A5017752685- Tactile and Sensory Interactions
- Neural Networks and Applications
- Multisensory perception and integration
- Rough Sets and Fuzzy Logic
- Fuzzy Logic and Control Systems
- Machine Learning and Data Classification
- Explainable Artificial Intelligence (XAI)
- Video Surveillance and Tracking Methods
- Cybercrime and Law Enforcement Studies
- Visual Attention and Saliency Detection
- Interactive and Immersive Displays
- Data Mining Algorithms and Applications
- Gaze Tracking and Assistive Technology
- Robotics and Sensor-Based Localization
- Anomaly Detection Techniques and Applications
- Advanced Vision and Imaging
- Imbalanced Data Classification Techniques
- Adversarial Robustness in Machine Learning
- Hearing Loss and Rehabilitation
- Blind Source Separation Techniques
- Machine Learning in Bioinformatics
- Visual perception and processing mechanisms
- Advanced Image and Video Retrieval Techniques
- AI in cancer detection
- Advanced Neural Network Applications
HES-SO University of Applied Sciences and Arts Western Switzerland
2016-2025
University of Geneva
2010-2020
Information Technology University
2015-2016
SIB Swiss Institute of Bioinformatics
2002-2011
Aristotle University of Thessaloniki
2006
Geneva College
2002
National University of Singapore
2002
Queensland University of Technology
2000
University of Padua
1968-2000
Trenitalia (Italy)
1997
Deep Learning (DL), a groundbreaking branch of Machine (ML), has emerged as driving force in both theoretical and applied Artificial Intelligence (AI). DL algorithms, rooted complex non-linear artificial neural systems, excel at extracting high-level features from data. demonstrated human-level performance real-world tasks, including clinical diagnostics, unlocked solutions to previously intractable problems virtual agent design, robotics, genomics, neuroimaging, computer vision, industrial...
N-terminal myristoylation is a post-translational modification that causes the addition of myristate to glycine in end amino acid chain. This work presents neural network (NN) models learn discriminate myristoylated and nonmyristoylated proteins. Ensembles 25 NNs decision trees were trained on 390 positive sequences 327 negative sequences. Experiments showed NN ensembles more accurate than tree ensembles. Our predictor evaluated by leave-one-out procedure, obtained false error rate equal...
Abstract Rule extraction from neural networks is a fervent research topic. In the last 20 years many authors presented number of techniques showing how to extract symbolic rules Multi Layer Perceptrons (MLPs). Nevertheless, very few were related ensembles and even less for trained by deep learning. On several datasets we performed rule Discretized Interpretable (DIMLP), DIMLPs The results obtained on Thyroid dataset Wisconsin Breast Cancer show that predictive accuracy extracted compare...
One way to make the knowledge stored in an artificial neural network more intelligible is extract symbolic rules. However, producing rules from Multilayer Perceptrons (MLPs) NP-hard problem. Many techniques have been introduced generate single networks, but very few were proposed for ensembles. Moreover, experiments rarely assessed by 10-fold cross-validation trials. In this work, based on Discretized Interpretable Perceptron (DIMLP), performed 10 repetitions of stratified trials over 25...
Deep connectionist models are characterized by many neurons grouped together in successive layers. As a result, their data classifications difficult to understand. We present two novel algorithms which explain the responses of several black-box machine learning models. The first is Fidex, local and thus applied single sample. second, called FidexGlo, global uses Fidex. Both generate explanations means propositional rules. In our framework, discriminative boundaries parallel input variables...
Deep connectionist models are characterized by many neurons grouped together in successive layers. As a result, their data classifications difficult to understand. We present two novel algorithms which explain the responses of several black-box machine learning models. The first is Fidex, local and thus applied single sample. second, called FidexGlo, global uses Fidex. Both generate explanations means propositional rules. In our framework, discriminative boundaries parallel input variables...
Although two proteins may be structurally similar, they not have significant sequence similarity. The recognition of protein fold structures without relying on similarity is a complex task. This work presents comparison study the 3-dimensional folds by Machine Learning models. Combinations neural networks were trained bagging and arcing with datasets available online (http://www.nersc.gov/). Our results improved average predictive accuracy obtained Support Vector Machines in previously...
Abstract The purpose of this study was to generate more concise rule extraction from the Recursive-Rule Extraction (Re-RX) algorithm by replacing C4.5 program currently employed in Re-RX with J48graft algorithm. Experiments were subsequently conducted determine rules for six different two-class mixed datasets having discrete and continuous attributes compare resulting accuracy, comprehensibility conciseness. When working CARD1, CARD2, CARD3, German, Bene1 Bene2 datasets, provided than...
The goal of the See ColOr project is to achieve a noninvasive mobility aid for blind users that will use auditory pathway represent in real-time frontal image scenes.We present and discuss here two processing methods were experimented this work: simplification by means segmentation, guiding focus attention through computation visual saliency.A mean shift segmentation technique gave best results, but constraints we simply implemented an quantification method based on HSL colour system.More...
Classification responses provided by Multi Layer Perceptrons (MLPs) can be explained means of propositional rules. So far, many rule extraction techniques have been proposed for shallow MLPs, but not Convolutional Neural Networks (CNNs). To fill this gap, work presents a new method applied to typical CNN architecture used in Sentiment Analysis (SA). We focus on the textual data which is trained with “tweets” movie reviews. Its includes an input layer representing words “word embeddings”,...
In machine learning, ensembles of models based on Multi-Layer Perceptrons (MLPs) or decision trees are considered successful models. However, explaining their responses is a complex problem that requires the creation new methods interpretation. A natural way to explain classifications transform them into propositional rules. this work, we focus random forests and gradient-boosted trees. Specifically, these converted an ensemble interpretable MLPs from which rules produced. The rule...
The problem of rule extraction from neural networks is NP-hard. This work presents a new technique to extract "if-then-else" rules ensembles DIMLP networks. Rules are extracted in polynomial time with respect the dimensionality problem, number examples, and size resulting network. Further, degree matching between network responses 100%. Ensembles were trained on four data sets public domain. Extracted average significantly more accurate than those C4.5 decision trees.
A natural way to determine the knowledge embedded within connectionist models is generate symbolic rules. Nevertheless, extracting rules from Multi Layer Perceptrons (MLPs) NP-hard. With advent of social networks, techniques applied Sentiment Analysis show a growing interest, but rule extraction in this context has been rarely performed because very high dimensionality input space. To fill gap we present case study on ensembles Neural Networks and Support Vector Machines (SVMs), purpose...
With the increasing proportion of senior citizens, many mobility aid devices have been developed such as rollator. However, under some circumstances, latter may cause accidents. The EyeWalker project aims to develop a small and autonomous device for rollators help elderly people, especially those with degree visual impairment, avoiding common dangers like obstacles hazardous ground changes, both outdoors indoors. We propose amethod real-time stereo obstacle detection using sparse 3D...
Purpose – The purpose of this paper is to overcome the limitations sensory substitution methods (SSDs) represent high-level or conceptual information involved in vision, which are mainly produced by biological mismatch between sight and substituting senses. Thus, provide visually impaired with a more practical functional SSD. Design/methodology/approach Unlike any other approach, SSD extends beyond sensing prototype, integrating computer vision produce reliable knowledge about physical world...
Rule extraction from neural networks is a fervent research topic. In the last 20 years many authors presented number of techniques showing how to extract symbolic rules Multi Layer Perceptrons (MLPs). Nevertheless, very few were related ensembles and even less for trained by deep learning. this work Discretized Perceptron (DIMLP) was learning, then extracted in an easier way with respect standard MLPs. We compared accuracy DIMLPs DIMLP on subset MNIST dataset. The former more accurate than...