Multi-scale stacked sequential learning

Sequential learning is the discipline of machine learning that deals with dependent data such that neighboring labels exhibit some kind of relationship. The paper main contribution is two-fold: first, we generalize the stacked sequential learning, highlighting the key role of neighboring interactions modeling. Second, we propose an effective and efficient way of capturing and exploiting sequential correlations that takes into account long-range interactions. We tested the method on two tasks: text lines classification and image pixel classification. Results on these tasks clearly show that our approach outperforms the standard stacked sequential learning as well as state-of-the-art conditional random fields.     

HMAP– A temporal data model managing intervals with different granularities and indeterminacy from natural language sentences

The granularity of given temporal information is the level of abstraction at which information is expressed. Different units of measure allow one to represent different granularities. Indeterminacy is often present in temporal information given at different granularities: temporal indeterminacy is related to incomplete knowledge of when the considered fact happened. Focusing on temporal databases, different granularities and indeterminacy have to be considered in expressing valid time, i.e., the time at which the information is true in the modeled reality. In this paper, we propose HMAP (The term is the transliteration of an ancient Greek poetical word meaning “day”.), a temporal data model extending the capability of defining valid times with different granularity and/or with indeterminacy. In HMAP, absolute intervals are explicitly represented by their start,end, and duration: in this way, we can represent valid times as “in December 1998 for five hours”, “from July 1995, for 15 days”, “from March 1997 to October 15, 1997, between 6 and 6:30 p.m.”. HMAP is based on a three-valued logic, for managing uncertainty in temporal relationships. Formulas involving different temporal relationships between intervals, instants, and durations can be defined, allowing one to query the database with different granularities, not necessarily related to that of data. In this paper, we also discuss the complexity of algorithms, allowing us to evaluate HMAP formulas, and show that the formulas can be expressed as constraint networks falling into the class of simple temporal problems, which can be solved in polynomial time. Received 6 August 1998 / Accepted 13 July 2000 Published online: 13 February 2001     

Experimental setup for lensless imaging via soft x-ray resonant scattering

We have developed a setup for measuring holographically formed interference patterns using an integrated sample-mask design. The direct space image of the sample is obtained via a two-dimensional Fourier transform of the X-ray diffraction pattern. We present the details of our setup, commenting on the influence of geometrical parameters on the imaging capabilities. As an example, we present and discuss the results of test experiments on a patterned Co film.     

Segmentation of news videos based on audio-video information

In this paper, we propose an innovative architecture to segment a news video into the so-called “stories” by both using the included video and audio information. Segmentation of news into stories is one of the key issues for achieving efficient treatment of news-based digital libraries. While the relevance of this research problem is widely recognized in the scientific community, we are in presence of a few established solutions in the field. In our approach, the segmentation is performed in two steps: first, shots are classified by combining three different anchor shot detection algorithms using video information only. Then, the shot classification is improved by using a novel anchor shot detection method based on features extracted from the audio track. Tests on a large database confirm that the proposed system outperforms each single video-based method as well as their combination.

The importance of different timings of excitatory and inhibitory pathways in neural field models

In this paper, we consider a neural field model comprised of two distinct populations of neurons, excitatory and inhibitory, for which both the velocities of action potential propagation and the time courses of synaptic processing are different. Using recently-developed techniques, we construct the Evans function characterising the stability of both stationary and travelling wave solutions, under the assumption that the firing rate function is the Heaviside step. We find that these differences in timing for the two populations can cause instabilities of these solutions, leading to, for example, stationary breathers. We also analyse "anti-pulses", a novel type of pattern for which all but a small interval of the domain (in moving coordinates) is active. These results extend previous work on neural fields with space-dependent delays, and demonstrate the importance of considering the effects of the different time-courses of excitatory and inhibitory neural activity.     

Kernel ridge regression for out-of-sample mapping in supervised manifold learning

Manifold learning methods for unsupervised nonlinear dimensionality reduction have proven effective in the visualization of high dimensional data sets. When dealing with classification tasks, supervised extensions of manifold learning techniques, in which class labels are used to improve the embedding of the training points, require an appropriate method for out-of-sample mapping.In this paper we propose multi-output kernel ridge regression (KRR) for out-of-sample mapping in supervised manifold learning, in place of general regression neural networks (GRNN) that have been adopted by previous studies on the subject. Specifically, we consider a supervised agglomerative variant of Isomap and compare the performance of classification methods when the out-of-sample embedding is based on KRR and GRNN, respectively. Extensive computational experiments, using support vector machines and k-nearest neighbors as base classifiers, provide statistical evidence that out-of-sample mapping based on KRR consistently dominates its GRNN counterpart, and that supervised agglomerative Isomap with KRR achieves a higher accuracy than direct classification methods on most data sets.     

Quantum Capacitance Effects in Carbon Nanotube Field-Effect Devices

Density functional numerical simulations are reported in which the non-equilibrium coherent transport properties of a coaxially gated carbon nanotube (CNT) are studied. Effects of screening properties of CNT on the channel conductance modulation are shown. Other factors which influence the CNT quantum capacitance beyond the density of states (DOS) are pointed out, first of all many-body effects which even cause the CNT to over-screen the gate field. Results stress the importance of a good electrostatic-design of the gate contact to obtain an effective field-effect current modulation, and suggest new interesting operation modes for quasi-one dimensional channel devices.     

Multirobot motion coordination in space and time

This paper describes a solution to the multirobot motion planning problem based on a decoupled analysis in the space domain and in the time domain. It investigates the practical use of the notion of motion plan quality and of the motion plan robustness measures for computing safe motions. The use of anytime algorithms allows one to evaluate the opportunity of looking for alternative solution paths by generating small variations of robot motions affecting both its geometrical path and its scheduled velocity. By using the concept of plan robustness, several alternative paths are generated and evaluated through various performance indices and impact factors, using heuristic rules. These indices allow one to know how much a variation affects a given plan. Finally, some recent experiments are outlined.     

An Attempt at the Classification of Energy Decaying Schemes for Structural and Multibody Dynamics

This paper analyzes the formulation of energy preserving/decaying schemes for dynamics problems. We argue that any energy preserving/decaying scheme can always be seen as composed of an underlying temporal discretization, that is then slightly modified in order to prove a discrete energy bound within a time step. While the details of the modified scheme depend in a critical way on the governing equations, the underlying discretization can in principle be applied to a variety of models. We review some of the temporal underlying schemes recently proposed in the literature, presenting them with a common notation. We show their similarities and highlight their differences.