@book {152, title = {Computer vision: models, learning, and inference}, year = {2012}, publisher = {Cambridge University Press}, organization = {Cambridge University Press}, url = {http://www.computervisionmodels.com/}, author = {Prince, Simon JD} } @conference {230, title = {Aplicaciones de Internet de las Cosas SIPIA6-Red de Sensores Inal{\'a}mbricos con IPv6}, booktitle = {XV Workshop de Investigadores en Ciencias de la Computaci{\'o}n}, year = {2013}, author = {Mercado, Gustavo and Borgo, Roberto and Gonzalez Antivilo, Francisco and Taffernaberry, Juan Carlos and Diedrichs, Ana and Aguirre, Mat{\'\i}as and Robles, Mar{\'\i}a In{\'e}s and Grunwaldt, Guillermo and Tabacchi, Germ{\'a}n and Tromer, Sebasti{\'a}n and others} } @conference {114, title = {Application of a Bayesian semi-supervised Learning Strategy to Network Intrution Detection}, booktitle = {Proceedings of ASAI 2010, Argentinean Symposioum of Artificial Intelligence}, year = {2010}, publisher = {SADIO}, organization = {SADIO}, address = {Buenos Aires}, author = {Catania, C. A. and Garcia Garino, C. and Bromberg, F.} } @conference {90, title = {Aprendizaje de independencias espec{\'\i}ficas del contexto en Markov random fields}, booktitle = {XVII Congreso Argentino de Ciencias de la Computaci{\'o}n}, year = {2011}, abstract = {
Los modelos no dirigidos o Markov random fields son ampliamente utilizados para problemas que aprenden una distribuci{\'o}n desconocida desde un conjunto de datos. Esto es porque permiten representar una distribuci{\'o}n eficientemente al hacer expl{\'\i}citas las independencias condicionales que pueden existir entre sus variables. Adem{\'a}s de estas independencias es posible representar otras, las Independencias Espec{\'\i}ficas del Contexto (CSIs) que a diferencia de las anteriores s{\'o}lo son v{\'a}lidas bajo ciertos valores que pueden tomar subconjuntos de sus variables. Debido a esto son complicadas de representar y aprenderlas desde datos. En este trabajo presentamos un enfoque para representar CSIs en modelos no dirigidos y un algoritmo que las aprende desde datos utilizando tests estad{\'\i}sticos. Mostramos resultados donde los modelos aprendidos por nuestro algoritmo resultan ser mejores o comparables a modelos aprendidos por otros sin utilizar CSIs.
}, author = {Edera, Alejandro and Bromberg, Facundo} } @conference {168, title = {Arm muscular effort estimation from images using Computer Vision and Machine Learning}, booktitle = {1st International Conference on Ambient Intelligence for Health}, year = {2015}, month = {Dec}, address = {Puerto Varas, Chile}, abstract = {A problem of great interest in disciplines like occupational\ medicine, ergonomics, and sports, is the measurement of biomechanical\ variables involved in human movement and balance such as internal\ muscle forces and joint torques. This problem \ is solved by a two-step process: data capturing using impractical, intrusive and expensive devices\ that is then used as input in \ complex models for obtaining the biomechanical variables of interest. In this work we present a first step towards\ capturing input \ data through a more automated, non-intrusive and economic process, specifically weight held by an arm subject to isometric\ contraction as a measure of muscular effort. We do so, by processing\ RGB images of the arm with computer vision (Local Binary Patterns\ and Color Histograms) and estimating the effort with machine learning\ algorithms (SVM and Random \ Forests). In the best case we obtained an\ FMeasure= 70.68\%, an Accuracy= 71.66\% and a mean absolute error in\ the predicted weights of 554.16 grs (over 3 possible levels of effort). Considering the difficulty of the problem , it is enlightening to achieve over\ random results indicating that, despite the simplicity of the approach,\ it is possible to extract meaningful information for the \ predictive task.\ Moreover, the simplicity of the approach suggests many lines of further\ improvements: on the image capturing \ side with other kind of images;\ on the feature extraction side with more sophisticated algorithms and\ features; and on the\ knowledge extraction side with more sophisticated\ learning algorithms.\
}, keywords = {color histograms, LBP, muscle arm effort, random forests, SVM}, author = {Abraham, Leandro and Bromberg, Facundo and Forradellas, Raymundo} } @conference {74, title = {Automated 3D segmentation and analysis of cotton plants}, booktitle = {Digital Image Computing Techniques and Applications (DICTA), 2011 International Conference on}, year = {2011}, publisher = {IEEE}, organization = {IEEE}, author = {Paproki, Anthony and Fripp, Jurgen and Salvado, Olivier and Sirault, Xavier and Berry, Scott and Furbank, Robert} } @conference {179, title = {An Autonomous labeling approach to SVM algorithms for network traffic anomaly detection}, booktitle = {Argentine Symposium of Artificial Intelligence (ASAI). Jornadas Argentinas de Inform{\'a}tica. Mar del Plata, Argentina.}, year = {2009}, month = {08/2009}, publisher = {Sociedad Argentina de Inform{\'a}tica}, organization = {Sociedad Argentina de Inform{\'a}tica}, address = {Mar del Plata, Argentina}, abstract = {In the past years, several support vector machines anomaly detection approaches have been proposed in the network intrusion detetion field. The main advantage of these approaches is that they\ can\ characterize normal traffic\ when trained using a data set\ containing not only normal traffic\ but also possible attacks. Unfortunately, these algorithms seem to be accurate only when the normal traffic\ vastly outnumbers the numbers of attacks or anomalies present in the dataset. This work presents an approach for autonomous labeling of normal traffic\ \ as a way of dealing with situations where\ class distributions do not present the required unbalance. The autonomous labeling process is made by SNORT, a misuse-based intrusion detection system. Experiments\ conducted on the 1998 DARPA dataset show the proposed autonomous labeling approach not only outperforms existing SVM alternatives but also obtains signicant improvement over SNORT itself.
}, author = {Brombereg, Facundo and Catania, Carlos A. and Garcia Garino, Carlos} } @conference {223, title = {A demo of the PEACH IoT-based frost event prediction system for precision agriculture}, booktitle = {Sensing, Communication, and Networking (SECON), 2016 13th Annual IEEE International Conference on}, year = {2016}, publisher = {IEEE}, organization = {IEEE}, doi = {10.1109/SAHCN.2016.7732963}, url = {https://hal.inria.fr/hal-01311527/document}, author = {Brun-Laguna, Keoma and Diedrichs, Ana Laura and Chaar, Javier Emilio and Dujovne, Diego and Taffernaberry, Juan Carlos and Mercado, Gustavo and Watteyne, Thomas} } @conference {246, title = {Exploring the Influence of Self-determination in the Collective Intelligence of Collaborative Organizations}, booktitle = {Symposium on implementing collaborative governance}, year = {2018}, month = {10/2018}, address = {Palermo, Italy.}, abstract = {This work introduces a novel technopolitical artifact that opens up the possibility for a system of
non-dominant citizen governance that eliminates the representative layer and puts citizens at the top
of the decision making process. A socio-political approach that arises by merging two disrupting
technologies: Sociocracy 3.0 as an organizational technology for efficiently governing large, self-
determinant organizations; and Blockchain as a digital reputation technology for overcoming the
limitations preventing standalone Sociocracy 3.0 of being used in organizations governing the
public domain.
In recent years, positive correlations between some factors of collaborative group task processes and the increasing of collective intelligence (CI) have been presented. This work introduces an hypothesis that argues the existence of a new factor of positive influence for the increasing of collective intelligence in collaborative group tasks operating in cooperative environments: self-determination. Therefore, we present an argumentation based on Cooperative Multiagent Systems that spotlights the significance of self-determination in these particular environments. Furthermore, we also introduce a preliminary design of an experimental setup and a methodological framework for validating the hypothesis empirically in human organizations. Our propose consists on measuring, on the one hand, the level of self-determination from the individuals that participate on the decision-making processes, and on the other hand, on measuring the level of collective intelligence achieved by performing collaborative group task. Finally, we propose to use statistical analysis to explore if there are positive correlations between self-determination and collective intelligence in cooperative environments, such as collaborative organizations.
}, author = {Ribas, Alexandre}, editor = {Bromberg, Facundo} } @conference {88, title = {The Grow-Shrink strategy for learning Markov network structures constrained by context-specific independences}, booktitle = {14th edition of the Ibero-American Conference on Artificial Intelligence}, year = {2014}, month = {11/2014}, publisher = {Lecture Notes in Computer Science/Lecture Notes in Artificial Intelligence LNCS/LNAI series}, organization = {Lecture Notes in Computer Science/Lecture Notes in Artificial Intelligence LNCS/LNAI series}, address = {Santiago de Chile}, abstract = {Markov networks are models for compactly representing complex probability distributions. They are composed by a structure and a set of numerical weights. The structure qualitatively describes independences in the distribution, which can be exploited to factorize the distribution into a set of compact functions. A key application for learning structures from data is to automatically discover knowledge. In practice, structure learning algorithms focused on "knowledge discovery" present a limitation: they use a coarse-grained representation of the structure. As a result, this representation cannot describe context-specific independences. Very recently, an algorithm called CSPC was designed to overcome this limitation, but it has a high computational complexity. This work tries to mitigate this downside presenting CSGS, an algorithm that uses the Grow-Shrink strategy for reducing unnecessary computations. On an empirical evaluation, the structures learned by CSGS achieve competitive accuracies and lower computational complexity with respect to those obtained by CSPC.
}, isbn = {978-3-319-12027-0}, url = {http://www.springer.com/computer/ai/book/978-3-319-12026-3}, author = {Edera, Alejandro and Strappa, Yanela and Bromberg, Facundo} } @conference {92, title = {Independence-based MAP for Markov networks structure discovery}, booktitle = {23rd IEEE International Conference on Tools with Artificial Intelligence (ICTAI)}, year = {2011}, month = {11/2013}, publisher = {IEEE}, organization = {IEEE}, author = {Bromberg, F. and Schl{\"u}ter, F. and Edera, A.} } @conference {103, title = {Interpolaci{\'o}n Espacial Mediante Aprendizaje de M{\'a}quinas en Vi{\~n}edos de la Provincia de Mendoza, Argentina.}, booktitle = {13th Argentine Symposium on Artificial Intelligence}, year = {2012}, isbn = {1850-2784}, author = {Bromberg, F and P{\'e}rez, Diego S} } @conference {86, title = {Learning Markov networks with context-specific independences}, booktitle = {IEEE 25th International Conference on Tools with Artificial Intelligence (ICTAI)}, year = {2013}, month = {Nov}, keywords = {Algorithm design and analysis, Complexity theory, conditional independences, Context, Context modeling, context-specific independences, CSPC algorithm, Encoding, independence relations, independence-based, independence-based algorithms, independence-based learning approach, Inference algorithms, learning (artificial intelligence), learning structures, log-linear model, machine learning, Markov network structure, Markov networks, Markov processes, Markov random fields, network theory (graphs), Structure learning, undirected graph}, doi = {10.1109/ICTAI.2013.88}, author = {Edera, A. and Schl{\"u}ter, F. and Bromberg, F.} } @conference {183, title = {Low-power wireless sensor network for frost monitoring in agriculture research}, booktitle = {Biennial Congress of Argentina (ARGENCON), 2014 IEEE}, year = {2014}, publisher = {IEEE}, organization = {IEEE}, address = {Bariloche}, abstract = {This work presents the development of a wireless sensor network (WSN), based on IEEE-802.15.4, in order to be used for frost characterization in precision agriculture by measuring temperature. Our key objective is to reduce the power consumption of the network to the minimum, allowing several measurement points per node and the remote monitoring of the sensors behaviour. For the communication interface between a WSN node and the sensors, we have developed a serial protocol inspired in SDI-12. Preliminary results show a low-cost and low-power WSN. The user can access and use the data for agronomic research.
}, isbn = {978-1-4799-4270-1}, doi = {10.1109/ARGENCON.2014.6868546}, url = {http://dx.doi.org/10.1109/ARGENCON.2014.6868546}, author = {Diedrichs, Ana Laura and Tabacchi, Germ{\'a}n and Grunwaldt, Guillermo and Pecchia, Mat{\'\i}as and Mercado, Gustavo and Gonzalez Antivilo, Francisco} } @conference {222, title = {(Not so) intuitive results from a smart agriculture low-power wireless mesh deployment}, booktitle = {Proceedings of the Eleventh ACM Workshop on Challenged Networks}, year = {2016}, publisher = {ACM}, organization = {ACM}, address = {New York city}, abstract = {This work presents S3LF, a socio-technical system in the form of a mobile application for
facilitating a digital self-determinant governance experience in collaborative organizations. It
consists of a technologically mediated adaptation of the decision making framework Sociocracy 3.0
a.k.a S3, turning it into an asynchronous and remote process with a facilitation mediated through the
user experience (UX) and interaction design. S3LF enables digital and distributed organizations
such as Collaborative Networked Organizations (CNOs) to adopt self-determinant governancemethodologies such as S3, and promises a faster onboarding into self-determinant governance and
higher throughput of self-determinant decision making. Furthermore, we introduce in S3LF a
self-diagnosis module that reports the level of self-determination perceived by each decision maker
of the group for each decision made, over different contextual situations; whose analysis has the
potential to allow organizations to self-correct dominant situations.
This work introduces a novel technopolitical artifact that opens up the possibility for a system of non-dominant citizen governance that eliminates the representative layer and puts citizens at the top of the decision making process. A socio-political approach that arises by merging two disrupting technologies: Sociocracy 3.0 as an organizational technology for efficiently governing large, self-determinant organizations; and Blockchain as a digital reputation technology for overcoming the limitations preventing standalone Sociocracy 3.0 of being used in organizations governing the public domain.
}, keywords = {Blockchain, citizen governance, reputation, self-determination, Sociocracy 3.0.}, author = {{\`A}lex Ribas and Facundo Bromberg} } @conference {105, title = {Variante de grow shrink para mejorar la calidad de markov blankets}, booktitle = {XXXV Latin American Informatics Conference (CLEI), Pelotas, Brasil}, year = {2009}, month = {10/2009}, author = {Bromberg, Facundo and Schl{\"u}ter, F} } @proceedings {178, title = {Characterization of LQI behavior in WSN for glacier area in Patagonia Argentina}, journal = {Embedded Systems (SASE/CASE), 2013 Fourth Argentine Symposium and Conference on}, year = {2013}, month = {08/2013}, pages = {1--6}, publisher = {IEEE}, address = {Buenos Aires, Argentina}, abstract = {One of the most important aspects before installing a Wireless Sensor Network (WSN) is a previous study of connectivity constraints that exist in the area to be covered. This study is critical to the final distribution of the sensors, with an important impact in the life of the network by reducing consumption, and on the robustness by contemplating redundancy of paths and sensors. In this paper, we present a summary of the most important aspects of a preliminary empirical study of the Link Quality Indicator (LQI), on different landscapes in the glaciers area of Patagonia Argentina. The landscapes covered varied in geographical structures with different levels of attenuation and extreme environmental conditions. Through the analysis of the Cumulative Distribution Function (CDF) of the measured LQI values, we can characterize the behavior of four different scenarios and correlate the combined effects of the environmental structure with the distance from the transmitter. The measurements performed were designed for characterizing the links at the physical layer with the purpose of defining models to estimate the Packet Error Rate (PER) for the WSN deployment stage.
}, issn = {978-1-4799-1098-4}, doi = {10.1109/SASE-CASE.2013.6636777}, url = {http://dx.doi.org/10.1109/SASE-CASE.2013.6636777}, author = {Diedrichs, Ana Laura and Robles, Mar{\'\i}a In{\'e}s and Bromberg, Facundo and Mercado, Gustavo and Dujovne, Diego} } @proceedings {174, title = {Efficient and Robust Independence-Based Markov Network Structure Discovery.}, journal = {20th International Joint Conference of Artificial Inteliigence (IJCAI)}, year = {2007}, pages = {2431-2436 }, publisher = {Morgan Kaufmann Publishers Inc.}, address = {San Francisco, CA}, abstract = {In this paper we introduce a novel algorithm for the induction of the Markov network structure of a domain from the outcome of conditional independence tests on data. Such algorithms work by successively restricting the set of possible structures until there is only a single structure consistent with the conditional independence tests executed. Existing independence-based algorithms have wellknown shortcomings, such as rigidly ordering the sequence of tests they perform, resulting in potential inefficiencies in the number of tests required, and committing fully to the test outcomes, resulting in lack of robustness in case of unreliable tests. We address both problems through a Bayesian particle filtering approach, which uses a population of Markov network structures to maintain the posterior probability distribution over them, given the outcomes of the tests performed. Instead of a fixed ordering, our approach greedily selects, at each step, the optimally informative from a pool of candidate tests according to information gain. In addition, it maintains multiple candidate structures weighed by posterior probability, which makes it more robust to errors in the test outcomes. The result is an approximate algorithm (due to the use of particle filtering) that is useful in domains where independence tests are uncertain (such as applications where little data is available) or expensive (such as cases of very large data sets and/or distributed data).
}, url = {http://www.aaai.org/Library/IJCAI/2007/ijcai07-391.php}, author = {Bromberg, Facundo and Margaritis, Dimitris} } @proceedings {176, title = {Efficient Markov network structure discovery using independence tests}, journal = {Proceedings of the SIAM Conference in Data Mining}, year = {2006}, pages = {141--152}, address = {Bethesda, Maryland, USA}, abstract = {We present two algorithms for learning the structure of a Markov network from discrete data: GSMN and GSIMN. Both algorithms use statistical conditional independence tests on data to infer the structure by successively constraining the set of structures consistent with the results of these tests. GSMN is a natural adaptation of the Grow-Shrink algorithm of Margaritis and Thrun for learning the structure of Bayesian networks. GSIMN extends GSMN by additionally exploiting Pearl{\textquoteright}s well-known properties of conditional independence relations to infer novel independencies from known independencies, thus avoiding the need to perform these tests. Experiments on artificial and real data sets show GSIMN can yield savings of up to 70\% with respect to GSMN, while generating a Markov network with comparable or in several cases considerably improved quality. In addition to GSMN, we also compare GSIMN to a forward-chaining implementation, called GSIMN-FCH, that produces all possible conditional independence results by repeatedly applying Pearl{\textquoteright}s theorems on the known conditional independence tests. The results of this comparison show that GSIMN is nearly optimal in terms of the number of tests it can infer, under a fixed ordering of the tests performed.
}, isbn = {978-0-89871-611-5}, doi = {10.1137-1.9781611972764.13}, url = {http://epubs.siam.org/doi/abs/10.1137/1.9781611972764.13}, author = {Bromberg, Facundo and Margaritis, Dimitris and Honavar, Vasant} } @proceedings {177, title = {Estimaci{\'o}n de carga muscular mediante im{\'a}genes}, journal = {Argentinean Symposium of Artificial Intelligence (ASAI) - Jornadas Argentinas de Inform{\'a}tica}, year = {2014}, pages = {91--98}, publisher = {Sociedad Argentina de Inform{\'a}tica}, address = {Buenos Aires, Argentina}, abstract = {Un problema de gran inter{\'e}s en disciplinas como la ocupacional, ergon{\'o}mica y deportiva, es la medici{\'o}n de variables biomec{\'a}nicas involucradas en\ el movimiento humano (como las fuerzas musculares internas y torque de articulaciones).\ Actualmente este problema se resuelve en un proceso de dos pasos.\ Primero capturando datos con dispositivos\ poco pr{\'a}cticos, intrusivos y costosos.\ Luego estos datos son usados como entrada en modelos complejos para obtener\ las variables biomec{\'a}nicas como salida. El presente trabajo representa una alternativa automatizada, no intrusiva y econ{\'o}mica al primer paso, proponiendo la\ captura de estos datos a trav{\'e}s de im{\'a}genes. En trabajos futuros la idea es automatizar todo el proceso de c{\'a}lculo de esas variables. En este trabajo elegimos un\ caso particular de medici{\'o}n de variables biomec{\'a}nicas: el problema de estimar\ el nivel discreto de carga muscular que est{\'a}n ejerciendo los m{\'u}sculos de un brazo. Para \ estimar a partir de im{\'a}genes est{\'a}ticas del brazo ejerciendo la fuerza de\ sostener la carga, el nivel de la misma, realizamos un\ proceso de clasificaci{\'o}n.\ Nuestro enfoque utiliza Support Vector Machines para clasificaci{\'o}n, combinada\ con una etapa de pre-procesamiento que extrae caracter{\'\i}sticas visuales utilizando variadas t{\'e}cnicas (Bag of Keypoints, Local Binary Patterns, Histogramas de\ Color, Momentos de Contornos) En los mejores casos (Local Binary Patterns y\ Momentos de Contornos) obtenemos medidas de performance en la clasificaci{\'o}n\ (Precision, Recall, F-Measure y Accuracy) superiores al 90 \%.
}, url = {http://43jaiio.sadio.org.ar/proceedings/ASAI/search.html}, author = {Abraham, Leandro and Bromberg, Facundo and Forradellas, Raymundo} } @proceedings {173, title = {Learning Markov Network Structure using Few Independence Tests.}, journal = {SIAM Data Mining}, year = {2008}, pages = {680--691}, abstract = {Weed density and composition are not uniform throughout the field, nevertheless, the conventional approach is to carry out a uniform application. Object Detection Networks have already arrived in agricultural applications that can be used for weed management. The current study developed a detection and classification of weeds system in a one-step procedure using RetinaNet Object Detection Network. The procedure was based on identifying Solanum nigrum L., Cyperus rotundus L. and Echinochloa crus-galli L. and two growth stages both for a broadleaf species (S. nigrum) as well as narrow-leaved species (C. rotundus) in a maize field. The predictions were evaluated by mAP metric. The result obtained was 0.88 with values between 0.98 and 0.75 depending on the class.
}, keywords = {deep learning, object detection networks, RetinaNet, site-specific weed management}, isbn = {978-90-8686-916-9}, issn = {978-90-8686-363-1}, doi = {https://doi.org/10.3920/978-90-8686-916-9}, url = {https://www.wageningenacademic.com/doi/abs/10.3920/978-90-8686-916-9}, author = {Correa, JM L{\'o}pez and Todeschini, M and P{\'e}rez, DS and Karouta, J and Bromberg, F and Ribeiro, A and And{\'u}jar, D} } @proceedings {225, title = {Sonda de temperatura para uso agr{\'\i}cola}, journal = {CASE Congreso Argentino de Sistemas Embebidos}, year = {2015}, edition = {2015}, issn = {978-987-45523-3-4}, author = {Guillermo Gr{\"u}nwaldt and Mat{\'\i}as Pecchia and Ana Laura Diedrichs and Germ{\'a}n Tabacchi and Gustavo Mercado} } @article {113, title = {An autonomous labeling approach to support vector machines algorithms for network traffic anomaly detection}, journal = {Expert Systems with Applications}, volume = {39}, year = {2012}, pages = {1822{\textendash}1829}, author = {Catania, Carlos A and Bromberg, Facundo and Garino, Carlos Garc{\'\i}a} } @article {211, title = {{B}lankets {J}oint {P}osterior score for learning {M}arkov network structures }, journal = {International Journal of Approximate Reasoning}, volume = {https://doi.org/10.1016/j.ijar.2017.10.018}, year = {2017}, month = {10/2017}, type = {Regular Issu}, abstract = {Markov networks are extensively used to model\ complex sequential, spatial, and relational interactions\ in a wide range of fields.\ By learning the Markov network independence structure of a domain, more accurate joint probability distributions can be obtained for inference tasks or, more directly, for interpreting the most significant relations among the variables. Recently, several researchers have investigated techniques for automatically learning the structure from data by obtaining the probabilistic maximum-a-posteriori structure given the available data. However, all the approximations proposed decompose the posterior of the whole structure into local sub-problems, by assuming that the posteriors of\ the Markov blankets of all the variables are mutually independent.\ In this work, we propose a scoring function for relaxing such assumption.\ The Blankets Joint Posterior\ score computes the\ joint posterior of structures as a joint distribution of the collection of its Markov blankets. Essentially, the whole posterior is obtained by computing the posterior of the blanket of each variable as a conditional distribution that takes into account information from other blankets in the network.\ We show in our experimental results that the proposed approximation\ can improve the sample complexity of state-of-the-art competitors\ when learning complex networks, where the independence assumption between blanket variables is clearly incorrect.\
}, keywords = {blankets posterior, irregular structures, Markov network, scoring function, Structure learning}, author = {Schl{\"u}ter, Federico and Strappa, Yanela and Bromberg, Facundo and Milone, Diego H.} } @article {170, title = {Efficient Markov network discovery using particle filters}, journal = {Computational Intelligence}, volume = {25}, year = {2009}, month = {11/2009}, pages = {367{\textendash}394}, abstract = {In this paper, we introduce an efficient independence-based algorithm for the induction of the Markov network (MN) structure of a domain from the outcomes of independence test conducted on data. Our algorithm utilizes a particle filter (sequential Monte Carlo) method to maintain a population of MN structures that represent the posterior probability distribution over structures, given the outcomes of the tests performed. This enables us to select, at each step, the maximally informative test to conduct next from a pool of candidates according to information gain, which minimizes the cost of the statistical tests conducted on data. This makes our approach useful in domains where independence tests are expensive, such as cases of very large data sets and/or distributed data. In addition, our method maintains multiple candidate structures weighed by posterior probability, which allows flexibility in the presence of potential errors in the test outcomes.
}, keywords = {graphical model structure learning, Markov networks, particle filters, sequential Monte Carlo}, doi = {10.1111/j.1467-8640.2009.00347.x}, url = {http://onlinelibrary.wiley.com/doi/10.1111/j.1467-8640.2009.00347.x/abstract;jsessionid=C95A98B67CD44AF9ABF59B3B0CCAA979.f01t02?userIsAuthenticated=false\&deniedAccessCustomisedMessage=}, author = {Margaritis, Dimitris and Bromberg, Facundo} } @article {171, title = {Efficient Markov network structure discovery using independence tests}, journal = {Journal of Artificial Intelligence Research}, volume = {35}, year = {2009}, pages = {449{\textendash}484}, abstract = {We present two algorithms for learning the structure of a Markov network from data: GSMN* and GSIMN. Both algorithms use statistical independence tests to infer the structure by successively constraining the set of structures consistent with the results of these tests. Until very recently, algorithms for structure learning were based on maximum likelihood estimation, which has been proved to be NP-hard for Markov networks due to the difficulty of estimating the parameters of the network, needed for the computation of the data likelihood. The independence-based approach does not require the computation of the likelihood, and thus both GSMN* and GSIMN can compute the structure efficiently (as shown in our experiments). GSMN* is an adaptation of the Grow-Shrink algorithm of Margaritis and Thrun for learning the structure of Bayesian networks. GSIMN extends GSMN* by additionally exploiting Pearls well-known properties of the conditional independence relation to infer novel independences from known ones, thus avoiding the performance of statistical tests to estimate them. To accomplish this efficiently GSIMN uses the Triangle theorem, also introduced in this work, which is a simplified version of the set of Markov axioms. Experimental comparisons on artificial and real-world data sets show GSIMN can yield significant savings with respect to GSMN*, while generating a Markov network with comparable or in some cases improved quality. We also compare GSIMN to a forward-chaining implementation, called GSIMN-FCH, that produces all possible conditional independences resulting from repeatedly applying Pearls theorems on the known conditional independence tests. The results of this comparison show that GSIMN, by the sole use of the Triangle theorem, is nearly optimal in terms of the set of independences tests that it infers.
}, doi = { 10.1613/jair.2773}, url = {http://www.jair.org/papers/paper2773.html}, author = {Bromberg, Facundo and Margaritis, Dimitris and Honavar, Vasant} } @article {219, title = {Ensemble of shape functions and support vector machines for the estimation of discrete arm muscle activation from external biceps 3D point clouds}, journal = {Computers in Biology and Medicine}, volume = {95}, year = {2018}, month = {04/2018}, pages = {129-139}, chapter = {129}, abstract = {Background: Muscle activation level is currently being captured using im- practical and expensive devices which make their use in telemedicine settings extremely difficult. To address this issue, a prototype is presented of a non- invasive, easy-to-install system for the estimation of a discrete level of muscle activation of the biceps muscle from 3D point clouds captured with RGB-D cameras.
Methods: A methodology is proposed that uses the ensemble of shape functions point cloud descriptor for the geometric characterization of 3D point clouds, together with support vector machines to learn a classifier that, based on this geometric characterization for some points of view of the biceps, provides a model for the estimation of muscle activation for all neighboring points of view. This results in a classifier that is robust to small perturba- tions in the point of view of the capturing device, greatly simplifying the installation process for end-users.
Results: In the discrimination of five levels of effort with values up to the maximum voluntary contraction (MVC) of the biceps muscle (3800 g),\ \ the best variant of the proposed methodology achieved mean absolute errors of about 9.21 \% MVC {\textemdash} an acceptable performance for telemedicine settings where the electric measurement of muscle activation is impractical.
Conclusions: The results prove that the correlations between the exter- nal geometry of the arm and biceps muscle activation are strong enough to consider computer vision and supervised learning an alternative with great potential for practical applications in tele-physiotherapy.\
In viticulture, there are several applications where 3D bud detection and localization in vineyards is a necessary task susceptible to automation: measurement of sunlight exposure, autonomous pruning, bud counting, type-of-bud classification, bud geometric characterization, internode length, and bud development stage.\ This paper presents a workflow to achieve quality 3D localizations of grapevine buds based on well-known computer vision and machine learning algorithms when provided with images captured in natural field conditions (i.e., natural sunlight and the addition of no artificial elements), during the winter season and using a mobile phone RGB camera. Our pipeline combines the Oriented FAST and Rotated BRIEF (ORB) for keypoint detection, a Fast Local Descriptor for Dense Matching (DAISY) for describing the keypoint, and the Fast Approximate Nearest Neighbor (FLANN) technique for matching keypoints, with the Structure from Motion multi-view scheme for generating consistent 3D point clouds. Next, it uses a 2D scanning window classifier based on Bag of Features and Support Vectors Machine for classification of 3D points in the cloud. Finally, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) for 3D bud localization is applied. Our approach resulted in a maximum precision\ of 1.0\ (i.e., no false detections), a maximum recall\ of 0.45\ (i.e. 45\%\ of the buds detected), and a localization error within the range of 259-554\ pixels (corresponding to approximately 3\ bud diameters, or 1.5cm) when evaluated over the whole range of user-given parameters of workflow components.
}, keywords = {computer vision, Grapevine bud detection, Precision viticulture}, issn = {0166-3615}, doi = {10.1016/j.compind.2018.03.033}, url = {https://www.sciencedirect.com/science/article/pii/S0166361517304815}, author = {Diaz, Carlos Ariel and P{\'e}rez, Diego Sebasti{\'a}n and Miatello, Humberto and Bromberg, Facundo} } @article {169, title = {Guest Editorial: 10th Argentinean Symposium on Artificial Intelligence (ASAI 2009)}, journal = {Inteligencia Artificial.}, volume = {13}, year = {2009}, pages = {4}, chapter = {3}, issn = { 1137-3601}, doi = {0.4114/ia.v13i44.1040}, url = {http://www.redalyc.org/articulo.oa?id=92513154001}, author = {Berd{\'u}n, Luis and Bromberg, Facundo} } @article {69, title = {The IBMAP approach for Markov network structure learning}, journal = {Annals of Mathematics and Artificial Intelligence}, volume = {72}, year = {2014}, month = {04/2014}, pages = {197--223}, chapter = {197}, abstract = {In this work we consider the problem of learning the structure of Markov networks from data. We present an approach for tackling this problem called IBMAP, together with an efficient instantiation of the approach: the IBMAP-HC algorithm, designed for avoiding important limitations of existing independence-based algorithms. These algorithms proceed by performing statistical independence tests on data, trusting completely the outcome of each test. In practice tests may be incorrect, resulting in potential cascading errors and the consequent reduction in the quality of the structures learned. IBMAP contemplates this uncertainty in the outcome of the tests through a probabilistic maximum-a-posteriori approach. The approach is instantiated in the IBMAP-HC algorithm, a structure selection strategy that performs a polynomial heuristic local search in the space of possible structures. We present an extensive empirical evaluation on synthetic and real data, showing that our algorithm outperforms significantly the current independence-based algorithms, in terms of data efficiency and quality of learned structures, with equivalent computational complexities. We also show the performance of IBMAP-HC in a real-world application of knowledge discovery: EDAs, which are evolutionary algorithms that use structure learning on each generation for modeling the distribution of populations. The experiments show that when IBMAP-HC is used to learn the structure, EDAs improve the convergence to the optimum.
}, keywords = {68T05, EDAs, Independence tests, Knowledge discovery, Markov network, Structure learning}, issn = {1012-2443}, doi = {10.1007/s10472-014-9419-5}, url = {http://dx.doi.org/10.1007/s10472-014-9419-5}, author = {Schl{\"u}ter, Federico and Bromberg, Facundo and Edera, Alejandro} } @article {210, title = {Image classification for detection of winter grapevine buds in natural conditions using scale-invariant features transform, bag of features and support vector machines}, journal = {Computers and Electronics in Agriculture}, volume = {135}, year = {2017}, month = {04/2017}, pages = {81-95}, chapter = {81}, abstract = {In viticulture, there are several applications where bud detection in vineyard images is a necessary task, susceptible of being automated through the use of computer vision methods. A common and effective family of visual detection algorithms are the scanning-window type, that slide a (usually) fixed size window along the original image, classifying each resulting windowed-patch as containing or not containing the target object. The simplicity of these algorithms finds its most challenging aspect in the classification stage. Interested in grapevine buds detection in natural field conditions, this paper presents a classification method for images of grapevine buds ranging 100 to 1600 pixels in diameter, captured in outdoor, under natural field conditions, in winter (i.e., no grape bunches, very few leaves, and dormant buds), without artificial background, and with minimum equipment requirements. The proposed method uses well-known computer vision technologies: Scale-Invariant Feature Transform for calculating low-level features, Bag of Features for building an image descriptor, and Support Vector Machines for training a classifier. When evaluated over images containing buds of at least 100 pixels in diameter, the approach achieves a recall higher than 0.9 and a precision of 0.86 over all windowed-patches covering the whole bud and down to 60\% of it, and scaled up to window patches containing a proportion of 20\%-80\% of bud versus background pixels. This robustness on the position and size of the window demonstrates its viability for use as the classification stage in a scanning-window detection algorithms.
}, keywords = {computer vision, Grapevine bud, Image classification, Precision viticulture, Scanning-window detection}, issn = {0168-1699}, doi = {10.1016/j.compag.2017.01.020}, url = {http://www.sciencedirect.com/science/article/pii/S0168169916301818}, author = {P{\'e}rez, Diego Sebasti{\'a}n and Bromberg, Facundo and Diaz, Carlos Ariel} } @article {172, title = {Improving the reliability of causal discovery from small data sets using argumentation}, journal = {The Journal of Machine Learning Research}, volume = {10}, year = {2009}, month = {02/2009}, pages = {301{\textendash}340}, abstract = {We address the problem of improving the reliability of independence-based causal discovery algorithms that results from the execution of statistical independence tests on small data sets, which typically have low reliability. We model the problem as a knowledge base containing a set of independence facts that are related through Pearl{\textquoteright}s well-known axioms. Statistical tests on finite data sets may result in errors in these tests and inconsistencies in the knowledge base. We resolve these inconsistencies through the use of an instance of the class of defeasible logics called argumentation, augmented with a preference function, that is used to reason about and possibly correct errors in these tests. This results in a more robust conditional independence test, called an argumentative independence test. Our experimental evaluation shows clear positive improvements in the accuracy of argumentative over purely statistical tests. We also demonstrate significant improvements on the accuracy of causal structure discovery from the outcomes of independence tests both on sampled data from randomly generated causal models and on real-world data sets.
}, keywords = {argumentation, causal Bayesian netw orks, Independence-based causal discovery, reliability improvement, Structure learning}, url = {http://www.jmlr.org/papers/v10/bromberg09a.html}, author = {Bromberg, Facundo and Margaritis, Dimitris} } @article {91, title = {Learning Markov networks networks with context-specific independences.}, journal = { International Journal on Artificial Intelligence Tools}, volume = {23}, year = {2014}, month = {12/2014}, chapter = {1460030}, abstract = {This work focuses on learning the structure of Markov networks from data. Markov networks are parametric models for compactly representing complex probability distributions. These models are composed by: a structure and numerical weights, where the structure describes independences that hold in the distribution. Depending on which is the goal of structure learning, learning algorithms can be divided into: density estimation algorithms, where structure is learned for answering inference queries; and knowledge discovery algorithms, where structure is learned for describing independences qualitatively. The latter algorithms present an important limitation for describing independences because they use a single graph; a coarse grain structure representation which cannot
represent flexible independences. For instance, context-specific independences cannot be described by a single graph. To overcome this limitation, this work proposes a new alternative representation named canonical model as well as the CSPC algorithm; a novel knowledge discovery algorithm for learning canonical models by using context-specific independences as constraints. On an extensive empirical evaluation, CSPC learns more accurate structures than state-of-the-art density estimation and knowledge discovery algorithms. Moreover, for answering inference queries, our approach obtains competitive results against density estimation algorithms, significantly outperforming knowledge discovery algorithms.
In 2013, 85\% of the peach production in the Mendoza region (Argentina) was lost because of frost. In a couple of hours, farmers can lose everything. Handling a frost event is possible, but it is hard to predict when it is going to happen. The goal of the PEACH project is to predict frost events by analyzing measurements from sensors deployed around an orchard. This article provides an in-depth description of a complete solution we designed and deployed: the low-power wireless network and the back-end system. The low-power wireless network is composed entirely of commercial o-the-shelf devices. We develop a methodology for deploying the network and present the open-source tools to assist with the deployment and to monitor the network. The deployed low-power wireless mesh network is 100\% reliable, with end-to-end latency below 2 s, and over 3 years of battery lifetime. This article discusses how the technology used is the right one for precision agriculture applications.
}, doi = {http://dx.doi.org/10.4108/eai.1-12-2016.151711}, author = {Watteyne, Thomas and Diedrichs, Ana Laura and Brun-Laguna, Keoma and Chaar, Javier Emilio and Dujovne, Diego and Taffernaberry, Juan Carlos and Mercado, Gustavo} } @article {242, title = {Prediction of frost events using machine learning and IoT sensing devices}, journal = {IEEE Internet of Things Journal}, volume = {5}, year = {2018}, month = {27 August 2018}, pages = { 4589-4597}, type = {Full}, chapter = {4589}, abstract = {Internet of Things (IoT) in agriculture applications have evolved to solve several relevant problems from producers. Here, we describe a component of an IoT-enabled frost prediction system. We follow current approaches for prediction that use machine learning algorithms trained by past readings of temperature and humidity sensors to predict future temperatures. However, contrary to current approaches, we assume that the surrounding thermodynamical conditions are informative for prediction. For that, a model was developed for each location, including in its training information of sensor readings of all other locations, autonomously selecting the most relevant ones (algorithm dependent). We evaluated our approach by training regression and classification models using several machine learning algorithms, many already proposed in the literature for the frost prediction problem, over data from five meteorological stations spread along the Mendoza Province of Argentina. Given the scarcity of frost events, data was augmented using the synthetic minority oversampling technique (SMOTE). The experimental results show that selecting the most relevant neighbors and training the models with SMOTE reduces the prediction errors of both regression predictors for all five locations, increases the performance of Random Forest classification predictors for four locations while keeping it unchanged for the remaining one, and produces inconclusive results for logistic regression predictor. These results demonstrate the main claim of these works: that thermodynamic information of neighboring locations can be informative for improving both regression and classification predictions, but also are good enough to suggest that the present approach is a valid and useful resource for decision makers and producers.
}, author = {Diedrichs, Ana Laura and Bromberg, Facundo and Dujovne, Diego and Brun-Laguna, Keoma and Watteyne, Thomas} } @article {72, title = {A survey on independence-based Markov networks learning}, journal = {Artificial Intelligence Review}, volume = {42}, year = {2012}, month = {06/2012}, pages = {1093}, chapter = {1069}, abstract = {The problem of learning\ the Markov network structure\ from data\ has become increasingly important in machine learning,
and in many other application fields.\ Markov networks are probabilistic graphical models,\ a widely used formalism for handling probability distributions in intelligent systems.\ This document focuses on a technology called \emph{independence-based} learning,\ which allows for the learning of the independence structure of Markov networks from data in an efficient and sound manner,\ whenever the dataset is sufficiently large, and data is a representative sample of the target distribution.\ In the analysis of such technology, this work surveys the current state-of-the-art algorithms,\ discussing its limitations, and posing a series of open problems\ where future work may produce some advances in the area, in terms of quality and efficiency.
In Viticulture, visual inspection of the plant is a necessary task for measuring relevant variables. In many cases, these visual inspections are susceptible to automation through computer vision methods. Bud detection is one such visual task, central for the measurement of important variables such as: measurement of bud sunlight exposure, autonomous pruning, bud counting, type-of-bud classification, bud geometric characterization, internode length, bud area, and bud development stage, among others. This paper presents a computer method for grapevine bud detection based on a Fully Convolutional Networks MobileNet architecture (FCN-MN). To validate its performance, this architecture was compared in the detection task with a strong method for bud detection, Scanning Windows (SW) based on a patch classifier, showing improvements over three aspects of detection: segmentation, correspondence identification and localization. The best version of FCN-MN showed a detection F1-measure of 88.6\% (for true positives defined as detected components whose intersection-over-union with the true bud is above 0.5), and false positives that are small and near the true bud. Splits {\textendash}false positives overlapping the true bud{\textendash} showed a mean segmentation precision of 89.3\%(21.7), while false alarms {\textendash}false positives not overlapping the true bud{\textendash} showed a mean pixel area of only 8\% the area of a true bud, and a distance (between mass centers) of 1.1 true bud diameters. The paper concludes by discussing how these results for FCN-MN would produce sufficiently accurate measurements of bud variables such as bud number, bud area, and internode length, suggesting a good performance in a practical setup.
}, keywords = {computer vision, Fully convolutional network, Grapevine bud detection, Precision viticulture}, issn = {0168-1699}, doi = {https://doi.org/10.1016/j.compag.2020.105947}, url = {https://www.sciencedirect.com/science/article/pii/S0168169920331525}, author = {Wenceslao Villegas Marset and Diego Sebastian Perez and Carlos Ariel D{\'\i}az and Facundo Bromberg} } @article {212, title = {An analysis of the impact of Kikuchi approximations}, year = {2016}, abstract = {Factorizations of probabilistic distributions serve as condensed and efficient representations for modeling problems with uncertainty.
In probabilistic graphical models, it is generally assumed that factorizations are consistent with a graphical structure which simplifies the probabilistic structure by exploiting the conditional independences present in the distribution. In this work we focus con Markov networks, a subtype of graphical models where the graphical structure is an undirected graph, and the factorization is characterized over the maximal cliques of the graph, according with the well-known Hammersley-Clifford Theorem. \ When using that factorization, factors may correspond to the maximal cliques and overlappings of a junction tree. The semantics of a junction tree guarantees that if marginals are consistent, whatever the accuracy of the approximation, the factorization will produce a valid probability distribution (e.g. the sum of the values associated by the factorization to all possible states of the system will add to $1$). However, the number of junction trees, and thus the number of (consistent) factorizations that can be obtained from them is relatively small in comparison with all possible products of marginal distributions, including those not necessarily producing valid factorizations.
\ In this work we investigate, from different perspectives, the quality of marginal factorizations as approximations of probability distributions.\
\ In particular, our analysis is focused on the Clique-based Kikuchi approximations (CBKA), a particular type of factorization in probability marginals. In this type of factorization, the marginals are completely determined by the independence graph.\
\ Our general goal is to identify whether, and under which conditions, the CBKA can produce accurate or acceptable approximations of an original probability distribution.\
\ We show that the class of CBKA includes a much larger set of approximations than those derived from junction trees (or equivalently, chordal graphs).\
\ Additionally, we show that the quality of CBKA can be measured by using the Kendall tau rank distance, a measure which considers scenarios where the values produced by the factorizations are mainly relevant to compare or rank the configurations of the system. We analyzed the quality of CBKA in terms of the Kendall distance and also in terms of the Kullback-Liebler divergence, and show that both measures are correlated. Our current results show that such correlation depends on the structure of the underlying distribution, and also on the strength of the dependences, e.g., the hyper-parameters of Dirichlet for synthetic distributions.\
Trunk diameter is a variable of agricultural interest, used mainly in the prediction of fruit trees production. It is correlated with leaf area and biomass of trees, and consequently gives a good estimate of the potential production of the plants. This work presents a low cost, high precision method for the measurement of trunk diameter of grapevines based on Computer Vision techniques. Several methods based on Computer Vision and other techniques are introduced in the literature. These methods present different advantages for crop management: they are amenable to be operated by unknowledgeable personnel, with lower operational costs; they result in lower stress levels to knowledgeable personnel, avoiding the deterioration of the measurement quality over time; and they make the measurement process amenable to be embedded in larger autonomous systems, allowing more measurements to be taken with equivalent costs. To date, all existing autonomous methods are either of low precision, or have a prohibitive cost for massive agricultural adoption, leaving the manual Vernier caliper or tape measure as the only choice in most situations. In this work we present a semi-autonomous measurement method that is susceptible to be fully automated, cost effective for mass adoption, and its precision is competitive (with slight improvements) over the caliper manual method.
}, keywords = {computer vision, Gaussian mixture model, Grapevine, Measurements, Trunk Diameter}, issn = {arXiv:1406.4845v2}, url = {http://arxiv.org/abs/1406.4845}, author = {P{\'e}rez, Diego Sebasti{\'a}n and Bromberg, Facundo and Gonzalez-Antivilo, Francisco} } @article {76, title = {Frontiers in Computer Vision: NSF White Paper}, year = {2010}, institution = {Frontiers in Computer Vision}, address = {Cambridge, EEUU}, abstract = {Markov random fields provide a compact representation of joint probability distributions by representing its independence properties in an undirected graph. The well-known Hammersley-Clifford theorem uses these conditional independences to factorize a Gibbs distribution into a set of factors. However, an important issue of using a graph to represent independences is that it cannot encode some types of independence relations, such as the context-specific independences (CSIs). They are a particular case of conditional independences that is true only for a certain assignment of its conditioning set; in contrast to conditional independences that must hold for all its assignments. This work presents a method for factorizing a Markov random field according to CSIs present in a distribution, and formally guarantees that this factorization is correct. This is presented in our main contribution, the context-specific Hammersley-Clifford theorem, a generalization to CSIs of the Hammersley-Clifford theorem that applies for conditional independences.
}, issn = {arXiv:1306.2295}, author = {Edera, Alejandro and Bromberg, Facundo and Schl{\"u}ter, Federico} } @article {227, title = {SDCP: Streaming Distributed Control Protocol}, year = {2016}, url = {https://datatracker.ietf.org/doc/draft-perez-dispatch-sdcp/}, author = {Cristian Perez and Ana Laura Diedrichs} } @mastersthesis {207, title = {Aprendizaje de redes de Markov basado en independencias espec{\'\i}ficas del contexto: formalizaci{\'o}n y aplicaci{\'o}n a datos espaciales}, year = {2016}, author = {Edera, Alejandro} } @mastersthesis {214, title = {Dise{\~n}o de Algoritmos de Aprendizajes de M{\'a}quinas y Visi{\'o}n Computacional para la Medici{\'o}n Aut{\'o}noma de Variables Vit{\i}́colas}, year = {2017}, author = {Diego Sebastian Perez} } @mastersthesis {161, title = {El enfoque IBMAP para aprendizaje de estructuras de redes de Markov}, volume = {Doctorado en Ciencias de la Computaci{\'o}n (PhD in Computer Science)}, year = {2014}, month = {11/2014}, pages = {138}, type = {Doctorado}, abstract = {Las redes de Markov\ son modelos probabil{\'\i}sticos gr{\'a}ficos,\ una herramienta computacional para el modelado eficiente de distribuciones de probabilidad\ que utiliza grafos para facilitar la representaci{\'o}n de problemas complejos.\ Estos modelos han sido dise{\~n}ados para ser manipulados por sistemas expertos,\ utilizando la teor{\'\i}a de la probabilidad para razonar eficientemente bajo condiciones de incertidumbre.\ Sin embargo, una limitaci{\'o}n importante para el uso de estos modelos es que en la pr{\'a}ctica\ resulta complejo dise{\~n}arlos manualmente, ya que el conocimiento de expertos no siempre es suficiente,\ sumado al hecho de que muchos dominios reales poseen una gran dimensionalidad.\ Por esto, el aprendizaje\ de estos modelos a partir de datos\ es un t{\'o}pico que ha tomado gran relevancia, ya que la disponibilidad de datos digitales es cada vez mayor,\ resultando adem{\'a}s en un mecanismo interesante para descubrir nuevo conocimiento a partir de datos digitales.
En este trabajo, la investigaci{\'o}n se centra en un enfoque espec{\'\i}fico de\ aprendizaje de redes de Markov:\ los algoritmos basados en independencia.\ Estos algoritmos est{\'a}n dise{\~n}ados para aprender la estructura de independencias del modelo,\ que es la componente que codifica de un modo compacto el conocimiento sobre la distribuci{\'o}n.\ Los algoritmos basados en independencia han sido dise{\~n}ados bajo lineamientos te{\'o}ricos\ que permiten aprender la estructura de un modo eficiente y robusto, a partir de\ la ejecuci{\'o}n de un conjunto de tests estad{\'\i}sticos de independencia\ sobre los datos.\ Com{\'u}nmente, los resultados de dichos tests son utilizados como restricciones\ que gu{\'\i}an una b{\'u}squeda en el espacio de las estructuras de independencia posibles,\ convergiendo a una estructura que satisface los resultados de todos los tests.\ Estos algoritmos garantizan que la estructura aprendida es correcta\ bajo la suposici{\'o}n de que los tests estad{\'\i}sticos son confiables.\ No obstante, un hecho muy com{\'u}n en la pr{\'a}ctica suele ser que los datos disponibles no son suficientes para obtener resultados correctos desde los tests estad{\'\i}sticos.\ Cuando esto sucede, los algoritmos basados en independencia\ acumulan y propagan suposiciones de\ independencia incorrectas,\ resultando en un aprendizaje con gran cantidad de errores estructurales.\ Este hecho, que resulta en una limitaci{\'o}n de real importancia\ a la hora de aplicar esta tecnolog{\'\i}a,\ es la motivaci{\'o}n de la presente tesis.
En este trabajo se presenta el enfoque de m{\'a}ximo a posteriori basado en independencias\ para aprendizaje de estructuras de redes de Markov (IBMAP, del ingl{\'e}s independence-based maximum-a-posteriori).\ Dado que los algoritmos tradicionales descartan la estructura correcta cada vez que ejecutan\ un test err{\'o}neo, se propone un enfoque que asigna probabilidades a las distintas estructuras,\ sin descartar ninguna. Para esto, se propone una funci{\'o}n de puntaje de estructuras basada en tests estad{\'\i}sticos denominada IB-score\ (puntaje basado en independencias).\ Esta funci{\'o}n permite computar de un modo aproximado la probabilidad a posteriori\ de una estructura dados los datos Pr(G|D), combinando los resultados de un conjunto de tests estad{\'\i}sticos.\ De este modo, las diferentes estructuras poseen un puntaje m{\'a}s alto o m{\'a}s bajo seg{\'u}n\ las probabilidades de las independencias que codifican.\ En resumen, el enfoque propuesto consiste en la maximizaci{\'o}n de la funci{\'o}n IB-score sobre el espacio de todas las estructuras posibles.\ A modo de instanciaci{\'o}n de este enfoque se presenta una srie de algoritmos que maximizan dicha funci{\'o}n\ con diversos mecanismos de optimizaci{\'o}n.\
Para validar el enfoque se evalu{\'o} el desempe{\~n}o de las distintas instanciaciones de la b{\'u}squeda,\ evaluando la calidad de las estructuras aprendidas con respecto\ a algoritmos basados en independencia que pertenecen al estado del arte.\ Se presentan resultados sistem{\'a}ticos sobre una gran variedad de dimensiones del problema,\ demostrando que este enfoque permite mejorar significativamente la calidad de las estructuras aprendidas.\ Se demuestra tambi{\'e}n que dichas mejoras pueden obtenerse a un costo computacional\ competitivo respecto de los algoritmos del estado del arte.\ Dicha experimentaci{\'o}n fue llevada a cabo sobre datos sint{\'e}ticos y datos del mundo real, y en una aplicaci{\'o}n de aprendizaje de estructuras para algoritmos evolutivos.
}, author = {Schl{\"u}ter, Federico and Bromberg, Facundo} } @mastersthesis {175, title = {Markov networks structure discovery using independence tests}, volume = {Doctor of Philosophy}, year = {2007}, pages = {182}, school = {Iowa State University}, address = {Ames, IA, USA}, abstract = {We investigate efficient algorithms for learning the structure of a Markov network from
data using the independence-based approach. Such algorithms conduct a series of conditional
independence tests on data, successively restricting the set of possible structures until there is
only a single structure consistent with the outcomes of the conditional independence tests exe-
cuted (if possible). As Pearl has shown, the instances of the conditional independence relation
in any domain are theoretically interdependent, made explicit in his well-known conditional
independence axioms. The first couple of algorithms we discuss, GSMN and GSIMN, exploit
Pearl{\textquoteright}s independence axioms to reduce the number of tests required to learn a Markov network.
This is useful in domains where independence tests are expensive, such as cases of very large
data sets or distributed data. Subsequently, we explore how these axioms can be exploited to
{\textquotedblleft}correct{\textquotedblright} the outcome of unreliable statistical independence tests, such as in applications where
little data is available. We show how the problem of incorrect tests can be mapped to inference
in inconsistent knowledge bases, a problem studied extensively in the field of non-monotonic
logic. We present an algorithm for inferring independence values based on a sub-class of non-
monotonic logics: the argumentation framework. Our results show the advantage of using our
approach in the learning of structures, with improvements in the accuracy of learned networks
of up to 20\%. As an alternative to logic-based interdependence among independence tests,
we also explore probabilistic interdependence. Our algorithm, called PFMN, takes a Bayesian
particle filtering approach, using a population of Markov network structures to maintain the
posterior probability distribution over them given the outcomes of the tests performed. The
result is an approximate algorithm (due to the use of particle filtering) that is useful in domains
where independence tests are expensive.
Este trabajo de tesis complementa los requisitos de la carrera de Doctorado en Ciencias de la Computaci{\'o}n de la Facultad de Ciencias Exactas de la Universidad Nacional del Centro de la Provincia de Buenos Aires. El problema general en el que se encuentra enmarcada la presente tesis es el de estimar el nivel de activaci{\'o}n muscular ejecutado por el b{\'\i}ceps de un brazo humano al ejercer diferentes niveles de esfuerzos discretos, por medio de un sistema de visi{\'o}n artificial alimentado exclusivamente con im{\'a}genes externas del brazo. El sistema mencionado tiene embebido un componente de aprendizaje de m{\'a}quinas supervisado que le permite predecir la medida deseada, a partir de ejemplos conocidos de la activaci{\'o}n real. En un primer paso del m{\'e}todo, se aplican t{\'e}cnicas de visi{\'o}n computacional para la descripci{\'o}n de im{\'a}genes de manera de generar vectores de caracter{\'\i}sticas de las mismas. Luego, estos vectores son usados como entrada en un proceso de aprendizaje de m{\'a}quinas supervisado por la fuerza externa aplicada, un valor altamente correlacionado al nivel de activaci{\'o}n real que el b{\'\i}ceps est{\'a} ejerciendo. Se evaluaron t{\'e}cnicas maduras y com{\'u}nmente usadas en estas {\'a}reas para las cuales existen implementaciones computacionales. Como respuesta a la pregunta de {\textquestiondown}por qu{\'e} es importante resolver el problema planteado?, se presenta una introducci{\'o}n al mismo y a las {\'a}reas disciplinares de aplicaci{\'o}n. Por otro lado, pretendiendo responder {\textquestiondown}qu{\'e} soluciones existen para el problema planteado?, se presenta una revisi{\'o}n bibliogr{\'a}fica de los trabajos relacionados al problema y el m{\'e}todo propuesto para resolverlo. Para responder a la pregunta de {\textquestiondown}qu{\'e} debo aprender para resolver el problema mediante el m{\'e}todo propuesto?, se presentan las tecnolog{\'\i}as de soporte evaluadas en esta tesis para resolver el problema usando el m{\'e}todo propuesto, puntualmente en las {\'a}reas de conocimiento de descripci{\'o}n de im{\'a}genes y aprendizaje de m{\'a}quinas supervisado. Finalmente para responder a la pregunta de {\textquestiondown}c{\'o}mo aplicar estas tecnolog{\'\i}as para resolver el problema?, se presentan distintas instancias del m{\'e}todo que pretenden resolver el problema de estimaci{\'o}n planteado. Una de las instancias del m{\'e}todo presentadas, logra obtener una precisi{\'o}n aceptable en la tarea de medici{\'o}n de la activaci{\'o}n en un setting lo suficientemente pr{\'a}ctico como para ser usado en ciertos ejercicios de tele-rehabilitaci{\'o}n. La precisi{\'o}n en la medici{\'o}n de la activaci{\'o}n as{\'\i} como la correlaci{\'o}n estimada entre las mediciones obtenidas con este m{\'e}todo y la real, son comparables a los m{\'e}todos actuales y a trabajos similares.\
article sent, status: under review
}, author = {Keoma Brun-Laguna and Ana Laura Diedrichs and Thomas Watteyne and Diego Dujovne and Carlos Taffernaberry and Remy Leone and Xavier Vilajosana} } @article {37, title = {Computer Vision Online}, year = {2014}, url = {http://www.computervisiononline.com/} } @article {121, title = {CV Resources}, year = {2014}, abstract = {Computer Vision and Machine Learning resources with a focus on 3D scene understanding, parsing and reconstruction.
}, url = {http://www.cvlibs.net/links.php}, author = {Geiger, Andreas} } @article {38, title = {CVonline}, year = {2014}, url = {http://homepages.inf.ed.ac.uk/rbf/CVonline/} } @article {39, title = {CVPapers}, year = {2014}, keywords = {computer vision, link, online, papers, resources}, url = {http://www.cvpapers.com/} } @article {40, title = {ImageNets}, year = {2010}, url = {http://imagenets.sourceforge.net/} } @article {71, title = {Point Cloud Library (PCL)}, year = {2011}, abstract = {The Point Cloud Library (PCL) is a standalone, large scale, open project for 2D/3D image and point cloud processing.
PCL is released under the terms of the BSD license, and thus free for commercial and research use. We are financially supported by a consortium of commercial companies, with our own non-profit organization, Open Perception.
}, keywords = {2D image, 3D image, PCL, point cloud, Point Cloud Library}, url = {http://pointclouds.org/} }