a fear conditioning experience. Indeed, an incubation of fear responses, that is an overall increase in fear responses following contextual fear conditioning, has repeatedly been reported. Such changes in the processing of emotionally relevant information might represent an alternative explanation of a time-dependent generalization of fear responses. Using immediate early genes imaging and behavioral assessment, our studies aimed at disentangling the processes supporting fear generalization over time. Our results suggest that under some circumstances, changes in the fear responses’ specificity might less be due to the dynamics of a memory system supporting the context representation than to alterations attributable to emotional information processing and/or expression. They point out dissociations in the latter changes between animals conditioned to a foreground or a background context. Consistent with our observation of a preserved detailed context memory trace, they also suggest that changes in the brain networks supporting the expression of a remotely acquired contextual fear might not reflect a time-dependent hippocampal-independency. Rather, the observed reorganization of neuronal networks might sustain changes in the associative and / or emotional information processing evoked by context exposure. Altogether, our results point out the need to take account of the high complexity of information processing leading to the onset of fear responses, when trying to infer the quality of a contextual representation on the basis of fear discrimination between contexts.