Bruno Caprile | Bruno Kessler Foundation (original) (raw)
Papers by Bruno Caprile
Journal of the Optical Society of America A, 1988
Two schemes for edge detection of real images based on gradient maxima are presented. Images are ... more Two schemes for edge detection of real images based on gradient maxima are presented. Images are filtered with narrow filters to increase localization. Experimental results and theoretical considerations suggest that the exact shape of the filter is not critical for good performance of the algorithm. Therefore a filter can be chosen to allow for a highly efficient hardware implementation, for example, a binary filter or a 4-bit finite-impulse-response filter. Because the digitized values of a binary filter are powers of 2, the hardware implementation does not require time-consuming computations, such as multiplication and time shift, but just appropriate addressings. The performance of this scheme, or a similar scheme using 4-bit filters, is as satisfactory as that of more sophisticated schemes. Therefore these low-cost schemes are likely to be more suitable for hardware implementation.
MAIA is the project aimed at the integration of the AI resources presently operating at IRST. The... more MAIA is the project aimed at the integration of the AI resources presently operating at IRST. The overall approach to the design of intelligent artificial systems that MAIA proposes is experimental not less than theoretical, and an experimental setup ( the experimental platform ) has consequently been defined in which a variety of mutually interacting functionalities can be tested in a common framework. Here, the overall architecture of the system is outlined, and one of the platform’s components — the Mobile Robot — is presented. Potentialities arising from the combined use of speech and autonomous navigation are also considered and discussed.
Proceedings of SPIE, May 4, 1993
MAIA (acronym for Modello Avanzato di Intelligenza Artificiale) is a project aimed at the integra... more MAIA (acronym for Modello Avanzato di Intelligenza Artificiale) is a project aimed at the integration of several AI resources being presently developed at IRST. The overall approach to the design of intelligent artificial systems that MAIA proposes is experimental not less than theoretical, and an experimental setup (that we call the experimental platform) has consequently been defined in which a variety of mutually interacting functionalities can be tested in a common framework. The experimental platform of MAIA consists of 'brains' and 'tentacles', and it is to one of such tentacles--the Mobile Robot--that the present paper is devoted to. At present, the mobile robot is equipped with two main modules: the Navigation Module, which gives the robot the capability of moving autonomously in an indoor environment, and the Speech Module, which allows the robot to communicate with humans by voice. Here the overall architecture of the system is described in detail and potentialities arising from the combined use of speech and navigation are considered.
ABSTRACT In this work a novel analysis methodology of SVMs optimal solutions is presented. Such a... more ABSTRACT In this work a novel analysis methodology of SVMs optimal solutions is presented. Such a methodology is based on a multiobjective optimization algorithm which exploits a genetic search paradigm. The application field is the design of smart microsensors, where both classification performance and complexity criteria have to be considered in order to balance accuracy and power consumption requirements
International Journal of Pattern Recognition and Artificial Intelligence, Aug 1, 2004
Computational Statistics & Data Analysis, Feb 1, 2007
AdaBoost is one of the most popular classification methods. In contrast to other ensemble methods... more AdaBoost is one of the most popular classification methods. In contrast to other ensemble methods (e.g., Bagging) the AdaBoost is inherently sequential. In many data intensive real-world situations this may limit the practical applicability of the method. P-AdaBoost is a novel scheme for the parallelization of AdaBoost, which builds upon earlier results concerning the dynamics of AdaBoost weights. P-AdaBoost yields approximations to the standard AdaBoost models that can be easily and efficiently distributed over a network of computing nodes. Properties of P-AdaBoost as a stochastic minimizer of the AdaBoost cost functional are discussed. Experiments are reported on both synthetic and benchmark data sets.
International Journal of Computer Vision, Mar 1, 1990
In this article a new method for the calibration of a vision system which consists of two (or mor... more In this article a new method for the calibration of a vision system which consists of two (or more) cameras is presented. The proposed method, which uses simple properties of vanishing points, is divided into two steps. In the first step, the intrinsic parameters of each camera, that is, the focal length and the location of the intersection between the optical axis and the image plane, are recovered from a single image of a cube. In the second step, the extrinsic parameters of a pair of cameras, that is, the rotation matrix and the translation vector which describe the rigid motion between the coordinate systems fixed in the two cameras, are estimated from an image stereo pair of a suitable planar pattern. Firstly, by matching the corresponding vanishing points in the two images the rotation matrix can be computed, then the translation vector is estimated by means of a simple triangulation. The robustness of the method against noise is discussed, and the conditions for optimal estimation of the rotation matrix are derived. Extensive experimentation shows that the precision that can be achieved with the proposed method is sufficient to efficiently perform machine vision tasks that require camera calibration, like depth from stereo and motion from image sequence.
Springer eBooks, 1995
Over the years, automated vehicles have evolved from reliable yet rigidly constrained AGVs to the... more Over the years, automated vehicles have evolved from reliable yet rigidly constrained AGVs to the fairly flexible ones of the Help Mate generation. It is interesting to observe that while such systems exhibit quite respectful autonomous navigation capabilities, their capacity of interacting with the users and the environment is still limited.
Robotics and Autonomous Systems, Nov 1, 1997
A robotic system is presented, which is able to autonomously explore unengineered indoor environm... more A robotic system is presented, which is able to autonomously explore unengineered indoor environments, thereby synthesising maps suitable for planning and navigation purposes. Map recovery takes place through interaction between the robot and the world, in which either sensing and acting are a ected by uncertainty. Kalman ltering is applied to maintain position best estimates, which are then fused with data coming from the observation of landmarks. The proposed method has been implemented on a robot equipped with ultrasonic range nders, and tested in a fairly simple, real environment.
Springer eBooks, 2002
The dynamical evolution of weights in the AdaBoost algorithm contains useful information about th... more The dynamical evolution of weights in the AdaBoost algorithm contains useful information about the rôle that the associated data points play in the built of the AdaBoost model. In particular, the dynamics induces a bipartition of the data set into two (easy/hard) classes. Easy points are ininfluential in the making of the model, while the varying relevance of hard points can be gauged in terms of an entropy value associated to their evolution. Smooth approximations of entropy highlight regions where classification is most uncertain. Promising results are obtained when methods proposed are applied in the Optimal Sampling framework.
IEEE Transactions on Instrumentation and Measurement, Sep 1, 2008
In this paper, we face the problem of designing accurate decision-making modules in measurement s... more In this paper, we face the problem of designing accurate decision-making modules in measurement systems that need to be implemented on resource-constrained platforms. We propose a methodology based on multiobjective optimization and genetic algorithms (GAs) for the analysis of support vector machine (SVM) solutions in the classification error-complexity space. Specific criteria for the choice of optimal SVM classifiers and experimental results on both real and synthetic data will also be discussed.
Neural Information Processing Systems, Oct 1, 1990
Learning an input-output mapping from a set of examples can be regarded as synthesizing an approx... more Learning an input-output mapping from a set of examples can be regarded as synthesizing an approximation of a multi-dimensional function. From this point of view, this form of learning is closely related to regularization theory, and we have previously shown (Poggio and Girosi, 1990a, 1990b) the equivalence between reglilari~at.ioll and a. class of three-layer networks that we call regularization networks. In this note, we ext.end the theory by introducing ways of <lealing with t.wo aspect.s of learning: learning in presence of unreliable examples or outliel•s, an<llearning from positive and negative examples.
Journal of Theoretical Biology, Sep 1, 2008
Sequences of epidemic waves have been observed in past influenza pandemics, such as the Spanish i... more Sequences of epidemic waves have been observed in past influenza pandemics, such as the Spanish influenza. Possible explanations may be sought either in mechanisms altering the structure of the network of contacts, such as those induced by changes in the rates of movement of people or by public health measures, or in the genetic drift of the influenza virus, since the appearance of new strains can reduce or eliminate herd immunity. The pandemic outbreaks may also be influenced by coinfection with other acute respiratory infections (ARI) that increase transmissibility of influenza virus (by coughing, sneezing, running nose). In fact, some viruses (e.g., Rhinovirus and Adenovirus) have been found to induce ''clouds'' of bacteria and increase the transmissibility of Staphylococcus aureus. Moreover, Rhinovirus and Adenovirus were detected in patients during past pandemics, and their presence is linked to superspreading events. In this paper, by assuming increased transmissibility in coinfected individuals, we propose and study a model where multiple pandemic waves are triggered by coinfection with ARI. The model agrees well with mortality excess data during the 1918 pandemic influenza, thereby providing indications for potential pandemic mitigation.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 1989
... cos e F,, = ( A - B)xg(x) +(-Y) + (A + B)g(x) . g(y) sin 0 cos 0 - (Aiig(ii)+(-Z) + Bug(u) +(... more ... cos e F,, = ( A - B)xg(x) +(-Y) + (A + B)g(x) . g(y) sin 0 cos 0 - (Aiig(ii)+(-Z) + Bug(u) +(U)) sin2 e (2.5.42) F,, = ( A - B)g(x) g(r) COS* e + (Bug(u) +(U) - Aiig(ii) +( -Z)) sin e cos 0 (2.5.43 ) F,, = -(A + B)g(x) g(y) sin 19 cos 0 - (AZg(E)
Journal of Theoretical Biology, Sep 1, 2009
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service... more This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Springer eBooks, Nov 28, 2012
Uncoordinated human behavioral responses triggered by risk perception can alter the evolution of ... more Uncoordinated human behavioral responses triggered by risk perception can alter the evolution of an epidemic outbreak further and beyond control measures imposed by public authorities. In fact, spontaneous behavioral changes could develop as a defensive response during the spread of an epidemic, thereby impacting the epidemic dynamics and affecting timing and overall number of cases. In this chapter, a model coupling the classic SIR disease transmission model with an imitation dynamics process is introduced which accounts for the diffusion of different behaviors in the population as a response to the epidemic threat. A detailed analysis of the model identifies the main determinants leading to remarkable alterations in infection dynamics in both risk perception and diffusion of human behavioral patterns. Empirical evidence points to the need of incorporating human behavior in prediction models informing public health decisions.
Mathematical and Computer Modelling, Sep 1, 1994
An extension of the one-parameter class of functionals considered in the Smoothing Spline Estimat... more An extension of the one-parameter class of functionals considered in the Smoothing Spline Estimate framework is proposed to better deal with scaling problems or anisotropies of data to be approximated. In particular, a matrix, L, of regularizing parameters is introduced in place of the typical parameter X. By means of an appropriate coordinates transformation, the resulting variational problem can be rephrased in terms of the "traditional" Smoothing Spline Estimate method, and the Generalized Cross Validation technique can consequently be applied to find optimal values for L. Following this approach, numerical experiments have been performed on synthetic 2D data, and the results compared with those obtained with Smoothing Spline Estimate and Hyper Basis Function methods. *The research work of B.C. is done within MAIA, the leading AI project being presently developed at 1-T. The authors feel deeply indebted to L. Tubaro, for many crucial suggestions and the constant support provided. C. Furlanello and F. Girosi carefully read the manuscript and made several useful comments.
Learning an input-output mapping from a set of examples can be regarded as synthesizing an approx... more Learning an input-output mapping from a set of examples can be regarded as synthesizing an approximation of a multi-dimensional function. From this point of view, this form of learning is closely related to regularization theory. In this note, we extend the theory by introducing ways of dealing with two aspects of learning: learning in the presence of unreliable examples and learning from positive {\it and} negative examples. The first extension corresponds to dealing with outliers among the sparse data. The second one corresponds to exploiting information about points or regions in the range of the function that are forbidden.
arXiv (Cornell University), Nov 3, 2000
An algorithm is presented which, with optimal efficiency, solves the problem of uniform random ge... more An algorithm is presented which, with optimal efficiency, solves the problem of uniform random generation of distribution functions for an n-valued random variable.
Journal of the Optical Society of America A, 1988
Two schemes for edge detection of real images based on gradient maxima are presented. Images are ... more Two schemes for edge detection of real images based on gradient maxima are presented. Images are filtered with narrow filters to increase localization. Experimental results and theoretical considerations suggest that the exact shape of the filter is not critical for good performance of the algorithm. Therefore a filter can be chosen to allow for a highly efficient hardware implementation, for example, a binary filter or a 4-bit finite-impulse-response filter. Because the digitized values of a binary filter are powers of 2, the hardware implementation does not require time-consuming computations, such as multiplication and time shift, but just appropriate addressings. The performance of this scheme, or a similar scheme using 4-bit filters, is as satisfactory as that of more sophisticated schemes. Therefore these low-cost schemes are likely to be more suitable for hardware implementation.
MAIA is the project aimed at the integration of the AI resources presently operating at IRST. The... more MAIA is the project aimed at the integration of the AI resources presently operating at IRST. The overall approach to the design of intelligent artificial systems that MAIA proposes is experimental not less than theoretical, and an experimental setup ( the experimental platform ) has consequently been defined in which a variety of mutually interacting functionalities can be tested in a common framework. Here, the overall architecture of the system is outlined, and one of the platform’s components — the Mobile Robot — is presented. Potentialities arising from the combined use of speech and autonomous navigation are also considered and discussed.
Proceedings of SPIE, May 4, 1993
MAIA (acronym for Modello Avanzato di Intelligenza Artificiale) is a project aimed at the integra... more MAIA (acronym for Modello Avanzato di Intelligenza Artificiale) is a project aimed at the integration of several AI resources being presently developed at IRST. The overall approach to the design of intelligent artificial systems that MAIA proposes is experimental not less than theoretical, and an experimental setup (that we call the experimental platform) has consequently been defined in which a variety of mutually interacting functionalities can be tested in a common framework. The experimental platform of MAIA consists of 'brains' and 'tentacles', and it is to one of such tentacles--the Mobile Robot--that the present paper is devoted to. At present, the mobile robot is equipped with two main modules: the Navigation Module, which gives the robot the capability of moving autonomously in an indoor environment, and the Speech Module, which allows the robot to communicate with humans by voice. Here the overall architecture of the system is described in detail and potentialities arising from the combined use of speech and navigation are considered.
ABSTRACT In this work a novel analysis methodology of SVMs optimal solutions is presented. Such a... more ABSTRACT In this work a novel analysis methodology of SVMs optimal solutions is presented. Such a methodology is based on a multiobjective optimization algorithm which exploits a genetic search paradigm. The application field is the design of smart microsensors, where both classification performance and complexity criteria have to be considered in order to balance accuracy and power consumption requirements
International Journal of Pattern Recognition and Artificial Intelligence, Aug 1, 2004
Computational Statistics & Data Analysis, Feb 1, 2007
AdaBoost is one of the most popular classification methods. In contrast to other ensemble methods... more AdaBoost is one of the most popular classification methods. In contrast to other ensemble methods (e.g., Bagging) the AdaBoost is inherently sequential. In many data intensive real-world situations this may limit the practical applicability of the method. P-AdaBoost is a novel scheme for the parallelization of AdaBoost, which builds upon earlier results concerning the dynamics of AdaBoost weights. P-AdaBoost yields approximations to the standard AdaBoost models that can be easily and efficiently distributed over a network of computing nodes. Properties of P-AdaBoost as a stochastic minimizer of the AdaBoost cost functional are discussed. Experiments are reported on both synthetic and benchmark data sets.
International Journal of Computer Vision, Mar 1, 1990
In this article a new method for the calibration of a vision system which consists of two (or mor... more In this article a new method for the calibration of a vision system which consists of two (or more) cameras is presented. The proposed method, which uses simple properties of vanishing points, is divided into two steps. In the first step, the intrinsic parameters of each camera, that is, the focal length and the location of the intersection between the optical axis and the image plane, are recovered from a single image of a cube. In the second step, the extrinsic parameters of a pair of cameras, that is, the rotation matrix and the translation vector which describe the rigid motion between the coordinate systems fixed in the two cameras, are estimated from an image stereo pair of a suitable planar pattern. Firstly, by matching the corresponding vanishing points in the two images the rotation matrix can be computed, then the translation vector is estimated by means of a simple triangulation. The robustness of the method against noise is discussed, and the conditions for optimal estimation of the rotation matrix are derived. Extensive experimentation shows that the precision that can be achieved with the proposed method is sufficient to efficiently perform machine vision tasks that require camera calibration, like depth from stereo and motion from image sequence.
Springer eBooks, 1995
Over the years, automated vehicles have evolved from reliable yet rigidly constrained AGVs to the... more Over the years, automated vehicles have evolved from reliable yet rigidly constrained AGVs to the fairly flexible ones of the Help Mate generation. It is interesting to observe that while such systems exhibit quite respectful autonomous navigation capabilities, their capacity of interacting with the users and the environment is still limited.
Robotics and Autonomous Systems, Nov 1, 1997
A robotic system is presented, which is able to autonomously explore unengineered indoor environm... more A robotic system is presented, which is able to autonomously explore unengineered indoor environments, thereby synthesising maps suitable for planning and navigation purposes. Map recovery takes place through interaction between the robot and the world, in which either sensing and acting are a ected by uncertainty. Kalman ltering is applied to maintain position best estimates, which are then fused with data coming from the observation of landmarks. The proposed method has been implemented on a robot equipped with ultrasonic range nders, and tested in a fairly simple, real environment.
Springer eBooks, 2002
The dynamical evolution of weights in the AdaBoost algorithm contains useful information about th... more The dynamical evolution of weights in the AdaBoost algorithm contains useful information about the rôle that the associated data points play in the built of the AdaBoost model. In particular, the dynamics induces a bipartition of the data set into two (easy/hard) classes. Easy points are ininfluential in the making of the model, while the varying relevance of hard points can be gauged in terms of an entropy value associated to their evolution. Smooth approximations of entropy highlight regions where classification is most uncertain. Promising results are obtained when methods proposed are applied in the Optimal Sampling framework.
IEEE Transactions on Instrumentation and Measurement, Sep 1, 2008
In this paper, we face the problem of designing accurate decision-making modules in measurement s... more In this paper, we face the problem of designing accurate decision-making modules in measurement systems that need to be implemented on resource-constrained platforms. We propose a methodology based on multiobjective optimization and genetic algorithms (GAs) for the analysis of support vector machine (SVM) solutions in the classification error-complexity space. Specific criteria for the choice of optimal SVM classifiers and experimental results on both real and synthetic data will also be discussed.
Neural Information Processing Systems, Oct 1, 1990
Learning an input-output mapping from a set of examples can be regarded as synthesizing an approx... more Learning an input-output mapping from a set of examples can be regarded as synthesizing an approximation of a multi-dimensional function. From this point of view, this form of learning is closely related to regularization theory, and we have previously shown (Poggio and Girosi, 1990a, 1990b) the equivalence between reglilari~at.ioll and a. class of three-layer networks that we call regularization networks. In this note, we ext.end the theory by introducing ways of <lealing with t.wo aspect.s of learning: learning in presence of unreliable examples or outliel•s, an<llearning from positive and negative examples.
Journal of Theoretical Biology, Sep 1, 2008
Sequences of epidemic waves have been observed in past influenza pandemics, such as the Spanish i... more Sequences of epidemic waves have been observed in past influenza pandemics, such as the Spanish influenza. Possible explanations may be sought either in mechanisms altering the structure of the network of contacts, such as those induced by changes in the rates of movement of people or by public health measures, or in the genetic drift of the influenza virus, since the appearance of new strains can reduce or eliminate herd immunity. The pandemic outbreaks may also be influenced by coinfection with other acute respiratory infections (ARI) that increase transmissibility of influenza virus (by coughing, sneezing, running nose). In fact, some viruses (e.g., Rhinovirus and Adenovirus) have been found to induce ''clouds'' of bacteria and increase the transmissibility of Staphylococcus aureus. Moreover, Rhinovirus and Adenovirus were detected in patients during past pandemics, and their presence is linked to superspreading events. In this paper, by assuming increased transmissibility in coinfected individuals, we propose and study a model where multiple pandemic waves are triggered by coinfection with ARI. The model agrees well with mortality excess data during the 1918 pandemic influenza, thereby providing indications for potential pandemic mitigation.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 1989
... cos e F,, = ( A - B)xg(x) +(-Y) + (A + B)g(x) . g(y) sin 0 cos 0 - (Aiig(ii)+(-Z) + Bug(u) +(... more ... cos e F,, = ( A - B)xg(x) +(-Y) + (A + B)g(x) . g(y) sin 0 cos 0 - (Aiig(ii)+(-Z) + Bug(u) +(U)) sin2 e (2.5.42) F,, = ( A - B)g(x) g(r) COS* e + (Bug(u) +(U) - Aiig(ii) +( -Z)) sin e cos 0 (2.5.43 ) F,, = -(A + B)g(x) g(y) sin 19 cos 0 - (AZg(E)
Journal of Theoretical Biology, Sep 1, 2009
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service... more This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Springer eBooks, Nov 28, 2012
Uncoordinated human behavioral responses triggered by risk perception can alter the evolution of ... more Uncoordinated human behavioral responses triggered by risk perception can alter the evolution of an epidemic outbreak further and beyond control measures imposed by public authorities. In fact, spontaneous behavioral changes could develop as a defensive response during the spread of an epidemic, thereby impacting the epidemic dynamics and affecting timing and overall number of cases. In this chapter, a model coupling the classic SIR disease transmission model with an imitation dynamics process is introduced which accounts for the diffusion of different behaviors in the population as a response to the epidemic threat. A detailed analysis of the model identifies the main determinants leading to remarkable alterations in infection dynamics in both risk perception and diffusion of human behavioral patterns. Empirical evidence points to the need of incorporating human behavior in prediction models informing public health decisions.
Mathematical and Computer Modelling, Sep 1, 1994
An extension of the one-parameter class of functionals considered in the Smoothing Spline Estimat... more An extension of the one-parameter class of functionals considered in the Smoothing Spline Estimate framework is proposed to better deal with scaling problems or anisotropies of data to be approximated. In particular, a matrix, L, of regularizing parameters is introduced in place of the typical parameter X. By means of an appropriate coordinates transformation, the resulting variational problem can be rephrased in terms of the "traditional" Smoothing Spline Estimate method, and the Generalized Cross Validation technique can consequently be applied to find optimal values for L. Following this approach, numerical experiments have been performed on synthetic 2D data, and the results compared with those obtained with Smoothing Spline Estimate and Hyper Basis Function methods. *The research work of B.C. is done within MAIA, the leading AI project being presently developed at 1-T. The authors feel deeply indebted to L. Tubaro, for many crucial suggestions and the constant support provided. C. Furlanello and F. Girosi carefully read the manuscript and made several useful comments.
Learning an input-output mapping from a set of examples can be regarded as synthesizing an approx... more Learning an input-output mapping from a set of examples can be regarded as synthesizing an approximation of a multi-dimensional function. From this point of view, this form of learning is closely related to regularization theory. In this note, we extend the theory by introducing ways of dealing with two aspects of learning: learning in the presence of unreliable examples and learning from positive {\it and} negative examples. The first extension corresponds to dealing with outliers among the sparse data. The second one corresponds to exploiting information about points or regions in the range of the function that are forbidden.
arXiv (Cornell University), Nov 3, 2000
An algorithm is presented which, with optimal efficiency, solves the problem of uniform random ge... more An algorithm is presented which, with optimal efficiency, solves the problem of uniform random generation of distribution functions for an n-valued random variable.