This master thesis presents the process of designing and implementing a CNN-based architecture for image recognition included in a larger project in the field of fashion recommendation with deep learning. Concretely, the presented network aims to perform localization and segmentation tasks. Therefore, an accurate analysis of the most well-known localization and segmentation networks in the state of the art has been performed. Afterwards, a multi-task network performing RoI pixel-wise segmentation has been created. This proposal solves the detected weaknesses of the pre-existing networks in the field of application, i.e. fashion recommendation. These weaknesses are basically related with the lack of a fine-grained quality of the segmentation and problems with computational efficiency. When it comes to improve the details of the segmentation, this network proposes to work pixel- wise, i.e. performing a classification task for each of the pixels of the image. Thus, the network is more suitable to detect all the details presented in the analysed images. However, a pixel-wise task requires working in pixel resolution, which implies that the number of operations to perform is usually large. To reduce the total number of operations to perform in the network and increase the computational efficiency, this pixel-wise segmentation is only done in the meaningful regions of the image (Regions of Interest), which are also computed in the network (RoI masks). Then, after a study of the more recent deep learning libraries, the network has been successfully implemented. Finally, to prove the correct operation of the design, a set of experiments have been satisfactorily conducted. In this sense, it must be noted that the evaluation of the results obtained during testing phase with respect to the most well-known architectures is out of the scope of this thesis as the experimental conditions, especially in terms of dataset, have not been suitable for doing so. Nevertheless, the proposed network is totally prepared to perform this evaluation in the future, when the required experimental conditions are available. / Denna examensarbete presenterar processen för att designa och implementera en CNN-baserad arkitektur för bildigenkänning som ingår i ett större projekt inom moderekommendation med djup inlärning. Konkret, det presenterade nätverket syftar till att utföra lokaliseringsoch segmenteringsuppgifter. Därför har en noggrann analys av de mest kända lokaliseringsoch segmenteringsnätena utförts inom den senaste tekniken. Därefter har ett multi-task-nätverk som utför RoI pixel-wise segmentering skapats. Detta förslag löser de upptäckta svagheterna hos de befintliga näten inom tillämpningsområdet, dvs modeanbefaling. Dessa svagheter är i grund och botten relaterade till bristen på en finkornad kvalitet på segmenteringen och problem med beräkningseffektivitet. När det gäller att förbättra detaljerna i segmenteringen, föreslår detta nätverk att arbeta pixelvis, dvs att utföra en klassificeringsuppgift för var och en av bildpunkterna i bilden. Nätverket är sålunda lämpligare att detektera alla detaljer som presenteras i de analyserade bilderna. En pixelvis uppgift kräver dock att man arbetar med pixelupplösning, vilket innebär att antalet operationer som ska utföras är vanligtvis stor. För att minska det totala antalet operationer som ska utföras i nätverket och öka beräkningseffektiviteten görs denna pixelvisa segmentering endast i de meningsfulla regionerna i bilden (intressanta regioner), som också beräknas i nätverket (RoI-masker) . Sedan, efter en studie av de senaste djuplärningsbiblioteken, har nätverket framgångsrikt implementerats. Slutligen, för att bevisa korrekt funktion av konstruktionen, har en uppsättning experiment genomförts på ett tillfredsställande sätt. I detta avseende måste det noteras att utvärderingen av de resultat som uppnåtts under testfasen i förhållande till de mest kända arkitekturerna ligger utanför denna avhandling, eftersom de experimentella förhållandena, särskilt vad gäller dataset, inte har varit lämpliga För att göra det. Ändå är det föreslagna nätverket helt beredd att utföra denna utvärdering i framtiden när de nödvändiga försöksvillkoren är tillgängliga. / En aquest treball de fi de màster es presenta el disseny i la implementació d’una arquitectura pel reconeixement d’imatges fent ús de CNN. Aquesta xarxa es troba inclosa en un projecte de major envergadura en el camp de la recomanació de moda. En concret, la xarxa presentada en aquest document s’encarrega de realitzar les tasques de localització i segmentació. Després d’un estudi a consciència de les xarxes més conegudes de l’estat de l’art, s’ha dissenyat una xarxa multi-tasca encarregada de realitzar una segmentació a resolució de píxel de les regions d’interès de la imatge, les quals han sigut prèviament calculades i emmascarades. Aquesta proposta soluciona les mancances detectades en les xarxes ja existents pel que fa a la tasca de recomanació de moda. Aquestes mancances es basen en la obtenció d’una segmentació sense prou nivell de detalls i en una rellevant complexitat computacional. Pel que fa a la qualitat de la segmentació, aquesta tesi proposa treballar en resolució de píxel, classificant tots els píxels de la imatge de forma individual, per tal de poder adaptar-se a tots els detalls que puguin aparèixer a la imatge analitzada. No obstant, treballar píxel a píxel implica la realització d’una gran quantitat d’operacions. Per reduir-les, proposem fer la segmentació píxel a píxel només a les regions d’interès de la imatge. A continuació, després d’un estudi detallat de les llibreries de deep learnign més destacades, el disseny ha sigut implementat. Finalment s’han dut a terme una sèrie d’experiments per provar el correcte funcionament del disseny. En aquest sentit és important destacar que aquesta tesi no té com a objectiu avaluar el disseny respecte d’altres xarxes ja existents. La raó és que les condicions d’experimentació, sobretot pel que fa a la base de dades, no són adequades per aquesta tasca. No obstant, la xarxa està perfectament preparada per fer aquesta avaluació un cop les condicions d’experimentació així ho permetin.

This thesis explores bodily representations in Late Iron Age Scandinavia (400–1050 AD). Non-human bodies, such as gold foil figures, and human bodies are analysed. The work starts with an examination and deconstruction of the sex/gender categories to the effect that they are considered to be of minor value for the purposes of the thesis. Three analytical concepts – masks, miniature, and metaphor – are deployed in order to interpret how and why the chosen bodies worked within their prehistoric contexts. The manipulations the figures sometimes have undergone are referred to as masking practices, discussed in Part One. It is shown that masks work and are powerful by being paradoxical; that they are vehicles for communication; and that they are, in effect, transitional objects bridging gaps that arise in continuity as a result of events such as symbolic or actual deaths. In Part Two miniaturization is discussed. Miniaturization contributes to making worlds intelligible, negotiable and communicative. Bodies in miniatures in comparison to other miniature objects are particularly potent. Taking gold foil figures under special scrutiny, it is claimed that gold, its allusions as well as its inherent properties conveyed numinosity. Consequently gold foil figures, regardless of the context, must be understood as extremely forceful agents. Part Three examines metaphorical thinking and how human and animal body parts were used in pro-creational acts, resulting in the birth of persons. However, these need not have been human, but could have been the outcomes of turning a deceased into an ancestor, iron into a steel sword, or clay into a ceramic urn, hence expanding and transforming the members of the family/household. Thus, bone in certain contexts acted as a transitional object or as a generative substance. It is concluded that the bodies of research are connected to transitions, and that the theme of transformation was one fundamental characteristic of the societies of study.

Les opérations géométriques appliquées aux maillages 3D introduisent des dis torsions géométriques qui peuvent être visibles pour un observateur humain. Dans cette thèse, nous étudions l’impact perceptuel de ces distorsions. Plus précisément, notre objectif est de calculer le seuil à partir duquel les distorsions géométriques locales deviennent visibles. Afin d’atteindre notre but, nous définissons tout d’abord des caractéristiques perceptuelles pour les maillages 3D. Nous avons ensuite effectué une étude expérimentale des propriétés du système visuel humain (sensibilité au contraste et effet du masquage visuel) en observant un maillage 3D. Les résultats de ces expériences sont finalement utilisés pour proposer un algorithme qui calcule le seuil de visibilité relatif à une distorsion locale. L’algorithme proposé s’adapte aux différentes conditions d’affichage (résolution et taille de l’écran), d’illumination et au type de rendu. Enfin, nous montrons l’utilité d’un tel algorithme en intégrant le seuil de visibilité dans le pipeline de plusieurs opérations géométriques (ex: simplification, subdivision adaptative) / Geometric operations applied to a 3D mesh introduce geometric distortion in the form of vertex displacement that can be visible to a human observer. In this thesis, we have studied the perceptual impact of these geometric distortions. More precisely, our goal is to compute the threshold beyond which a local geometric distortion becomes visible. In order to reach this goal, we start by evaluating perceptually relevant properties on 3D meshes. We have then performed a series of psychophysical experiments in which we measured the visibility threshold relative to various properties of the Human Visual System (contrast sensitivity and visual masking). The results of these experiments allowed us to propose an algorithm that computes the visibility threshold relative to a local geometric distortion. This algorithm is capable of adapting to the different display condition of 3D meshes (resolution, display size, illumination condition and rendering). Finally, we showcase the utility of our work by integrating the developed perceptual method in several geometric operations such as mesh simplification and adaptive subdivision

Dans les sociétés industrialisées, les humains sont exposés à multiples sources de mouvements vibratoires. Jusqu'à présent, l’analyse de l'effet des stimuli vibratoire dans son ensemble a été bien développée, plus précisément en ce qui concerne le seuil de perception. Au cours des dernières années, diverses études ont analysé l'influence de certains facteurs.Cependant, l'effet d'une des composantes du stimulus sur la perception d'autres composants a été rarement rapporté. Dans ce cas, très peu d'information a été trouvée dans la littérature. Dans les deux études trouvées, le seuil absolu a été réévalué en considérant un second signal sinusoïdal. La valeur du seuil était plus élevée que ce qui a été estimé avec le stimulus simple. L'augmentation du seuil a été attribuée à l'effet de masquage. Basé sur ces observations, cette étude examine l'existence de l'effet de masquage dans les vibrations vertical du corps entière. Afin d'examiner l'effet de masquage, la différence entre le seuil absolu et le seuil masqué est nécessaire. Deux types de stimuli ont été utilisés : le stimulus masquage et le stimulus de test. Le masque est un signal de bruit à bande étroite entre 10 et 20 Hz, à trois niveaux d'amplitude différents. La sélection du stimulus de masque a été basée sur le modèle de masquage largement étudié dans certains domaines tels que l'acoustique, le domaine visuel et vibrotactile. Le stimulus de test est un signal sinusoïdale d'amplitude modulée à six fréquences différentes (30, 35, 40, 50, 60, 80 Hz. L'effet de la fréquence sur le seuil absolu estimé est similaire à ce qui a été indiqué par d'autres études. La détection du stimulus en présence du signal de masquage a été empêchée. Par conséquent, il est certainement clair que le phénomène de masquage a eu lieu. Les valeurs des seuils masqués étaient considérablement plus élevées que ceux correspondant aux seuils absolus. Le seuil décroît en fonction de la fréquence de test et l'effet de masque est plus important à haute qu’à faible niveau d'intensité du masque. Le démasquage est lié à l’écart entre les fréquences des signaux masquant et de test. Lorsque l’écart est faible, prévaut la reconnaissance de l’augmentation de l’intensité globale d'énergie. Si l’écart augmente, la discrimination des signaux prédomine. / In industrialized societies, humans are exposed to multiple sources of vibratory motions. To date, the analyses of the impact of vibratory stimuli as a whole have been well developed, more precisely with respect to perception threshold. Along recent years various studies have been analyzed the influence of some factors involved.However, the effect of one component of the stimulus in to perception of other components has been rarely reported. In this case, little information was found in the literature. In two studies found, the absolute threshold was revalued by integrating a second sinusoidal signal. The threshold value was higher that which has been estimated with the simple stimulus. The increase of threshold has been attributed to a masking effect. Based on these remarks, this study investigates the existence of masking effect for vertical whole body vibration, in particular to sitting position. To examine the masking effect, the difference between absolute and masked thresholds is required. Two types of stimuli have been used: the masking and the test stimulus. The masker is a narrow-band noise from 10 to 20 Hz at three different magnitude levels. The selection of mask stimulus was based on the masking model widely studied in acoustic, visual and vibrotactile areas. The test stimulus is a sinusoidal modulated amplitude signal at six different frequencies (30, 35, 40, 50, 60, 80 Hz). The frequency dependence of the absolute threshold estimated is similar to the results presented in other studies. The stimulus detection in the presence of the masking signal has been prevented. Therefore, it is certainly clear that the masking phenomenon occurred. The masked thresholds values were considerably higher than those absolute thresholds. The threshold decreases as a function of test frequency and the mask effect is more important at high than at low intensity masker levels. The unmasking is related to the difference between the mask and test signals frequencies. When the difference is small, the recognition of the increase in the overall energy intensity predominates. However, if the difference increases, then the discrimination of signals is dominant.

Bei einem Spaziergang im Mondlicht nehmen wir Ereignisse verschiedenster Intensität wahr. Vom blendenden Licht eines vorbeifahrenden Motorrads bis hin zu undeutlichen Schatten. Verarbeiten wir diese Ereignisse alle gleich? In dieser Arbeit untersuchte ich, wie das visuelle System auf Stimuli reagiert, die so niedrig in ihrer Intensität sind, dass wir sie nicht wahrnehmen. Frühere somatosensorische Studien zeigten eine kortikale Deaktivierung aufgrund subliminaler Stimulation. Diese wurde als Hemmungsmechanismus interpretiert zur Unterdrückung kortikalen Rauschens. Unterstützt wurde diese Aussage durch ein Verhaltensexperiment, in dem der Schwellwert für somatosensorische Stimuli bei subliminaler Stimulation erhöht war. Gibt es im visuellen System äquivalente Hemmungsmechanismen? In Studie I untersuchte ich die Wahrnehmung schwellnaher visueller Zielreize: die Wahrnehmung verschlechterte sich bei gleichzeitiger subliminaler Stimulation im selben Hemifield. Inhibitorische Interneurone könnten diesen Effekt hervorrufen. Gleichzeitig zeigten Nervenzell-Studien die Degeneration intrakortikaler Inhibition mit dem fortschreitendem Alter von Affen. In Studie II untersuchte ich daher Unterschiede inhibitorischer Mechanismen einer kleinen Gruppe von älteren Probanden und verglich diese mit den jüngeren Probanden aus Studie I. In Studie III, einer elektrophysiologischen Studie, führt subliminale Stimulation zu einer Verstärkung des Alpha-Rhythmus. Supraliminale Stimulation führt zu einer Verstärkung niedriger Frequenzen und einer Abschwächung des Alpha-Rhythmus. Die spezifische neuronale Signatur aufgrund subliminaler Stimulation deutet darauf hin, dass die neuronale Verarbeitung des Stimulus zu einer Verringerung der Aktivität in involvierten Arealen führt. Ein Rauschunterdrückungs-Mechanismus wurde im somatosensorischen System beschrieben und könnte für die verringerte Wahrnehmung der schwellnahen Zielreize bei subliminaler Stimulation verantwortlich sein. / Walking along a street on a moonlit night, we can perceive visual events at a wide range of intensities – from the blinding light of a passing motorcycle to faint shadows. Does the visual system react similarly to all of these events? Here, I investigated how the visual system reacts to stimuli that are so low in their intensity that they are not perceived. It has been shown that subliminal low-intensity somatosensory stimuli lead to a cortical deactivation or inhibition. This deactivation was interpreted as inhibition mechanism that usually protects the cortex against activation by noise. Also, a behavioural experiment showed an increased sensitivity threshold for peri-liminal stimuli during subliminal stimulation. Does a similar mechanism exist within the visual System? In Study I, I investigated the perception of visual peri-liminal target-stimuli under different conditions. The threshold for target-stimuli significantly increased when presented during subliminal stimulation on the same side as the target-stimulus. The underlying mechanism could be mediated through intracortical inhibition. Concurrently, studies in macaque senescent neurons suggest a degradation of intracortical inhibition with age. In Study II, I therefore investigated differences of inhibitory responses in a group of elderly subjects and compared the results to the young participants in Study I. In Study III, using electroencephalography, I show that subliminal stimulation leads to an alpha-band power increase, whereas supraliminal stimulation leads to a lower frequency increase and an alpha-band power decrease. Specific neural signature in response to subliminal stimulation indicate neural processing of the stimulus that lead to a down-regulation of areas involved in stimulus processing. This mechanism could serve a suppression of input noise that has been described in the somatosensory system and may lead to decreased detection of peri-liminal target-stimuli during subliminal stimulation.

Das visuelle System erreicht enorme Verarbeitungsmengen, wenn wir unsere Augen auf ein Objekt richten. Mehrere Prozesse sind aktiv bevor unser Blick das neue Objekt erreicht. Diese Arbeit erforscht die räumlichen und zeitlichen Eigenschaften drei solcher Prozesse: 1. aufmerksamkeitsbedingte Steigerung der neuronalen Aktivität und sakkadische Suppression; 2. aufmerksamkeitsbasierte Auswahl des Zielreizes bei einer visuellen Suchaufgabe; 3. zeitliche Entwicklung der Detektiongenauigkeit bei der Objekt-Substitutionsmaskierung. Wir untersuchten diese Prozesse mit einer Kombination aus humaner Elektroenzephalografie (EEG), eye tracking und psychophysischen Verhaltensmessungen. Zuerst untersuchten wir, wie die neuronale Repräsentation eines Reizes von seiner zeitlichen Nähe zur Sakkade geprägt wird. Wir zeigten, dass direkt vor der Sakkade erscheinende Reize am meisten durch Aufmerksamkeit und Suppression geprägt sind. In Studie 2 wurde die Sichtbarkeit des Reizes mit der Objekt-Substitutionsmaskierung verringert, und wir analysierten das Verhältnis zwischen sakkadischen Reaktionszeiten und ihrer Genauigkeit. Dazu erfassten wir neuronale Marker der Aufmerksamkeitslenkung zum Zielreiz und eine subjektive Bewertung seiner Wahrnehmbarkeit. Wir stellten fest, dass schnelle Sakkaden der Maskierung entgingen und Genauigkeit sowie subjektive Wahrnehmbarkeit erhöhten. Dies zeigt, dass bereits in frühen Verarbeitungsstadien eine bewusste und korrekte Wahrnehmung des Reizes entstehen kann. Wir replizierten diesen Befund für manuelle Antworten, um eine Verfälschung der Ergebnisse durch sakkadenspezifische Prozesse auszuschließen. Neben ihrer theoretischen Bedeutung liefern diese Studien einen methodischen Beitrag zum Forschungsgebiet der EEG-Augenbewegung: Entfernung sakkadischer Artefakte aus dem EEG bzw. Erstellung eines künstlichen Vergleichsdatensatzes. Die Arbeit stellt mehrere Ansätze zur Untersuchung der Dynamik visueller Wahrnehmung sowie Lösungen für zukünftige Studien dar. / The visual system achieves a tremendous amount of processing as soon as we set eyes on a new object. Numerous processes are active already before eyes reach the object. This thesis explores the spatio-temporal properties of three such processes: attentional enhancement and saccadic suppression that accompany saccades to target; attentional selection of target in a visual search task; the timecourse of target detection accuracy under object-substitution masking. We monitored these events using a combination of human electrophysiology (EEG), eye tracking and behavioral psychophysics. We first studied how the neural representation of a visual stimulus is affected by its temporal proximity to saccade onset. We show that stimuli immediately preceding a saccade show strongest effects of attentional enhancement and saccadic suppression. Second, using object-substitution masking to reduce visibility, we analyzed the relationship between saccadic reaction times and response accuracy. We also collected subjective visibility ratings and observed neural markers of attentional selection, such as the negative, posterior-contralateral deflection at 200 ms (N2pc). We found that fast saccades escaped the effects of masking, resulted in higher response accuracy and higher awareness ratings. This indicates that early visual processing can trigger awareness and correct behavior. Finally, we replicated this finding with manual responses. Discovering a similar accuracy timecourse in a different modality ruled out saccade-specific mechanisms, such as saccadic suppression and retinal shift, as a potential confound. Next to their theoretical impact, all studies make a methodological contribution to EEG-eye movement research, such as removal of large-scale saccadic artifacts from EEG data and composition of matched surrogate data. In sum, this work uses multiple approaches to describe the dynamics of visual perisaccadic perception and offers solutions for future studies in this field.

<p>This thesis explores bodily representations in Late Iron Age Scandinavia (400–1050 AD). Non-human bodies, such as gold foil figures, and human bodies are analysed. The work starts with an examination and deconstruction of the sex/gender categories to the effect that they are considered to be of minor value for the purposes of the thesis. Three analytical concepts – masks, miniature, and metaphor – are deployed in order to interpret how and why the chosen bodies worked within their prehistoric contexts.</p><p>The manipulations the figures sometimes have undergone are referred to as masking practices, discussed in Part One. It is shown that masks work and are powerful by being paradoxical; that they are vehicles for communication; and that they are, in effect, transitional objects bridging gaps that arise in continuity as a result of events such as symbolic or actual deaths.</p><p>In Part Two miniaturization is discussed. Miniaturization contributes to making worlds intelligible, negotiable and communicative. Bodies in miniatures in comparison to other miniature objects are particularly potent. Taking gold foil figures under special scrutiny, it is claimed that gold, its allusions as well as its inherent properties conveyed numinosity. Consequently gold foil figures, regardless of the context, must be understood as extremely forceful agents.</p><p>Part Three examines metaphorical thinking and how human and animal body parts were used in pro-creational acts, resulting in the birth of persons. However, these need not have been human, but could have been the outcomes of turning a deceased into an ancestor, iron into a steel sword, or clay into a ceramic urn, hence expanding and transforming the members of the family/household. Thus, bone in certain contexts acted as a transitional object or as a generative substance.</p><p>It is concluded that the bodies of research are connected to transitions, and that the theme of transformation was one fundamental characteristic of the societies of study.</p>

S'ha investigat molt tant els mecanismes neuronals bàsics de l'audició com l'organització psicològica de la percepció de la parla. Tanmateix, en ambdós temes n'hi ha una relativa escassetat en quant a modelització. Aquí describim dos treballs de modelització. Un d'ells proposa un nou mecanisme de millora de selectivitat de freqüències que explica resultats de experiments neurofisiològics investigant manifestacions de forward masking y sobretot auditory streaming en l'escorça auditiva principal (A1). El mecanisme funciona en una xarxa feed-forward amb depressió sináptica entre el tàlem y l'escorça, però mostrem que és robust a l'introducció d'una organització realista del circuit de A1, que per la seva banda explica cantitat de dades neurofisiològics. L'altre treball descriu un mecanisme candidat d'explicar la trobada en estudis psicofísics de diferències en la percepció de paraules entre bilinguës primerencs y simultànis. Simulant tasques de decisió lèxica y discriminació de fonemes, fortifiquem l'hipòtesi de que persones sovint exposades a variacions dialectals de paraules poden guardar aquestes en el seu lèxic, sense alterar representacions fonemàtiques . / Though much experimental research exists on both basic neural mechanisms of hearing and the psychological organization of language perception, there is a relative paucity of modelling work on these subjects. Here we describe two modelling efforts. One proposes a novel mechanism of frequency selectivity improvement that accounts for results of neurophysiological experiments investigating manifestations of forward masking and above all auditory streaming in the primary auditory cortex (A1). The mechanism works in a feed-forward network with depressing thalamocortical synapses, but is further showed to be robust to a realistic organization of the neural circuitry in A1, which accounts for a wealth of neurophysiological data. The other effort describes a candidate mechanism for explaining differences in word/non-word perception between early and simultaneous bilinguals found in psychophysical studies. By simulating lexical decision and phoneme discrimination tasks in an attractor neural network model, we strengthen the hypothesis that people often exposed to dialectal word variations can store these in their lexicons, without altering their phoneme representations. / Se ha investigado mucho tanto los mecanismos neuronales básicos de la audición como la organización psicológica de la percepción del habla. Sin embargo, en ambos temas hay una relativa escasez en cuanto a modelización. Aquí describimos dos trabajos de modelización. Uno propone un nuevo mecanismo de mejora de selectividad de frecuencias que explica resultados de experimentos neurofisiológicos investigando manifestaciones de forward masking y sobre todo auditory streaming en la corteza auditiva principal (A1). El mecanismo funciona en una red feed-forward con depresión sináptica entre el tálamo y la corteza, pero mostramos que es robusto a la introducción de una organización realista del circuito de A1, que a su vez explica cantidad de datos neurofisiológicos. El otro trabajo describe un mecanismo candidato de explicar el hallazgo en estudios psicofísicos de diferencias en la percepción de palabras entre bilinguës tempranos y simultáneos. Simulando tareas de decisión léxica y discriminación de fonemas, fortalecemos la hipótesis de que personas expuestas a menudo a variaciones dialectales de palabras pueden guardar éstas en su léxico, sin alterar representaciones fonémicas.

Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal

There are large gaps in our understanding how fish populations are affected by the anthropogenic noise and the alteration of habitat caused by the construction and operation of offshore wind farms. These issues are of great importance as the construction of offshore wind farms will increase all over the world in the near future. This thesis studies these effects with a focus on fish. The wind turbine foundations function as artificial reefs and are colonized by invertebrates, algae and fish. The epibenthic assemblages are influenced by factors such as hydrographical parameters, time of submergence, distance to natural hard bottom, material and texture (PAPER I, II). Once an epibenthic assemblage has been developed, fish utilize it for different ecosystem services such as food, shelter, and spawning and nursery area. Benthic and semi-pelagic species show a stronger response to the introduced foundation than pelagic species, as it is the bottom habitat that has mainly been altered (PAPER I, II). Pelagic species could be positively affected by the increased food availability - but it takes time and the effect is local. Construction noise like pile driving creates high levels of sound pressure and acoustic particle motion in the water and seabed. This noise induces behavioural reactions in cod (Gadus morhua) and sole (Solea solea). These reactions could occur up to tens of kilometres distance from the source (PAPER III). During power production, the wind turbines generate a broadband noise with a few dominating tones (PAPER IV, V), which are detectable by sound pressure sensitive fish at a distance of several kilometres even though intense shipping occurs in the area. Motion sensitive species will only detect the turbine noise at around a ten meter distance. Sound levels are only high enough to possibly cause a behavioural reaction within meters from a turbine (PAPER IV, V). / At the time of the doctoral defense, the following papers were unpublished and had a status as follows: Paper 3, 4 and 5: Manuscripts.