Generating Discriminative Objective Proposals via Submodular Ranking

Object recognition has long been a core problem in computer vision. To improve object spatial support and speed up object localization for object recognition, generating high-quality category-independent object proposals as the input for object recognition system has drawn attention recently. Given an image, we generate a limited number of high-quality and category-independent object proposals in advance and used as inputs for many computer vision tasks. We present an efficient dictionary-based model for image classification task. We further extend the work to a discriminative dictionary learning method for tensor sparse coding. In the first part, a multi-scale greedy-based object proposal generation approach is presented. Based on the multi-scale nature of objects in images, our approach is built on top of a hierarchical segmentation. We first identify the representative and diverse exemplar clusters within each scale. Object proposals are obtained by selecting a subset from the multi-scale segment pool via maximizing a submodular objective function, which consists of a weighted coverage term, a single-scale diversity term and a multi-scale reward term. The weighted coverage term forces the selected set of object proposals to be representative and compact; the single-scale diversity term encourages choosing segments from different exemplar clusters so that they will cover as many object patterns as possible; the multi-scale reward term encourages the selected proposals to be discriminative and selected from multiple layers generated by the hierarchical image segmentation. The experimental results on the Berkeley Segmentation Dataset and PASCAL VOC2012 segmentation dataset demonstrate the accuracy and efficiency of our object proposal model. Additionally, we validate our object proposals in simultaneous segmentation and detection and outperform the state-of-art performance. To classify the object in the image, we design a discriminative, structural low-rank framework for image classification. We use a supervised learning method to construct a discriminative and reconstructive dictionary. By introducing an ideal regularization term, we perform low-rank matrix recovery for contaminated training data from all categories simultaneously without losing structural information. A discriminative low-rank representation for images with respect to the constructed dictionary is obtained. With semantic structure information and strong identification capability, this representation is good for classification tasks even using a simple linear multi-classifier.

A supervised machine learning method for word sense disambiguation of Portuguese nouns

Dissertação de Mestrado, Ciências da Linguagem, Faculdade de Ciências Humanas e Sociais, Universidade do Algarve, 2010

SARASOM: a supervised architecture based on the recurrent associative SOM

We present and evaluate a novel supervised recurrent neural network architecture, the SARASOM, based on the associative self-organizing map. The performance of the SARASOM is evaluated and compared with the Elman network as well as with a hidden Markov model (HMM) in a number of prediction tasks using sequences of letters, including some experiments with a reduced lexicon of 15 words. The results were very encouraging with the SARASOM learning better and performing with better accuracy than both the Elman network and the HMM.

An investigation of imputation methods for discrete databases and multi-variate time series

When it comes to information sets in real life, often pieces of the whole set may not be available. This problem can find its origin in various reasons, describing therefore different patterns. In the literature, this problem is known as Missing Data. This issue can be fixed in various ways, from not taking into consideration incomplete observations, to guessing what those values originally were, or just ignoring the fact that some values are missing. The methods used to estimate missing data are called Imputation Methods. The work presented in this thesis has two main goals. The first one is to determine whether any kind of interactions exists between Missing Data, Imputation Methods and Supervised Classification algorithms, when they are applied together. For this first problem we consider a scenario in which the databases used are discrete, understanding discrete as that it is assumed that there is no relation between observations. These datasets underwent processes involving different combina- tions of the three components mentioned. The outcome showed that the missing data pattern strongly influences the outcome produced by a classifier. Also, in some of the cases, the complex imputation techniques investigated in the thesis were able to obtain better results than simple ones. The second goal of this work is to propose a new imputation strategy, but this time we constrain the specifications of the previous problem to a special kind of datasets, the multivariate Time Series. We designed new imputation techniques for this particular domain, and combined them with some of the contrasted strategies tested in the pre- vious chapter of this thesis. The time series also were subjected to processes involving missing data and imputation to finally propose an overall better imputation method. In the final chapter of this work, a real-world example is presented, describing a wa- ter quality prediction problem. The databases that characterized this problem had their own original latent values, which provides a real-world benchmark to test the algorithms developed in this thesis.

The sensitivity of mapping methods to reference data quality:training supervised image classifications with imperfect reference data

The accuracy of a map is dependent on the reference dataset used in its construction. Classification analyses used in thematic mapping can, for example, be sensitive to a range of sampling and data quality concerns. With particular focus on the latter, the effects of reference data quality on land cover classifications from airborne thematic mapper data are explored. Variations in sampling intensity and effort are highlighted in a dataset that is widely used in mapping and modelling studies; these may need accounting for in analyses. The quality of the labelling in the reference dataset was also a key variable influencing mapping accuracy. Accuracy varied with the amount and nature of mislabelled training cases with the nature of the effects varying between classifiers. The largest impacts on accuracy occurred when mislabelling involved confusion between similar classes. Accuracy was also typically negatively related to the magnitude of mislabelled cases and the support vector machine (SVM), which has been claimed to be relatively insensitive to training data error, was the most sensitive of the set of classifiers investigated, with overall classification accuracy declining by 8% (significant at 95% level of confidence) with the use of a training set containing 20% mislabelled cases.

An automatic classifier of emotions built from entropy of noise

The electrocardiogram (ECG) signal has been widely used to study the physiological substrates of emotion. However, searching for better filtering techniques in order to obtain a signal with better quality and with the maximum relevant information remains an important issue for researchers in this field. Signal processing is largely performed for ECG analysis and interpretation, but this process can be susceptible to error in the delineation phase. In addition, it can lead to the loss of important information that is usually considered as noise and, consequently, discarded from the analysis. The goal of this study was to evaluate if the ECG noise allows for the classification of emotions, while using its entropy as an input in a decision tree classifier. We collected the ECG signal from 25 healthy participants while they were presented with videos eliciting negative (fear and disgust) and neutral emotions. The results indicated that the neutral condition showed a perfect identification (100%), whereas the classification of negative emotions indicated good identification performances (60% of sensitivity and 80% of specificity). These results suggest that the entropy of noise contains relevant information that can be useful to improve the analysis of the physiological correlates of emotion.

Educational software in english to the awareness in a preschool education supervised context

Information and communication technologies play an increasingly important role in society, in the sense that all areas and professions make use of digital resources. The school can not be brushed off this reality, aim to create full subjects and integrated in society today. Educational software can be used very early in the education of children, but they must be carefully and monitoring. This article aims to present the results of the use of educational software in English to the awareness of context with children of pre-school education in kindergarten, nursery center Redemptorist Fathers - The smallest fox in White Castle, a 21 group children under 5 years. Early awareness of foreign language such as English can be started with digital multimedia capabilities and various software available on the market. However, the small study described the case reveals some resistance from parents and educators, in the preparation of these to choose and monitor the use of ICT by children, in addition to also highlight the self-interest of the children involved and their learning a few words in English language in different contexts of daily worked. The study opens perspectives on close monitoring needs of such uses and training of educators in the field of use of resources multilingual awareness in pre-school education.

The use of digital educational resources : the contribution of research in the supervised teaching practice in basic education

This article aims to reflect on the impact of Information and Communication Technologies (ICT) in the educational context, focusing on the potential contributions of the use of Digital Educational Resources (RED) in the process of teaching and learning. For this purpose, the results of the use of the RED will be presented:. Digital Classroom - The World's 1st Year Carochinha The study was accomplished in a class of the 1st grade of the 1st CEB, composed of 27 students, aged 6-7 years in Castelo Branco City Schools Group within the Supervised Teaching Practice. The results obtained after the analysis and processing of the data showed that when using this RED students show they have acquired the content covered by the fact that they enhanced levels of greater interest, commitment, motivation, commitment and initiative in the course of activities proposals. But, perhaps because they are students of 1st year of the 1st CEB, do not neglect the presence and monitoring of the teacher and the use of paper-based resources. This means that there should be a complementarity that reconciles the human factor (teacher), with the use of digital media resources and paper support resources (Manual).

Detecting outliers and asserting consistency in agriculture ground truth information by using temporal VI data from modis.

Collecting ground truth data is an important step to be accomplished before performing a supervised classification. However, its quality depends on human, financial and time ressources. It is then important to apply a validation process to assess the reliability of the acquired data. In this study, agricultural infomation was collected in the Brazilian Amazonian State of Mato Grosso in order to map crop expansion based on MODIS EVI temporal profiles. The field work was carried out through interviews for the years 2005-2006 and 2006-2007. This work presents a methodology to validate the training data quality and determine the optimal sample to be used according to the classifier employed. The technique is based on the detection of outlier pixels for each class and is carried out by computing Mahalanobis distances for each pixel. The higher the distance, the further the pixel is from the class centre. Preliminary observations through variation coefficent validate the efficiency of the technique to detect outliers. Then, various subsamples are defined by applying different thresholds to exclude outlier pixels from the classification process. The classification results prove the robustness of the Maximum Likelihood and Spectral Angle Mapper classifiers. Indeed, those classifiers were insensitive to outlier exclusion. On the contrary, the decision tree classifier showed better results when deleting 7.5% of pixels in the training data. The technique managed to detect outliers for all classes. In this study, few outliers were present in the training data, so that the classification quality was not deeply affected by the outliers.

Comparison of multitemporal MODIS-EVI smoothing algorithms and its contribution to crop monitoring.

Crop monitoring and more generally land use change detection are of primary importance in order to analyze spatio-temporal dynamics and its impacts on environment. This aspect is especially true in such a region as the State of Mato Grosso (south of the Brazilian Amazon Basin) which hosts an intensive pioneer front. Deforestation in this region as often been explained by soybean expansion in the last three decades. Remote sensing techniques may now represent an efficient and objective manner to quantify how crops expansion really represents a factor of deforestation through crop mapping studies. Due to the special characteristics of the soybean productions' farms in Mato Grosso (area varying between 1000 hectares and 40000 hectares and individual fields often bigger than 100 hectares), the Moderate Resolution Imaging Spectroradiometer (MODIS) data with a near daily temporal resolution and 250 m spatial resolution can be considered as adequate resources to crop mapping. Especially, multitemporal vegetation indices (VI) studies have been currently used to realize this task [1] [2]. In this study, 16-days compositions of EVI (MODQ13 product) data are used. However, although these data are already processed, multitemporal VI profiles still remain noisy due to cloudiness (which is extremely frequent in a tropical region such as south Amazon Basin), sensor problems, errors in atmospheric corrections or BRDF effect. Thus, many works tried to develop algorithms that could smooth the multitemporal VI profiles in order to improve further classification. The goal of this study is to compare and test different smoothing algorithms in order to select the one which satisfies better to the demand which is classifying crop classes. Those classes correspond to 6 different agricultural managements observed in Mato Grosso through an intensive field work which resulted in mapping more than 1000 individual fields. The agricultural managements above mentioned are based on combination of soy, cotton, corn, millet and sorghum crops sowed in single or double crop systems. Due to the difficulty in separating certain classes because of too similar agricultural calendars, the classification will be reduced to 3 classes : Cotton (single crop), Soy and cotton (double crop), soy (single or double crop with corn, millet or sorghum). The classification will use training data obtained in the 2005-2006 harvest and then be tested on the 2006-2007 harvest. In a first step, four smoothing techniques are presented and criticized. Those techniques are Best Index Slope Extraction (BISE) [3], Mean Value Iteration (MVI) [4], Weighted Least Squares (WLS) [5] and Savitzky-Golay Filter (SG) [6] [7]. These techniques are then implemented and visually compared on a few individual pixels so that it allows doing a first selection between the five studied techniques. The WLS and SG techniques are selected according to criteria proposed by [8]. Those criteria are: ability in eliminating frequent noises, conserving the upper values of the VI profiles and keeping the temporality of the profiles. Those selected algorithms are then programmed and applied to the MODIS/TERRA EVI data (16-days composition periods). Tests of separability are realized based on the Jeffries-Matusita distance in order to see if the algorithms managed in improving the potential of differentiation between the classes. Those tests are realized on the overall profile (comprising 23 MODIS images) as well as on each MODIS sub-period of the profile [1]. This last test is a double interest process because it allows comparing the smoothing techniques and also enables to select a set of images which carries more information on the separability between the classes. Those selected dates can then be used to realize a supervised classification. Here three different classifiers are tested to evaluate if the smoothing techniques as a particular effect on the classification depending on the classifiers used. Those classifiers are Maximum Likelihood classifier, Spectral Angle Mapper (SAM) classifier and CHAID Improved Decision tree. It appears through the separability tests on the overall process that the smoothed profiles don't improve efficiently the potential of discrimination between classes when compared with the original data. However, the same tests realized on the MODIS sub-periods show better results obtained with the smoothed algorithms. The results of the classification confirm this first analyze. The Kappa coefficients are always better with the smoothing techniques and the results obtained with the WLS and SG smoothed profiles are nearly equal. However, the results are different depending on the classifier used. The impact of the smoothing algorithms is much better while using the decision tree model. Indeed, it allows a gain of 0.1 in the Kappa coefficient. While using the Maximum Likelihood end SAM models, the gain remains positive but is much lower (Kappa improved of 0.02 only). Thus, this work's aim is to prove the utility in smoothing the VI profiles in order to improve the final results. However, the choice of the smoothing algorithm has to be made considering the original data used and the classifier models used. In that case the Savitzky-Golay filter gave the better results.

Sperimentazione di Deep Metric Loss per Self-Supervised Information Retrieval Systems su CORD19.

Dopo lo sviluppo dei primi casi di Covid-19 in Cina nell’autunno del 2019, ad inizio 2020 l’intero pianeta è precipitato in una pandemia globale che ha stravolto le nostre vite con conseguenze che non si vivevano dall’influenza spagnola. La grandissima quantità di paper scientifici in continua pubblicazione sul coronavirus e virus ad esso affini ha portato alla creazione di un unico dataset dinamico chiamato CORD19 e distribuito gratuitamente. Poter reperire informazioni utili in questa mole di dati ha ulteriormente acceso i riflettori sugli information retrieval systems, capaci di recuperare in maniera rapida ed efficace informazioni preziose rispetto a una domanda dell'utente detta query. Di particolare rilievo è stata la TREC-COVID Challenge, competizione per lo sviluppo di un sistema di IR addestrato e testato sul dataset CORD19. Il problema principale è dato dal fatto che la grande mole di documenti è totalmente non etichettata e risulta dunque impossibile addestrare modelli di reti neurali direttamente su di essi. Per aggirare il problema abbiamo messo a punto nuove soluzioni self-supervised, a cui abbiamo applicato lo stato dell'arte del deep metric learning e dell'NLP. Il deep metric learning, che sta avendo un enorme successo soprattuto nella computer vision, addestra il modello ad "avvicinare" tra loro immagini simili e "allontanare" immagini differenti. Dato che sia le immagini che il testo vengono rappresentati attraverso vettori di numeri reali (embeddings) si possano utilizzare le stesse tecniche per "avvicinare" tra loro elementi testuali pertinenti (e.g. una query e un paragrafo) e "allontanare" elementi non pertinenti. Abbiamo dunque addestrato un modello SciBERT con varie loss, che ad oggi rappresentano lo stato dell'arte del deep metric learning, in maniera completamente self-supervised direttamente e unicamente sul dataset CORD19, valutandolo poi sul set formale TREC-COVID attraverso un sistema di IR e ottenendo risultati interessanti.

Self-supervised information retrieval: a novel approach based on Deep Metric Learning and Neural Language Models

Most of the existing open-source search engines, utilize keyword or tf-idf based techniques to find relevant documents and web pages relative to an input query. Although these methods, with the help of a page rank or knowledge graphs, proved to be effective in some cases, they often fail to retrieve relevant instances for more complicated queries that would require a semantic understanding to be exploited. In this Thesis, a self-supervised information retrieval system based on transformers is employed to build a semantic search engine over the library of Gruppo Maggioli company. Semantic search or search with meaning can refer to an understanding of the query, instead of simply finding words matches and, in general, it represents knowledge in a way suitable for retrieval. We chose to investigate a new self-supervised strategy to handle the training of unlabeled data based on the creation of pairs of ’artificial’ queries and the respective positive passages. We claim that by removing the reliance on labeled data, we may use the large volume of unlabeled material on the web without being limited to languages or domains where labeled data is abundant.

Supervised and weakly supervised counting-by-segmentation: the fluorescent microscopy use case

This thesis focuses on automating the time-consuming task of manually counting activated neurons in fluorescent microscopy images, which is used to study the mechanisms underlying torpor. The traditional method of manual annotation can introduce bias and delay the outcome of experiments, so the author investigates a deep-learning-based procedure to automatize this task. The author explores two of the main convolutional-neural-network (CNNs) state-of-the-art architectures: UNet and ResUnet family model, and uses a counting-by-segmentation strategy to provide a justification of the objects considered during the counting process. The author also explores a weakly-supervised learning strategy that exploits only dot annotations. The author quantifies the advantages in terms of data reduction and counting performance boost obtainable with a transfer-learning approach and, specifically, a fine-tuning procedure. The author released the dataset used for the supervised use case and all the pre-training models, and designed a web application to share both the counting process pipeline developed in this work and the models pre-trained on the dataset analyzed in this work.

Directly training spiking neural networks for cyber-physical systems: from supervised to reinforcement learning

Spiking Neural Networks (SNNs) are bio-inspired Artificial Neural Networks (ANNs) utilizing discrete spiking signals, akin to neuron communication in the brain, making them ideal for real-time and energy-efficient Cyber-Physical Systems (CPSs). This thesis explores their potential in Structural Health Monitoring (SHM), leveraging low-cost MEMS accelerometers for early damage detection in motorway bridges. The study focuses on Long Short-Term SNNs (LSNNs), although their complex learning processes pose challenges. Comparing LSNNs with other ANN models and training algorithms for SHM, findings indicate LSNNs' effectiveness in damage identification, comparable to ANNs trained using traditional methods. Additionally, an optimized embedded LSNN implementation demonstrates a 54% reduction in execution time, but with longer pre-processing due to spike-based encoding. Furthermore, SNNs are applied in UAV obstacle avoidance, trained directly using a Reinforcement Learning (RL) algorithm with event-based input from a Dynamic Vision Sensor (DVS). Performance evaluation against Convolutional Neural Networks (CNNs) highlights SNNs' superior energy efficiency, showing a 6x decrease in energy consumption. The study also investigates embedded SNN implementations' latency and throughput in real-world deployments, emphasizing their potential for energy-efficient monitoring systems. This research contributes to advancing SHM and UAV obstacle avoidance through SNNs' efficient information processing and decision-making capabilities within CPS domains.

Depth Estimation in Stereo Biomedical Images via Proxy-Supervised Deep Learning

In order to estimate depth through supervised deep learning-based stereo methods, it is necessary to have access to precise ground truth depth data. While the gathering of precise labels is commonly tackled by deploying depth sensors, this is not always a viable solution. For instance, in many applications in the biomedical domain, the choice of sensors capable of sensing depth at small distances with high precision on difficult surfaces (that present non-Lambertian properties) is very limited. It is therefore necessary to find alternative techniques to gather ground truth data without having to rely on external sensors. In this thesis, two different approaches have been tested to produce supervision data for biomedical images. The first aims to obtain input stereo image pairs and disparities through simulation in a virtual environment, while the second relies on a non-learned disparity estimation algorithm in order to produce noisy disparities, which are then filtered by means of hand-crafted confidence measures to create noisy labels for a subset of pixels. Among the two, the second approach, which is referred in literature as proxy-labeling, has shown the best results and has even outperformed the non-learned disparity estimation algorithm used for supervision.