Graph-based Vision-Language Model for Multi-Modal Retrieval of Fashion Products

Artificial Intelligence is reshaping the field of fashion industry in different ways. E-commerce retailers exploit their data through AI to enhance their search engines, make outfit suggestions and forecast the success of a specific fashion product. However, it is a challenging endeavour as the data they possess is huge, complex and multi-modal. The most common way to search for fashion products online is by matching keywords with phrases in the product's description which are often cluttered, inadequate and differ across collections and sellers. A customer may also browse an online store's taxonomy, although this is time-consuming and doesn't guarantee relevant items. With the advent of Deep Learning architectures, particularly Vision-Language models, ad-hoc solutions have been proposed to model both the product image and description to solve this problems. However, the suggested solutions do not exploit effectively the semantic or syntactic information of these modalities, and the unique qualities and relations of clothing items. In this work of thesis, a novel approach is proposed to address this issues, which aims to model and process images and text descriptions as graphs in order to exploit the relations inside and between each modality and employs specific techniques to extract syntactic and semantic information. The results obtained show promising performances on different tasks when compared to the present state-of-the-art deep learning architectures.

Adaptive Brain-Computer Interface Systems For Communication in People with Severe Neuromuscular Disabilities

Brain-computer interfaces (BCI) have the potential to restore communication or control abilities in individuals with severe neuromuscular limitations, such as those with amyotrophic lateral sclerosis (ALS). The role of a BCI is to extract and decode relevant information that conveys a user's intent directly from brain electro-physiological signals and translate this information into executable commands to control external devices. However, the BCI decision-making process is error-prone due to noisy electro-physiological data, representing the classic problem of efficiently transmitting and receiving information via a noisy communication channel.

This research focuses on P300-based BCIs which rely predominantly on event-related potentials (ERP) that are elicited as a function of a user's uncertainty regarding stimulus events, in either an acoustic or a visual oddball recognition task. The P300-based BCI system enables users to communicate messages from a set of choices by selecting a target character or icon that conveys a desired intent or action. P300-based BCIs have been widely researched as a communication alternative, especially in individuals with ALS who represent a target BCI user population. For the P300-based BCI, repeated data measurements are required to enhance the low signal-to-noise ratio of the elicited ERPs embedded in electroencephalography (EEG) data, in order to improve the accuracy of the target character estimation process. As a result, BCIs have relatively slower speeds when compared to other commercial assistive communication devices, and this limits BCI adoption by their target user population. The goal of this research is to develop algorithms that take into account the physical limitations of the target BCI population to improve the efficiency of ERP-based spellers for real-world communication.

In this work, it is hypothesised that building adaptive capabilities into the BCI framework can potentially give the BCI system the flexibility to improve performance by adjusting system parameters in response to changing user inputs. The research in this work addresses three potential areas for improvement within the P300 speller framework: information optimisation, target character estimation and error correction. The visual interface and its operation control the method by which the ERPs are elicited through the presentation of stimulus events. The parameters of the stimulus presentation paradigm can be modified to modulate and enhance the elicited ERPs. A new stimulus presentation paradigm is developed in order to maximise the information content that is presented to the user by tuning stimulus paradigm parameters to positively affect performance. Internally, the BCI system determines the amount of data to collect and the method by which these data are processed to estimate the user's target character. Algorithms that exploit language information are developed to enhance the target character estimation process and to correct erroneous BCI selections. In addition, a new model-based method to predict BCI performance is developed, an approach which is independent of stimulus presentation paradigm and accounts for dynamic data collection. The studies presented in this work provide evidence that the proposed methods for incorporating adaptive strategies in the three areas have the potential to significantly improve BCI communication rates, and the proposed method for predicting BCI performance provides a reliable means to pre-assess BCI performance without extensive online testing.

User Behavioral Modeling of Web-based Systems for Continuous User Authentication

Authentication plays an important role in how we interact with computers, mobile devices, the web, etc. The idea of authentication is to uniquely identify a user before granting access to system privileges. For example, in recent years more corporate information and applications have been accessible via the Internet and Intranet. Many employees are working from remote locations and need access to secure corporate files. During this time, it is possible for malicious or unauthorized users to gain access to the system. For this reason, it is logical to have some mechanism in place to detect whether the logged-in user is the same user in control of the user's session. Therefore, highly secure authentication methods must be used. We posit that each of us is unique in our use of computer systems. It is this uniqueness that is leveraged to "continuously authenticate users" while they use web software. To monitor user behavior, n-gram models are used to capture user interactions with web-based software. This statistical language model essentially captures sequences and sub-sequences of user actions, their orderings, and temporal relationships that make them unique by providing a model of how each user typically behaves. Users are then continuously monitored during software operations. Large deviations from "normal behavior" can possibly indicate malicious or unintended behavior. This approach is implemented in a system called Intruder Detector (ID) that models user actions as embodied in web logs generated in response to a user's actions. User identification through web logs is cost-effective and non-intrusive. We perform experiments on a large fielded system with web logs of approximately 4000 users. For these experiments, we use two classification techniques; binary and multi-class classification. We evaluate model-specific differences of user behavior based on coarse-grain (i.e., role) and fine-grain (i.e., individual) analysis. A specific set of metrics are used to provide valuable insight into how each model performs. Intruder Detector achieves accurate results when identifying legitimate users and user types. This tool is also able to detect outliers in role-based user behavior with optimal performance. In addition to web applications, this continuous monitoring technique can be used with other user-based systems such as mobile devices and the analysis of network traffic.

Techniques to Improve the word alignments in Statistical Machine Translation from English to Malayalam

In Statistical Machine Translation from English to Malayalam, an unseen English sentence is translated into its equivalent Malayalam translation using statistical models like translation model, language model and a decoder. A parallel corpus of English-Malayalam is used in the training phase. Word to word alignments has to be set up among the sentence pairs of the source and target language before subjecting them for training. This paper is deals with the techniques which can be adopted for improving the alignment model of SMT. Incorporating the parts of speech information into the bilingual corpus has eliminated many of the insignificant alignments. Also identifying the name entities and cognates present in the sentence pairs has proved to be advantageous while setting up the alignments. Moreover, reduction of the unwanted alignments has brought in better training results. Experiments conducted on a sample corpus have generated reasonably good Malayalam translations and the results are verified with F measure, BLEU and WER evaluation metrics

Statistical model-based segmentation of the proximal femur in digital antero-posterior (AP) pelvic radiographs.

PURPOSE    Segmentation of the proximal femur in digital antero-posterior (AP) pelvic radiographs is required to create a three-dimensional model of the hip joint for use in planning and treatment. However, manually extracting the femoral contour is tedious and prone to subjective bias, while automatic segmentation must accommodate poor image quality, anatomical structure overlap, and femur deformity. A new method was developed for femur segmentation in AP pelvic radiographs. METHODS    Using manual annotations on 100 AP pelvic radiographs, a statistical shape model (SSM) and a statistical appearance model (SAM) of the femur contour were constructed. The SSM and SAM were used to segment new AP pelvic radiographs with a three-stage approach. At initialization, the mean SSM model is coarsely registered to the femur in the AP radiograph through a scaled rigid registration. Mahalanobis distance defined on the SAM is employed as the search criteria for each annotated suggested landmark location. Dynamic programming was used to eliminate ambiguities. After all landmarks are assigned, a regularized non-rigid registration method deforms the current mean shape of SSM to produce a new segmentation of proximal femur. The second and third stages are iteratively executed to convergence. RESULTS    A set of 100 clinical AP pelvic radiographs (not used for training) were evaluated. The mean segmentation error was [Formula: see text], requiring [Formula: see text] s per case when implemented with Matlab. The influence of the initialization on segmentation results was tested by six clinicians, demonstrating no significance difference. CONCLUSIONS    A fast, robust and accurate method for femur segmentation in digital AP pelvic radiographs was developed by combining SSM and SAM with dynamic programming. This method can be extended to segmentation of other bony structures such as the pelvis.

Phrase-based statistical machine translation: explanation of its processes and statistical models and evaluation of the English to Spanish translations produced

Statistical machine translation (SMT) is an approach to Machine Translation (MT) that uses statistical models whose parameter estimation is based on the analysis of existing human translations (contained in bilingual corpora). From a translation student’s standpoint, this dissertation aims to explain how a phrase-based SMT system works, to determine the role of the statistical models it uses in the translation process and to assess the quality of the translations provided that system is trained with in-domain goodquality corpora. To that end, a phrase-based SMT system based on Moses has been trained and subsequently used for the English to Spanish translation of two texts related in topic to the training data. Finally, the quality of this output texts produced by the system has been assessed through a quantitative evaluation carried out with three different automatic evaluation measures and a qualitative evaluation based on the Multidimensional Quality Metrics (MQM).

Child programming: an adequate domain specific language for programming specific robots

Dissertação para obtenção do Grau de Mestre em Engenharia Informática

Statistical Modeling of the Eye for Multimodal Treatment Planning for External Beam Radiation Therapy of Intraocular Tumors.

PURPOSE: Ocular anatomy and radiation-associated toxicities provide unique challenges for external beam radiation therapy. For treatment planning, precise modeling of organs at risk and tumor volume are crucial. Development of a precise eye model and automatic adaptation of this model to patients' anatomy remain problematic because of organ shape variability. This work introduces the application of a 3-dimensional (3D) statistical shape model as a novel method for precise eye modeling for external beam radiation therapy of intraocular tumors. METHODS AND MATERIALS: Manual and automatic segmentations were compared for 17 patients, based on head computed tomography (CT) volume scans. A 3D statistical shape model of the cornea, lens, and sclera as well as of the optic disc position was developed. Furthermore, an active shape model was built to enable automatic fitting of the eye model to CT slice stacks. Cross-validation was performed based on leave-one-out tests for all training shapes by measuring dice coefficients and mean segmentation errors between automatic segmentation and manual segmentation by an expert. RESULTS: Cross-validation revealed a dice similarity of 95% ± 2% for the sclera and cornea and 91% ± 2% for the lens. Overall, mean segmentation error was found to be 0.3 ± 0.1 mm. Average segmentation time was 14 ± 2 s on a standard personal computer. CONCLUSIONS: Our results show that the solution presented outperforms state-of-the-art methods in terms of accuracy, reliability, and robustness. Moreover, the eye model shape as well as its variability is learned from a training set rather than by making shape assumptions (eg, as with the spherical or elliptical model). Therefore, the model appears to be capable of modeling nonspherically and nonelliptically shaped eyes.

Statistical Analysis of Spectral Images

Tämä diplomityö liittyy Spektrikuvien tutkimiseen tilastollisen kuvamallin näkökulmasta. Diplomityön ensimmäisessä osassa tarkastellaan tilastollisten parametrien jakaumien vaikutusta väreihin ja korostumiin erilaisissa valaistusolosuhteissa. Havaittiin, että tilastollisten parametrien väliset suhteet eivät riipu valaistusolosuhteista, mutta riippuvat kuvan häiriöttömyydestä. Ilmeni myös, että korkea huipukkuus saattaa aiheutua värikylläisyydestä. Lisäksi työssä kehitettiin tilastolliseen spektrimalliin perustuvaa tekstuurinyhdistämisalgoritmia. Sillä saavutettiin hyviä tuloksia, kun tilastollisten parametrien väliset riippuvuussuhteet olivat voimassa. Työn toisessa osassa erilaisia spektrikuvia tutkittiin käyttäen itsenäistä komponenttien analyysia (ICA). Seuraavia itsenäiseen komponenttien analyysiin tarkoitettuja algoritmia tarkasteltiin: JADE, kiinteän pisteen ICA ja momenttikeskeinen ICA. Tutkimuksissa painotettiin erottelun laatua. Paras erottelu saavutettiin JADE- algoritmilla, joskin erot muiden algoritmien välillä eivät olleet merkittäviä. Algoritmi jakoi kuvan kahteen itsenäiseen, joko korostuneeseen ja korostumattomaan tai kromaattiseen ja akromaattiseen, komponenttiin. Lopuksi pohditaan huipukkuuden suhdetta kuvan ominaisuuksiin, kuten korostuneisuuteen ja värikylläisyyteen. Työn viimeisessä osassa ehdotetaan mahdollisia jatkotutkimuskohteita.

Statistical Methods for Conservation and Alignment Quality in Proteins

Construction of multiple sequence alignments is a fundamental task in Bioinformatics. Multiple sequence alignments are used as a prerequisite in many Bioinformatics methods, and subsequently the quality of such methods can be critically dependent on the quality of the alignment. However, automatic construction of a multiple sequence alignment for a set of remotely related sequences does not always provide biologically relevant alignments.Therefore, there is a need for an objective approach for evaluating the quality of automatically aligned sequences. The profile hidden Markov model is a powerful approach in comparative genomics. In the profile hidden Markov model, the symbol probabilities are estimated at each conserved alignment position. This can increase the dimension of parameter space and cause an overfitting problem. These two research problems are both related to conservation. We have developed statistical measures for quantifying the conservation of multiple sequence alignments. Two types of methods are considered, those identifying conserved residues in an alignment position, and those calculating positional conservation scores. The positional conservation score was exploited in a statistical prediction model for assessing the quality of multiple sequence alignments. The residue conservation score was used as part of the emission probability estimation method proposed for profile hidden Markov models. The results of the predicted alignment quality score highly correlated with the correct alignment quality scores, indicating that our method is reliable for assessing the quality of any multiple sequence alignment. The comparison of the emission probability estimation method with the maximum likelihood method showed that the number of estimated parameters in the model was dramatically decreased, while the same level of accuracy was maintained. To conclude, we have shown that conservation can be successfully used in the statistical model for alignment quality assessment and in the estimation of emission probabilities in the profile hidden Markov models.

Quantum tunneling fragmentation model

A nonthermal quantum mechanical statistical fragmentation model based on tunneling of particles through potential barriers is studied in compact two- and three-dimensional systems. It is shown that this fragmentation dynamics gives origin to several static and dynamic scaling relations. The critical exponents are found and compared with those obtained in classical statistical models of fragmentation of general interest, in particular with thermal fragmentation involving classical processes over potential barriers. Besides its general theoretical interest, the fragmentation dynamics discussed here is complementary to classical fragmentation dynamics of interest in chemical kinetics and can be useful in the study of a number of other dynamic processes such as nuclear fragmentation. ©2000 The American Physical Society.

Statistical quark models for the nucleon structure functions

We consider some existing relativistic models for the nucleon structure functions, relying on statistical approaches instead of perturbative ones. These models are based on the Fermi-Dirac distribution for the confined quarks, where a density of energy levels is obtained from an effective confining potential. In this context, it is presented some results obtained with a recent statistical quark model for the sea-quark asymmetry in the nucleon. It is shown, within this model, that experimental available observables, such as the ratio and difference between proton and neutron structure functions, are quite well reproduced with just three parameters: two chemical potentials used to reproduce the valence up and down quark numbers in the nucleon, and a temperature that is being used to reproduce the Gottfried sum rule violation. © 2010 American Institute of Physics.

Metodologia para o planejamento da operação de sistemas hidrotérmicos em mercado elétrico competitivo

Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)

The assessment of change in rural landscape and the analysis of the driving forces. Proposal of a spatial model for the rural built environment

The presented study carried out an analysis on rural landscape changes. In particular the study focuses on the understanding of driving forces acting on the rural built environment using a statistical spatial model implemented through GIS techniques. It is well known that the study of landscape changes is essential for a conscious decision making in land planning. From a bibliography review results a general lack of studies dealing with the modeling of rural built environment and hence a theoretical modelling approach for such purpose is needed. The advancement in technology and modernity in building construction and agriculture have gradually changed the rural built environment. In addition, the phenomenon of urbanization of a determined the construction of new volumes that occurred beside abandoned or derelict rural buildings. Consequently there are two types of transformation dynamics affecting mainly the rural built environment that can be observed: the conversion of rural buildings and the increasing of building numbers. It is the specific aim of the presented study to propose a methodology for the development of a spatial model that allows the identification of driving forces that acted on the behaviours of the building allocation. In fact one of the most concerning dynamic nowadays is related to an irrational expansion of buildings sprawl across landscape. The proposed methodology is composed by some conceptual steps that cover different aspects related to the development of a spatial model: the selection of a response variable that better describe the phenomenon under study, the identification of possible driving forces, the sampling methodology concerning the collection of data, the most suitable algorithm to be adopted in relation to statistical theory and method used, the calibration process and evaluation of the model. A different combination of factors in various parts of the territory generated favourable or less favourable conditions for the building allocation and the existence of buildings represents the evidence of such optimum. Conversely the absence of buildings expresses a combination of agents which is not suitable for building allocation. Presence or absence of buildings can be adopted as indicators of such driving conditions, since they represent the expression of the action of driving forces in the land suitability sorting process. The existence of correlation between site selection and hypothetical driving forces, evaluated by means of modeling techniques, provides an evidence of which driving forces are involved in the allocation dynamic and an insight on their level of influence into the process. GIS software by means of spatial analysis tools allows to associate the concept of presence and absence with point futures generating a point process. Presence or absence of buildings at some site locations represent the expression of these driving factors interaction. In case of presences, points represent locations of real existing buildings, conversely absences represent locations were buildings are not existent and so they are generated by a stochastic mechanism. Possible driving forces are selected and the existence of a causal relationship with building allocations is assessed through a spatial model. The adoption of empirical statistical models provides a mechanism for the explanatory variable analysis and for the identification of key driving variables behind the site selection process for new building allocation. The model developed by following the methodology is applied to a case study to test the validity of the methodology. In particular the study area for the testing of the methodology is represented by the New District of Imola characterized by a prevailing agricultural production vocation and were transformation dynamic intensively occurred. The development of the model involved the identification of predictive variables (related to geomorphologic, socio-economic, structural and infrastructural systems of landscape) capable of representing the driving forces responsible for landscape changes.. The calibration of the model is carried out referring to spatial data regarding the periurban and rural area of the study area within the 1975-2005 time period by means of Generalised linear model. The resulting output from the model fit is continuous grid surface where cells assume values ranged from 0 to 1 of probability of building occurrences along the rural and periurban area of the study area. Hence the response variable assesses the changes in the rural built environment occurred in such time interval and is correlated to the selected explanatory variables by means of a generalized linear model using logistic regression. Comparing the probability map obtained from the model to the actual rural building distribution in 2005, the interpretation capability of the model can be evaluated. The proposed model can be also applied to the interpretation of trends which occurred in other study areas, and also referring to different time intervals, depending on the availability of data. The use of suitable data in terms of time, information, and spatial resolution and the costs related to data acquisition, pre-processing, and survey are among the most critical aspects of model implementation. Future in-depth studies can focus on using the proposed model to predict short/medium-range future scenarios for the rural built environment distribution in the study area. In order to predict future scenarios it is necessary to assume that the driving forces do not change and that their levels of influence within the model are not far from those assessed for the time interval used for the calibration.

Single fluoroscopic image based 2D/3D reconstruction of a scaled, patient-specific vertebral model

Purpose Accurate three-dimensional (3D) models of lumbar vertebrae can enable image-based 3D kinematic analysis. The common approach to derive 3D models is by direct segmentation of CT or MRI datasets. However, these have the disadvantages that they are expensive, timeconsuming and/or induce high-radiation doses to the patient. In this study, we present a technique to automatically reconstruct a scaled 3D lumbar vertebral model from a single two-dimensional (2D) lateral fluoroscopic image. Methods Our technique is based on a hybrid 2D/3D deformable registration strategy combining a landmark-to-ray registration with a statistical shape model-based 2D/3D reconstruction scheme. Fig. 1 shows different stages of the reconstruction process. Four cadaveric lumbar spine segments (total twelve lumbar vertebrae) were used to validate the technique. To evaluate the reconstruction accuracy, the surface models reconstructed from the lateral fluoroscopic images were compared to the associated ground truth data derived from a 3D CT-scan reconstruction technique. For each case, a surface-based matching was first used to recover the scale and the rigid transformation between the reconstructed surface model Results Our technique could successfully reconstruct 3D surface models of all twelve vertebrae. After recovering the scale and the rigid transformation between the reconstructed surface models and the ground truth models, the average error of the 2D/3D surface model reconstruction over the twelve lumbar vertebrae was found to be 1.0 mm. The errors of reconstructing surface models of all twelve vertebrae are shown in Fig. 2. It was found that the mean errors of the reconstructed surface models in comparison to their associated ground truths after iterative scaled rigid registrations ranged from 0.7 mm to 1.3 mm and the rootmean squared (RMS) errors ranged from 1.0 mm to 1.7 mm. The average mean reconstruction error was found to be 1.0 mm. Conclusion An accurate, scaled 3D reconstruction of the lumbar vertebra can be obtained from a single lateral fluoroscopic image using a statistical shape model based 2D/3D reconstruction technique. Future work will focus on applying the reconstructed model for 3D kinematic analysis of lumbar vertebrae, an extension of our previously-reported imagebased kinematic analysis. The developed method also has potential applications in surgical planning and navigation.