The present paper presents the framework and the results of Project 2: "Multimodal tools and inte... more The present paper presents the framework and the results of Project 2: "Multimodal tools and interfaces for the intercommunication between visually impaired and "deaf and mute" people", which has been developed during the eNTERFACE-2006 summer workshop in the context of the SIMILAR NoE. The developed system aims to provide alternative tools and interfaces to blind and deaf-and-mute persons so as to enable their intercommunication as well as their interaction with the computer. All the involved technologies are integrated into a treasure hunting game application that is jointly played by the blind and deaf-and-mute user. The reason for choosing to integrate the multimodal interfaces into a game application is that it serves both as an entertainment and as a pleasant education tool to its users. The proposed application integrates haptics, audio, visual output as well as computer vision, sign language analysis and synthesis, speech recognition and synthesis, in order to provide an interactive environment where the blind and deaf and mute users can collaborate in order to play the treasure hunting game.
International Conference on Artificial Neural Networks, 2016
The Recursive-Rule eXtraction (Re-RX) algorithm family includes the Re-RX algorithm, the Re-RX al... more The Recursive-Rule eXtraction (Re-RX) algorithm family includes the Re-RX algorithm, the Re-RX algorithm with both discrete and continuous attributes (Continuous Re-RX [1]), the Re-RX algorithm with J48graft [2], Re-RX with J48graft combined with Sampling Selection Techniques (Sampling Re-RX with J48graft [4]), and the Re-RX algorithm with a trained neural network (Sampling Re-RX [3]). In this study, we compare the performance of the Re-RX algorithm family with various previous algorithms. One issue that always remains important in rule extraction is Pareto optimality, or in other words, an ideally balanced trade-off. In rule extraction, the trade-off is between the classification accuracy and interpretability of extracted rules. Our goal is to obtain a wider viable region for the Pareto optimal curve that will enable improvements in both the accuracy and interpretability of extracted rules. We vividly demonstrate Pareto-optimal curves between the accuracies and number of rules obtained for German and Australian datasets by 10 runs of 10-fold cross validation of the Re-RX algorithm family and those obtained using other algorithms. The Re-RX algorithm family has proven effective for extracting concise and interpretable rules from medical [1, 2, 4] and financial [3] datasets
This article is an open access article distributed under the terms and conditions of the Creative... more This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY
Making diagnosis by learning from examples is a typical field of artificial neural networks. Howe... more Making diagnosis by learning from examples is a typical field of artificial neural networks. However, justifications of network responses are difficult to obtain, especially when input examples have analog variables. We propose a particular multi-layer Perceptron model in which explanations of responses are obtained through symbolic rules. The originality of this model consists in its architecture. Experiments using three datasets related to breast cancer diagnosis, coronary heart disease and thyroid dysfunctions have shown high mean predictive accuracy (respectively: 96.3%, 90.0%, 99.3%). Comparisons with the C4.5 algorithm, which builds inductive decision trees, have shown that the predictive accuracy of both approaches is roughly the same, with neural networks slightly more accurate.
Millions of elderly people around the world use the walker for their mobility; nevertheless, thes... more Millions of elderly people around the world use the walker for their mobility; nevertheless, these devices may lead to an accident. One of the cause of these accidents is misjudge the terrain. The main objective of this work is the implementation of a ground change detector in real time on a small and light embedded system that can be clipped on a rollator. As a long-term goal, this device will allow users to anticipate entering dangerous situations. We implemented an algorithm to detect ground changes based on color histograms and texture descriptor given as inputs to multi-layer perceptrons. Experiments were performed both off-line and with an embedded system. The obtained results indicated that it is possible to have an accurate detector which is able to distinguish ground changes in real-time.
1Computer Science Department, University of Geneva, Battelle Campus, 7 Route de Drize, 1227 Carou... more 1Computer Science Department, University of Geneva, Battelle Campus, 7 Route de Drize, 1227 Carouge (Geneva), Switzerland 2Computer Science Department, University of Applied Studies (HES-SO), 4 Rue de la Prairie, 1202 Geneva, Switzerland 3Center for Research and Technology Hellas (ITI/CERTH), Informatics and Telematics Institute, 1st Km Thermi-Panorama Road, P.O. Box 361, 57001 Thermi-Thessaloniki, Greece
Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02., 2002
ABSTRACT Although two proteins may be structurally similar, they may not have significant sequenc... more ABSTRACT Although two proteins may be structurally similar, they may not have significant sequence similarity. The recognition of protein fold structures without relying on sequence similarity is a complex task. This work presents a comparison study on the recognition of 3-dimensional protein folds by Machine Learning models. Combinations of neural networks were trained by bagging and arcing with two datasets available online (http://www.nersc.gov/). Our results improved the average predictive accuracy obtained by Support Vector Machines in previously published work.
Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks. IJCNN 2000. Neural Computing: New Challenges and Perspectives for the New Millennium, 2000
ABSTRACT We tackle the problem of rule extraction from multilayer perceptrons. Our approach consi... more ABSTRACT We tackle the problem of rule extraction from multilayer perceptrons. Our approach consists of characterising discriminant hyper-plane frontiers built by a special neural network model, denoted as a discretized interpretable multilayer perceptron (DIMLP). Rules are extracted in polynomial time with respect to the size of the problem. Further, the degree of matching between extracted rules and neural network responses is 100%. We apply DIMLP to five data sets of the public domain in which for some of them it gives better average predictive accuracy than standard multilayer perceptrons and C4.5 decision trees
Proceedings of the Third International Conference on Computer Vision Theory and Applications, 2008
The context of this work is the development of a mobility aid for visually impaired persons. We p... more The context of this work is the development of a mobility aid for visually impaired persons. We present here an original approach for a real time alerting system, based on the use of feature maps for detecting visual salient parts in images. In order to improve the quality of this method, we propose here to benefit from a new feature map constructed from the depth gradient. A specific distance function is described, which takes into account both stereoscopic camera limitations and user's choices. We demonstrate here that this additional depth-based feature map allows the system to detect the salient regions with good accuracy in most situations, even with noisy disparity maps.
ABSTRACT In this work the purpose is to determine discriminant hyperplanes of a neural network in... more ABSTRACT In this work the purpose is to determine discriminant hyperplanes of a neural network in order to extract possible valuable knowledge by means of symbolic rules. We define a special neural network model denoted to as Discretized Interpretable Multi Layer Perceptron (DIMLP). As a result, rules are extracted in polynomial time with respect to the size of the problem and the size of the network. Further, the degree of matching between extracted rules and neural network responses is 100%. Our network model was tested on 7 classification problems of the public domain. It turned out that DIMLPs were significantly more accurate than C4.5 decision trees on average.
Artificial Neural Nets Problem Solving Methods, 2003
This work presents ensembles of neural network models that learn to discriminate images from diff... more This work presents ensembles of neural network models that learn to discriminate images from different categorical scenes. The basic idea was to use ICA filter energies and to train neural network ensem- bles. The presented results improved the predictive accuracy of previ- ously published work on the second classification problem. Finally, rules generated from ensembles in the less complex classification task showed that a few filters are sufficient to reach a good recognition rate, whereas many more filters are represented in the rule antecedents of the most difficult classification problem.
The problem of rule extraction from neural networks is NP-hard. This work presents a new techniqu... more The problem of rule extraction from neural networks is NP-hard. This work presents a new technique to extract If-Then-Else rules from linear combinations of discretised interpretable multilayer perceptron (DIMLP) neural networks. Rules are extracted in polynomial time with respect to the dimensionality of the problem, the number of examples, and the size of the resulting network. Further, the degree of matching between extracted rules and neural network responses is 100%. Linear combinations of DIMLP networks were trained on 4 data sets related to the public domain. The extracted rules obtained are more accurate than those extracted from C4.5 decision trees on average
ABSTRACT « EyeWalker » est un projet qui a pour but de développer un appareil léger et compact s’... more ABSTRACT « EyeWalker » est un projet qui a pour but de développer un appareil léger et compact s’adaptant facilement à n’importe quel déambulateur et alertant un/e utilisateur/trice avant qu’il/elle se trouve dans une situation dangereuse pouvant entraîner sa chute. Dans ce travail, nous traitons la détection de deux types de situations présentant un risque. Le premier est lié à l’état affectif de la personne et le deuxième est défini par la présence d’escaliers descendants.
The present paper presents the framework and the results of Project 2: "Multimodal tools and inte... more The present paper presents the framework and the results of Project 2: "Multimodal tools and interfaces for the intercommunication between visually impaired and "deaf and mute" people", which has been developed during the eNTERFACE-2006 summer workshop in the context of the SIMILAR NoE. The developed system aims to provide alternative tools and interfaces to blind and deaf-and-mute persons so as to enable their intercommunication as well as their interaction with the computer. All the involved technologies are integrated into a treasure hunting game application that is jointly played by the blind and deaf-and-mute user. The reason for choosing to integrate the multimodal interfaces into a game application is that it serves both as an entertainment and as a pleasant education tool to its users. The proposed application integrates haptics, audio, visual output as well as computer vision, sign language analysis and synthesis, speech recognition and synthesis, in order to provide an interactive environment where the blind and deaf and mute users can collaborate in order to play the treasure hunting game.
International Conference on Artificial Neural Networks, 2016
The Recursive-Rule eXtraction (Re-RX) algorithm family includes the Re-RX algorithm, the Re-RX al... more The Recursive-Rule eXtraction (Re-RX) algorithm family includes the Re-RX algorithm, the Re-RX algorithm with both discrete and continuous attributes (Continuous Re-RX [1]), the Re-RX algorithm with J48graft [2], Re-RX with J48graft combined with Sampling Selection Techniques (Sampling Re-RX with J48graft [4]), and the Re-RX algorithm with a trained neural network (Sampling Re-RX [3]). In this study, we compare the performance of the Re-RX algorithm family with various previous algorithms. One issue that always remains important in rule extraction is Pareto optimality, or in other words, an ideally balanced trade-off. In rule extraction, the trade-off is between the classification accuracy and interpretability of extracted rules. Our goal is to obtain a wider viable region for the Pareto optimal curve that will enable improvements in both the accuracy and interpretability of extracted rules. We vividly demonstrate Pareto-optimal curves between the accuracies and number of rules obtained for German and Australian datasets by 10 runs of 10-fold cross validation of the Re-RX algorithm family and those obtained using other algorithms. The Re-RX algorithm family has proven effective for extracting concise and interpretable rules from medical [1, 2, 4] and financial [3] datasets
This article is an open access article distributed under the terms and conditions of the Creative... more This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY
Making diagnosis by learning from examples is a typical field of artificial neural networks. Howe... more Making diagnosis by learning from examples is a typical field of artificial neural networks. However, justifications of network responses are difficult to obtain, especially when input examples have analog variables. We propose a particular multi-layer Perceptron model in which explanations of responses are obtained through symbolic rules. The originality of this model consists in its architecture. Experiments using three datasets related to breast cancer diagnosis, coronary heart disease and thyroid dysfunctions have shown high mean predictive accuracy (respectively: 96.3%, 90.0%, 99.3%). Comparisons with the C4.5 algorithm, which builds inductive decision trees, have shown that the predictive accuracy of both approaches is roughly the same, with neural networks slightly more accurate.
Millions of elderly people around the world use the walker for their mobility; nevertheless, thes... more Millions of elderly people around the world use the walker for their mobility; nevertheless, these devices may lead to an accident. One of the cause of these accidents is misjudge the terrain. The main objective of this work is the implementation of a ground change detector in real time on a small and light embedded system that can be clipped on a rollator. As a long-term goal, this device will allow users to anticipate entering dangerous situations. We implemented an algorithm to detect ground changes based on color histograms and texture descriptor given as inputs to multi-layer perceptrons. Experiments were performed both off-line and with an embedded system. The obtained results indicated that it is possible to have an accurate detector which is able to distinguish ground changes in real-time.
1Computer Science Department, University of Geneva, Battelle Campus, 7 Route de Drize, 1227 Carou... more 1Computer Science Department, University of Geneva, Battelle Campus, 7 Route de Drize, 1227 Carouge (Geneva), Switzerland 2Computer Science Department, University of Applied Studies (HES-SO), 4 Rue de la Prairie, 1202 Geneva, Switzerland 3Center for Research and Technology Hellas (ITI/CERTH), Informatics and Telematics Institute, 1st Km Thermi-Panorama Road, P.O. Box 361, 57001 Thermi-Thessaloniki, Greece
Proceedings of the 9th International Conference on Neural Information Processing, 2002. ICONIP '02., 2002
ABSTRACT Although two proteins may be structurally similar, they may not have significant sequenc... more ABSTRACT Although two proteins may be structurally similar, they may not have significant sequence similarity. The recognition of protein fold structures without relying on sequence similarity is a complex task. This work presents a comparison study on the recognition of 3-dimensional protein folds by Machine Learning models. Combinations of neural networks were trained by bagging and arcing with two datasets available online (http://www.nersc.gov/). Our results improved the average predictive accuracy obtained by Support Vector Machines in previously published work.
Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks. IJCNN 2000. Neural Computing: New Challenges and Perspectives for the New Millennium, 2000
ABSTRACT We tackle the problem of rule extraction from multilayer perceptrons. Our approach consi... more ABSTRACT We tackle the problem of rule extraction from multilayer perceptrons. Our approach consists of characterising discriminant hyper-plane frontiers built by a special neural network model, denoted as a discretized interpretable multilayer perceptron (DIMLP). Rules are extracted in polynomial time with respect to the size of the problem. Further, the degree of matching between extracted rules and neural network responses is 100%. We apply DIMLP to five data sets of the public domain in which for some of them it gives better average predictive accuracy than standard multilayer perceptrons and C4.5 decision trees
Proceedings of the Third International Conference on Computer Vision Theory and Applications, 2008
The context of this work is the development of a mobility aid for visually impaired persons. We p... more The context of this work is the development of a mobility aid for visually impaired persons. We present here an original approach for a real time alerting system, based on the use of feature maps for detecting visual salient parts in images. In order to improve the quality of this method, we propose here to benefit from a new feature map constructed from the depth gradient. A specific distance function is described, which takes into account both stereoscopic camera limitations and user's choices. We demonstrate here that this additional depth-based feature map allows the system to detect the salient regions with good accuracy in most situations, even with noisy disparity maps.
ABSTRACT In this work the purpose is to determine discriminant hyperplanes of a neural network in... more ABSTRACT In this work the purpose is to determine discriminant hyperplanes of a neural network in order to extract possible valuable knowledge by means of symbolic rules. We define a special neural network model denoted to as Discretized Interpretable Multi Layer Perceptron (DIMLP). As a result, rules are extracted in polynomial time with respect to the size of the problem and the size of the network. Further, the degree of matching between extracted rules and neural network responses is 100%. Our network model was tested on 7 classification problems of the public domain. It turned out that DIMLPs were significantly more accurate than C4.5 decision trees on average.
Artificial Neural Nets Problem Solving Methods, 2003
This work presents ensembles of neural network models that learn to discriminate images from diff... more This work presents ensembles of neural network models that learn to discriminate images from different categorical scenes. The basic idea was to use ICA filter energies and to train neural network ensem- bles. The presented results improved the predictive accuracy of previ- ously published work on the second classification problem. Finally, rules generated from ensembles in the less complex classification task showed that a few filters are sufficient to reach a good recognition rate, whereas many more filters are represented in the rule antecedents of the most difficult classification problem.
The problem of rule extraction from neural networks is NP-hard. This work presents a new techniqu... more The problem of rule extraction from neural networks is NP-hard. This work presents a new technique to extract If-Then-Else rules from linear combinations of discretised interpretable multilayer perceptron (DIMLP) neural networks. Rules are extracted in polynomial time with respect to the dimensionality of the problem, the number of examples, and the size of the resulting network. Further, the degree of matching between extracted rules and neural network responses is 100%. Linear combinations of DIMLP networks were trained on 4 data sets related to the public domain. The extracted rules obtained are more accurate than those extracted from C4.5 decision trees on average
ABSTRACT « EyeWalker » est un projet qui a pour but de développer un appareil léger et compact s’... more ABSTRACT « EyeWalker » est un projet qui a pour but de développer un appareil léger et compact s’adaptant facilement à n’importe quel déambulateur et alertant un/e utilisateur/trice avant qu’il/elle se trouve dans une situation dangereuse pouvant entraîner sa chute. Dans ce travail, nous traitons la détection de deux types de situations présentant un risque. Le premier est lié à l’état affectif de la personne et le deuxième est défini par la présence d’escaliers descendants.
Uploads
Papers by Guido Bologna