10 research outputs found
Convergent Dynamics of Nonreciprocal Differential Variational Inequalities Modeling Neural Networks
No full textThe paper addresses convergence of solutions for a class of differential inclusions termed differential variational inequalities (DVIs). Each DVI describes the dynamics of a neural network (NN) evolving in a closed hypercube of and defined by a continuously differentiable, {\em cooperative\/} and (possibly) nonreciprocal vector field . The main result in the paper is that under a new condition on , which is called strong Kamke-Muller condition, the solution semiflow generated by the DVI is strongly order preserving (SOP) and hence it satisfies a {\sc Limit Set Dichotomy} and enjoys generic convergence properties. A characterization of the SKM condition is given in terms of the interconnection properties of the Jacobian matrix of . In the case where is an affine, or a linear, vector field the considered DVIs include two relevant classes of NNs, namely, the linear systems operating on a closed hypercube, also known as linear systems in saturated mode (LSSMs), and the full-range (FR) model of cellular neural networks (CNNs). By applying the results to LSSMs it is obtained that any cooperative LSSM with a (possibly) nonsymmetric and fully interconnected matrix is generically convergent. Analogous results hold for FRCNNs. All the obtained convergence results hold in the general case where the DVIs, and the LSSMs and FRCNNs, possess multiple equilibrium points
An experimental study on long transient oscillations in cooperative CNN rings
No full textThe paper considers a class of one-dimensional circular standard cellular neural network (CNN) arrays
with a typical three-segment piecewise linear activation and two-sided {em cooperative/} (positive) interactions (a cooperative CNN ring). Numerical simulations show that in a wide range of interconnection parameters, and for a wide set of initial conditions, the solutions of a cooperative CNN ring display unexpectedly long oscillations, lasting even hundreds of cycles, before they eventually converge toward an equilibrium point. The goal of this paper is to confirm the presence
of such long-transient oscillations through laboratory experiments on a simple
discrete-component prototype of a cooperative CNN ring with 16 cells and to analyze
some of their salient features. Analytical results are also provided to
support the numerical and experimental findings
Monotonicity of semiflows generated by cooperative delayed full-range CNNs
No full textThe paper considers the full-range (FR) model of cellular neural networks (CNNs) with ideal hard-limiter non-linearities that limit the allowable range of the neuron state variables. It is also supposed that there is a concentrated delay (D) in the neuron interconnections. Due to the presence of multivalued nonlinearities the D-FRCNN model is mathematically described by a retarded differential inclusion. The main result is a rigorous proof that, in the case of nonsymmetric cooperative (nonnegative) interconnections, and delayed interconnections, the semiflow generated by D-FRCNNs is monotone, and that monotonicity implies some basic restrictions on the long-term behavior of the solutions. The result is compared with recent results in the literature on semiflows generated by cooperative standard CNNs, with and without delays
Oscillatory Circuits with a Real Non-Volatile Stanford Memristor Model
No full textStanford memristor model is a widely used model that accurately characterizes real non-volatile metal-oxide resistive random access memory (RRAM) devices with bipolar switching characteristics. The paper studies for the first time the dynamics and bifurcations in a class of nonlinear oscillators with real non-volatile memristor devices obeying Stanford model. This is in contrast with papers in the literature considering oscillators with ideal, abstract, or artificial memristor models, that are unable to describe physical memristors implemented in nanotechnology. One main new idea in the paper is to use the memristor as a programmable nonlinear resistor. Namely, two principal modes of operation are considered. 1) Analogue transient phase: the oscillator is designed so that in the transient oscillations the voltage on the memristor is below threshold, hence the main memristor state variable, i.e., the gap of the insulating material, is almost constant and the memristor behaves as a static nonlinear resistor. 2) Programming phase: the nonlinear characteristic of the memristor, which depends on the gap, can be changed via the application of voltages above threshold. The paper studies nonlinear oscillations in the transient phase for a fixed gap as well as the bifurcations phenomena displayed when the gap is varied. The paper also discusses the differences between the approach in the paper and those to design other memristor oscillators with non-volatile memristors
Intervertebral disc regeneration: morphological investigation of an in vitro reconstructed tissue
No full textGenerating Bounding Box Supervision for Semantic Segmentation with Deep Learning
No full textMost of the leading Convolutional Neural Network (CNN) models for semantic segmentation exploit a large number of pixel–level annotations. Such a human based labeling requires a considerable effort that complicates the creation of large–scale datasets. In this paper, we propose a deep learning approach that uses bounding box annotations to train a semantic segmentation network. Indeed, the bounding box supervision, even though less accurate, is a valuable alternative, effective in reducing the dataset collection costs. The proposed method is based on a two stage training procedure: first, a deep neural network is trained to distinguish the relevant object from the background inside a given bounding box; then, the output of the network is used to provide a weak supervision for a multi–class segmentation CNN. The performances of our approach have been assessed on the Pascal–VOC 2012 segmentation dataset, obtaining competitive results compared to a fully supervised setting
Adipose derived stem cell therapy for intervertebral disc regeneration : an in vitro reconstructed tissue in alginate capsule
No full textCombining deep learning and symbolic processing for extracting knowledge from raw text
No full textThis paper faces the problem of extracting knowledge from raw text. We present a deep architecture in the framework of Learning from Constraints [5] that is trained to identify mentions to entities and relations belonging to a given ontology. Each input word is encoded into two latent representations with different coverage of the local context, that are exploited to predict the type of entity and of relation to which the word belongs. Our model combines an entropy-based regularizer and a set of First-Order Logic formulas that bridge the predictions on entity and relation types accordingly to the ontology structure. As a result, the system generates symbolic descriptions of the raw text that are interpretable and well-suited to attach human-level knowledge. We evaluate the model on a dataset composed of sentences about simple facts, that we make publicly available. The proposed system can efficiently learn to discover mentions with very few human supervisions and that the relation to knowledge in the form of logic constraints improves the quality of the system predictions
ATM Protection Using Embedded Deep Learning Solutions
No full textLast decade advances in Deep Learning methods lead to sensible improvements in state of the art results in many real world applications, thanks to the exploitation of particular Artificial Neural Networks architectures. In this paper we present an investigation of the application of such kind of structures to a Video Surveillance case of study, in which the special nature and the small amount of available data increases the difficulties during the training phase. The analyzed scenario involves the protection of Automatic Teller Machines (ATM), representing a sensitive problem in the world of both banking and public security. Because of the critical issues related to this environment, even apparently small improvements in either accuracy or responsiveness of surveillance systems can produce a fundamental contribution. Even if the experimentation has been reproduced in an artificial scenario, the results show that the implemented architecture is able to classify depth data in real-time on an embedded system, detecting all the test attacks in a few seconds
Inductive–Transductive Learning with Graph Neural Networks
No full textGraphs are a natural choice to encode data in many real–world applications. In fact, a graph can describe a given pattern as a complex structure made up of parts (the nodes) and relationships between them (the edges). Despite their rich representational power, most of machine learning approaches cannot deal directly with inputs encoded by graphs. Indeed, Graph Neural Networks (GNNs) have been devised as an extension of recursive models, able to process general graphs, possibly undirected and cyclic. In particular, GNNs can be trained to approximate all the “practically useful” functions on the graph space, based on the classical inductive learning approach, realized within the supervised framework. However, the information encoded in the edges can actually be used in a more refined way, to switch from inductive to transductive learning. In this paper, we present an inductive–transductive learning scheme based on GNNs. The proposed approach is evaluated both on artificial and real–world datasets showing promising results. The recently released GNN software, based on the Tensorflow library, is made available for interested users
