We consider a new class of non Markovian processes with a countable number of interacting components, both in discrete and continuous time. Each component is represented by a point process indicating if it has a spike or not at a given time. The system evolves as follows. For each component, the rate (in continuous time) or the probability (in discrete time) of having a spike depends on the entire time evolution of the system since the last spike time of the component. In discrete time this class of systems extends in a non trivial way both Spitzer’s interacting particle systems, which are Markovian, and Rissanen’s stochastic chains with memory of variable length which have finite state space. In continuous time they can be seen as a kind of Rissanen’s variable length memory version of the class of self-exciting point processes which are also called “Hawkes processes”, however with infinitely many components. These features make this class a good candidate to describe the time evolution of networks of spiking neurons. In this article we present a critical reader’s guide to recent papers dealing with this class of models, both in discrete and in continuous time. We briefly sketch results concerning perfect simulation and existence issues, de-correlation between successive interspike intervals, the longtime behavior of finite non-excited systems and propagation of chaos in mean field systems.
https://arxiv.org/abs/1502.06446
Axions are well motivated particles that could make up most or all of the dark matter if they have masses below 100 $\mu$eV. Microwave cavity techniques comprised of closed resonant structures immersed in solenoid magnets are sensitive to dark matter axions with masses of a few $\mu$eV, but face difficulties scaling to higher masses. We present the a novel detector architecture consisting of an open, Fabry-Pérot resonator and a series of current-carrying wire planes, and demonstrate this technique with a search for dark matter axion-like particles called Orpheus. This search excludes dark matter axion-like particles with masses between 68.2 and 76.5 $\mu$eV and axion-photon couplings greater than $4\times10^{-7} \mathrm{GeV}^{-1}$. We project that the fundamental sensitivity of this technique could be extended to be sensitive to couplings below $1\times10^{-15} \mathrm{GeV}^{-1}$, consistent with the DFSZ model of QCD axions.
https://arxiv.org/abs/1403.3121
We present a simple regularization technique for Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) units. Dropout, the most successful technique for regularizing neural networks, does not work well with RNNs and LSTMs. In this paper, we show how to correctly apply dropout to LSTMs, and show that it substantially reduces overfitting on a variety of tasks. These tasks include language modeling, speech recognition, image caption generation, and machine translation.
我们提出了一个简单的正则化技术用于具有长期短期记忆（LSTM）单元的递归神经网络（RNN）。辍学是规范神经网络最成功的技术，对于RNN和LSTM来说效果不佳。在本文中，我们展示了如何正确地将丢失应用到LSTM，并显示它大大减少了对各种任务的过度拟合。这些任务包括语言建模，语音识别，图像标题生成和机器翻译。
https://arxiv.org/abs/1409.2329
In computer interfaces in general, especially in information retrieval tasks, it is important to be able to quickly find and retrieve information. State of the art approach, used, for example, in search engines, is not effective as it introduces losses of meanings due to context to keywords back and forth translation. Authors argue it increases the time and reduces the accuracy of information retrieval compared to what it could be in the system that employs modern information retrieval and text mining methods while presenting results in an adaptive human- computer interface where system effectively learns what operator needs through iterative interaction. In current work, a combination of adaptive navigational interface and real time collaborative feedback analysis for documents relevance weighting is proposed as an viable alternative to prevailing “telegraphic” approach in information retrieval systems. Adaptive navigation is provided through a dynamic links panel controlled by an evolutionary algorithm. Documents relevance is initially established with standard information retrieval techniques and is further refined in real time through interaction of users with the system. Introduced concepts of multidimensional Knowledge Map and Weighted Point of Interest allow finding relevant documents and users with common interests through a trivial calculation. Browsing search approach, the ability of the algorithm to adapt navigation to users interests, collaborative refinement and the self-organising features of the system are the main factors making such architecture effective in various fields where non-structured knowledge shall be represented to the users.
https://arxiv.org/abs/1502.05535
We introduce algorithms to visualize feature spaces used by object detectors. Our method works by inverting a visual feature back to multiple natural images. We found that these visualizations allow us to analyze object detection systems in new ways and gain new insight into the detector’s failures. For example, when we visualize the features for high scoring false alarms, we discovered that, although they are clearly wrong in image space, they do look deceptively similar to true positives in feature space. This result suggests that many of these false alarms are caused by our choice of feature space, and supports that creating a better learning algorithm or building bigger datasets is unlikely to correct these errors. By visualizing feature spaces, we can gain a more intuitive understanding of recognition systems.
https://arxiv.org/abs/1502.05461
We introduce a highly-parallelizable architecture for estimating parameters of compact binary coalescence using gravitational-wave data and waveform models. Using a spherical harmonic mode decomposition, the waveform is expressed as a sum over modes that depend on the intrinsic parameters (e.g. masses) with coefficients that depend on the observer dependent extrinsic parameters (e.g. distance, sky position). The data is then prefiltered against those modes, at fixed intrinsic parameters, enabling efficiently evaluation of the likelihood for generic source positions and orientations, independent of waveform length or generation time. We efficiently parallelize our intrinsic space calculation by integrating over all extrinsic parameters using a Monte Carlo integration strategy. Since the waveform generation and prefiltering happens only once, the cost of integration dominates the procedure. Also, we operate hierarchically, using information from existing gravitational-wave searches to identify the regions of parameter space to emphasize in our sampling. As proof of concept and verification of the result, we have implemented this algorithm using standard time-domain waveforms, processing each event in less than one hour on recent computing hardware. For most events we evaluate the marginalized likelihood (evidence) with statistical errors of less than about 5%, and even smaller in many cases. With a bounded runtime independent of the waveform model starting frequency, a nearly-unchanged strategy could estimate NS-NS parameters in the 2018 advanced LIGO era. Our algorithm is usable with any noise curve and existing time-domain model at any mass, including some waveforms which are computationally costly to evolve.
https://arxiv.org/abs/1502.04370
Many realistic networks are scale-free, with small characteristic path lengths, high clustering, and power law in their degree distribution. They can be obtained by dynamical networks in which a preferential attachment process takes place. However, this mechanism is non-local, in the sense that it requires knowledge of the whole graph in order for the graph to be updated. Instead, if preferential attachment and realistic networks occur in physical systems, these features need to emerge from a local model. In this paper, we propose a local model and show that a possible ingredient (which is often underrated) for obtaining scale-free networks with local rules is memory. Such a model can be realised in solid-state circuits, using non-linear passive elements with memory such as memristors, and thus can be tested experimentally.
https://arxiv.org/abs/1312.2289
In this paper, we propose an approach that exploits object segmentation in order to improve the accuracy of object detection. We frame the problem as inference in a Markov Random Field, in which each detection hypothesis scores object appearance as well as contextual information using Convolutional Neural Networks, and allows the hypothesis to choose and score a segment out of a large pool of accurate object segmentation proposals. This enables the detector to incorporate additional evidence when it is available and thus results in more accurate detections. Our experiments show an improvement of 4.1% in mAP over the R-CNN baseline on PASCAL VOC 2010, and 3.4% over the current state-of-the-art, demonstrating the power of our approach.
https://arxiv.org/abs/1502.04275
We investigate the transport of dipolar indirect excitons along the growth plane of polar (Al,Ga)N/GaN quantum well structures by means of spatially- and time-resolved photoluminescence spectroscopy. The transport in these strongly disordered quantum wells is activated by dipole-dipole repulsion. The latter induces an emission blue shift that increases linearly with exciton density, whereas the radiative recombination rate increases exponentially. Under continuous, localized excitation, we measure a continuous red shift of the emission, as excitons propagate away from the excitation spot. This shift corresponds to a steady-state gradient of exciton density, measured over several tens of micrometers. Time-resolved micro-photoluminescence experiments provide information on the dynamics of recombination and transport of dipolar excitons. We account for the ensemble of experimental results by solving the nonlinear drift-diffusion equation. Quantitative analysis suggests that in such structures, exciton propagation on the scale of 10 to 20 microns is mainly driven by diffusion, rather than by drift, due to the strong disorder and the presence of nonradiative defects. Secondary exciton creation, most probably by the intense higher-energy luminescence, guided along the sample plane, is shown to contribute to the exciton emission pattern on the scale up to 100 microns. The exciton propagation length is strongly temperature dependent, the emission being quenched beyond a critical distance governed by nonradiative recombination.
https://arxiv.org/abs/1502.03483
The optical properties of a stack of GaN/AlN quantum discs (QDiscs) in a GaN nanowire have been studied by spatially resolved cathodoluminescence (CL) at the nanoscale (nanoCL) using a Scanning Transmission Electron Microscope (STEM) operating in spectrum imaging mode. For the electron beam excitation in the QDisc region, the luminescence signal is highly localized with spatial extension as low as 5 nm due to the high band gap difference between GaN and AlN. This allows for the discrimination between the emission of neighbouring QDiscs and for evidencing the presence of lateral inclusions, about 3 nm thick and 20 nm long rods (quantum rods, QRods), grown unintentionally on the nanowire sidewalls. These structures, also observed by STEM dark-field imaging, are proven to be optically active in nanoCL, emitting at similar, but usually shorter, wavelengths with respect to most QDiscs.
https://arxiv.org/abs/1209.2545
This paper proposes a reconfigurable model to recognize and detect multiclass (or multiview) objects with large variation in appearance. Compared with well acknowledged hierarchical models, we study two advanced capabilities in hierarchy for object modeling: (i) “switch” variables(i.e. or-nodes) for specifying alternative compositions, and (ii) making local classifiers (i.e. leaf-nodes) shared among different classes. These capabilities enable us to account well for structural variabilities while preserving the model compact. Our model, in the form of an And-Or Graph, comprises four layers: a batch of leaf-nodes with collaborative edges in bottom for localizing object parts; the or-nodes over bottom to activate their children leaf-nodes; the and-nodes to classify objects as a whole; one root-node on the top for switching multiclass classification, which is also an or-node. For model training, we present an EM-type algorithm, namely dynamical structural optimization (DSO), to iteratively determine the structural configuration, (e.g., leaf-node generation associated with their parent or-nodes and shared across other classes), along with optimizing multi-layer parameters. The proposed method is valid on challenging databases, e.g., PASCAL VOC 2007 and UIUC-People, and it achieves state-of-the-arts performance.
https://arxiv.org/abs/1502.00744
This paper studies a novel discriminative part-based model to represent and recognize object shapes with an “And-Or graph”. We define this model consisting of three layers: the leaf-nodes with collaborative edges for localizing local parts, the or-nodes specifying the switch of leaf-nodes, and the root-node encoding the global verification. A discriminative learning algorithm, extended from the CCCP [23], is proposed to train the model in a dynamical manner: the model structure (e.g., the configuration of the leaf-nodes associated with the or-nodes) is automatically determined with optimizing the multi-layer parameters during the iteration. The advantages of our method are two-fold. (i) The And-Or graph model enables us to handle well large intra-class variance and background clutters for object shape detection from images. (ii) The proposed learning algorithm is able to obtain the And-Or graph representation without requiring elaborate supervision and initialization. We validate the proposed method on several challenging databases (e.g., INRIA-Horse, ETHZ-Shape, and UIUC-People), and it outperforms the state-of-the-arts approaches.
https://arxiv.org/abs/1502.00741
In this paper, we investigate a novel reconfigurable part-based model, namely And-Or graph model, to recognize object shapes in images. Our proposed model consists of four layers: leaf-nodes at the bottom are local classifiers for detecting contour fragments; or-nodes above the leaf-nodes function as the switches to activate their child leaf-nodes, making the model reconfigurable during inference; and-nodes in a higher layer capture holistic shape deformations; one root-node on the top, which is also an or-node, activates one of its child and-nodes to deal with large global variations (e.g. different poses and views). We propose a novel structural optimization algorithm to discriminatively train the And-Or model from weakly annotated data. This algorithm iteratively determines the model structures (e.g. the nodes and their layouts) along with the parameter learning. On several challenging datasets, our model demonstrates the effectiveness to perform robust shape-based object detection against background clutter and outperforms the other state-of-the-art approaches. We also release a new shape database with annotations, which includes more than 1500 challenging shape instances, for recognition and detection.
https://arxiv.org/abs/1502.00341
This work shows that the combination of ultrathin highly strained GaN quantum wells embedded in an AlN matrix, with controlled isotopic concentrations of Nitrogen enables a dual marker method for Raman spectroscopy. By combining these techniques, we demonstrate the effectiveness in studying strain in the vertical direction. This technique will enable the precise probing of properties of buried active layers in heterostructures, and can be extended in the future to vertical devices such as those used for optical emitters, and for power electronics.
https://arxiv.org/abs/1502.00072
Most object detection methods operate by applying a binary classifier to sub-windows of an image, followed by a non-maximum suppression step where detections on overlapping sub-windows are removed. Since the number of possible sub-windows in even moderately sized image datasets is extremely large, the classifier is typically learned from only a subset of the windows. This avoids the computational difficulty of dealing with the entire set of sub-windows, however, as we will show in this paper, it leads to sub-optimal detector performance. In particular, the main contribution of this paper is the introduction of a new method, Max-Margin Object Detection (MMOD), for learning to detect objects in images. This method does not perform any sub-sampling, but instead optimizes over all sub-windows. MMOD can be used to improve any object detection method which is linear in the learned parameters, such as HOG or bag-of-visual-word models. Using this approach we show substantial performance gains on three publicly available datasets. Strikingly, we show that a single rigid HOG filter can outperform a state-of-the-art deformable part model on the Face Detection Data Set and Benchmark when the HOG filter is learned via MMOD.
https://arxiv.org/abs/1502.00046
We consider a simple Markovian class of the stochastic Wilson-Cowan type models of neuronal network dynamics, which incorporates stochastic delay caused by the existence of a refractory period of neurons. From the point of view of the dynamics of the individual elements, we are dealing with a network of non-Markovian stochastic two-state oscillators with memory which are coupled globally in a mean-field fashion. This interrelation of a higher-dimensional Markovian and lower-dimensional non-Markovian dynamics is discussed in its relevance to the general problem of the network dynamics of complex elements possessing memory. The simplest model of this class is provided by a three-state Markovian neuron with one refractory state, which causes firing delay with an exponentially decaying memory within the two-state reduced model. This basic model is used to study critical avalanche dynamics (the noise sustained criticality) in a balanced feedforward network consisting of the excitatory and inhibitory neurons. Such avalanches emerge due to the network size dependent noise (mesoscopic noise). Numerical simulations reveal an intermediate power law in the distribution of avalanche sizes with the critical exponent around -1.16. We show that this power law is robust upon a variation of the refractory time over several orders of magnitude. However, the avalanche time distribution is biexponential. It does not reflect any genuine power law dependence.
https://arxiv.org/abs/1501.07448
The problem of sharing entanglement over large distances is crucial for implementations of quantum cryptography. A possible scheme for long-distance entanglement sharing and quantum communication exploits networks whose nodes share Einstein-Podolsky-Rosen (EPR) pairs. In Perseguers et al. [Phys. Rev. A 78, 062324 (2008)] the authors put forward an important isomorphism between storing quantum information in a dimension $D$ and transmission of quantum information in a $D+1$-dimensional network. We show that it is possible to obtain long-distance entanglement in a noisy two-dimensional (2D) network, even when taking into account that encoding and decoding of a state is exposed to an error. For 3D networks we propose a simple encoding and decoding scheme based solely on syndrome measurements on 2D Kitaev topological quantum memory. Our procedure constitutes an alternative scheme of state injection that can be used for universal quantum computation on 2D Kitaev code. It is shown that the encoding scheme is equivalent to teleporting the state, from a specific node into a whole two-dimensional network, through some virtual EPR pair existing within the rest of network qubits. We present an analytic lower bound on fidelity of the encoding and decoding procedure, using as our main tool a modified metric on space-time lattice, deviating from a taxicab metric at the first and the last time slices.
https://arxiv.org/abs/1202.1016
We present an atomistic description of the electronic and optical properties of $\text{In}{0.25}\text{Ga}{0.75}$N/GaN quantum wells. Our analysis accounts for fluctuations of well width, local alloy composition, strain and built-in field fluctuations as well as Coulomb effects. We find a strong hole and much weaker electron wave function localization in InGaN random alloy quantum wells. The presented calculations show that while the electron states are mainly localized by well-width fluctuations, the holes states are already localized by random alloy fluctuations. These localization effects affect significantly the quantum well optical properties,leading to strong inhomogeneous broadening of the lowest interband transition energy. Our results are compared with experimental literature data.
https://arxiv.org/abs/1501.05482
The recent observation of superconducting state at atomic scale has motivated the pursuit of exotic condensed phases in two-dimensional (2D) systems. Here we report on a superconducting phase in two-monolayer crystalline Ga films epitaxially grown on wide band-gap semiconductor GaN(0001). This phase exhibits a hexagonal structure and only 0.552 nm in thickness, nevertheless, brings about a superconducting transition temperature Tc as high as 5.4 K, confirmed by in situ scanning tunneling spectroscopy, and ex situ electrical magneto-transport and magnetization measurements. The anisotropy of critical magnetic field and Berezinski-Kosterlitz-Thouless-like transition are observed, typical for the 2D superconductivity. Our results demonstrate a novel platform for exploring atomic-scale 2D superconductor, with great potential for understanding of the interface superconductivity.
https://arxiv.org/abs/1403.5936
This article proposes the use of Vector Symbolic Architectures for implementing Hierarchical Graph Neuron, an architecture for memorizing patterns of generic sensor stimuli. The adoption of a Vector Symbolic representation ensures a one-layered design for the approach, while maintaining the previously reported properties and performance characteristics of Hierarchical Graph Neuron, and also improving the noise resistance of the architecture. The proposed architecture enables a linear (with respect to the number of stored entries) time search for an arbitrary sub-pattern.
https://arxiv.org/abs/1501.03784
Learning sparse combinations is a frequent theme in machine learning. In this paper, we study its associated optimization problem in the distributed setting where the elements to be combined are not centrally located but spread over a network. We address the key challenges of balancing communication costs and optimization errors. To this end, we propose a distributed Frank-Wolfe (dFW) algorithm. We obtain theoretical guarantees on the optimization error $\epsilon$ and communication cost that do not depend on the total number of combining elements. We further show that the communication cost of dFW is optimal by deriving a lower-bound on the communication cost required to construct an $\epsilon$-approximate solution. We validate our theoretical analysis with empirical studies on synthetic and real-world data, which demonstrate that dFW outperforms both baselines and competing methods. We also study the performance of dFW when the conditions of our analysis are relaxed, and show that dFW is fairly robust.
http://arxiv.org/abs/1404.2644
Solid material in protoplanetary discs will suffer one of two fates after the epoch of planet formation; either being bound up into planetary bodies, or remaining in smaller planetesimals to be ground into dust. These end states are identified through detection of sub-stellar companions by periodic radial velocity (or transit) variations of the star, and excess emission at mid- and far-infrared wavelengths, respectively. Since the material that goes into producing the observable outcomes of planet formation is the same, we might expect these components to be related both to each other and their host star. Heretofore, our knowledge of planetary systems around other stars has been strongly limited by instrumental sensitivity. In this work, we combine observations at far-infrared wavelengths by IRAS, Spitzer, and Herschel with limits on planetary companions derived from non-detections in the 16-year Anglo-Australian Planet Search to clarify the architectures of these (potential) planetary systems and search for evidence of correlations between their constituent parts. We find no convincing evidence of such correlations, possibly owing to the dynamical history of the disk systems, or the greater distance of the planet-search targets. Our results place robust limits on the presence of Jupiter analogs which, in concert with the debris disk observations, provides insights on the small-body dynamics of these nearby systems.
https://arxiv.org/abs/1501.02508
It has become apparent that a Gaussian center bias can serve as an important prior for visual saliency detection, which has been demonstrated for predicting human eye fixations and salient object detection. Tseng et al. have shown that the photographer’s tendency to place interesting objects in the center is a likely cause for the center bias of eye fixations. We investigate the influence of the photographer’s center bias on salient object detection, extending our previous work. We show that the centroid locations of salient objects in photographs of Achanta and Liu’s data set in fact correlate strongly with a Gaussian model. This is an important insight, because it provides an empirical motivation and justification for the integration of such a center bias in salient object detection algorithms and helps to understand why Gaussian models are so effective. To assess the influence of the center bias on salient object detection, we integrate an explicit Gaussian center bias model into two state-of-the-art salient object detection algorithms. This way, first, we quantify the influence of the Gaussian center bias on pixel- and segment-based salient object detection. Second, we improve the performance in terms of F1 score, Fb score, area under the recall-precision curve, area under the receiver operating characteristic curve, and hit-rate on the well-known data set by Achanta and Liu. Third, by debiasing Cheng et al.’s region contrast model, we exemplarily demonstrate that implicit center biases are partially responsible for the outstanding performance of state-of-the-art algorithms. Last but not least, as a result of debiasing Cheng et al.’s algorithm, we introduce a non-biased salient object detection method, which is of interest for applications in which the image data is not likely to have a photographer’s center bias (e.g., image data of surveillance cameras or autonomous robots).
https://arxiv.org/abs/1501.03383
In recent years development in the area of Single Board Computing has been advancing rapidly. At Wolters Kluwer’s Corporate Legal Services Division a prototyping effort was undertaken to establish the utility of such devices for practical and general computing needs. This paper presents the background of this work, the design and construction of a 64 core 96 GHz cluster, and their possibility of yielding approximately 400 GFLOPs from a set of small footprint InSignal boards created for just over $2,300. Additionally this paper discusses the software environment on the cluster, the use of a standard Beowulf library and its operation, as well as other software application uses including Elastic Search and ownCloud. Finally, consideration will be given to the future use of such technologies in a business setting in order to introduce new Open Source technologies, reduce computing costs, and improve Time to Market. Index Terms: Single Board Computing, Raspberry Pi, InSignal Exynos 5420, Linaro Ubuntu Linux, High Performance Computing, Beowulf clustering, Open Source, MySQL, MongoDB, ownCloud, Computing Architectures, Parallel Computing, Cluster Computing
https://arxiv.org/abs/1501.00039
Prediction of protein secondary structure from the amino acid sequence is a classical bioinformatics problem. Common methods use feed forward neural networks or SVMs combined with a sliding window, as these models does not naturally handle sequential data. Recurrent neural networks are an generalization of the feed forward neural network that naturally handle sequential data. We use a bidirectional recurrent neural network with long short term memory cells for prediction of secondary structure and evaluate using the CB513 dataset. On the secondary structure 8-class problem we report better performance (0.674) than state of the art (0.664). Our model includes feed forward networks between the long short term memory cells, a path that can be further explored.
https://arxiv.org/abs/1412.7828
The reconstructions of the Ga polarity GaN(0 0 0 1) surface with and without trace amounts of arsenic and prepared by molecular beam epitaxy (MBE) have been studied with in situ reflection high-energy electron diffraction (RHEED) and scanning tunneling microscopy (STM). Various reconstructions are observed with RHEED by analyzing patterns while the substrate is exposed to a fixed NH3 flux or after depositing known amounts of Ga as a function of substrate temperature. In situ STM images reveal that only a few of these reconstructions yield long-range periodicity in real space. The controversial role of arsenic on Ga induced reconstructions was also investigated using two independent MBE chambers and X-ray photoelectron spectroscopy.
https://arxiv.org/abs/1501.00143
We present the results obtained by using an evolution of our CUDA-based solution for the exploration, via a Breadth First Search, of large graphs. This latest version exploits at its best the features of the Kepler architecture and relies on a combination of techniques to reduce both the number of communications among the GPUs and the amount of exchanged data. The final result is a code that can visit more than 800 billion edges in a second by using a cluster equipped with 4096 Tesla K20X GPUs.
https://arxiv.org/abs/1408.1605
This dissertation explores the roles of polarities and focussing in various aspects of Computational Logic. These concepts play a key role in the the interpretation of proofs as programs, a.k.a. the Curry-Howard correspondence, in the context of classical logic. Arising from linear logic, they allow the construction of meaningful semantics for cut-elimination in classical logic, some of which relate to the Call-by-Name and Call-by-Value disciplines of functional programming. The first part of this dissertation provides an introduction to these interpretations, highlighting the roles of polarities and focussing. For instance: proofs of positive formulae provide structured data, while proofs of negative formulae consume such data; focussing allows the description of the interaction between the two kinds of proofs as pure pattern-matching. This idea is pushed further in the second part of this dissertation, and connected to realisability semantics, where the structured data is interpreted algebraically, and the consumption of such data is modelled with the use of an orthogonality relation. Most of this part has been proved in the Coq proof assistant. Polarities and focussing were also introduced with applications to logic programming in mind, where computation is proof-search. In the third part of this dissertation, we push this idea further by exploring the roles that these concepts can play in other applications of proof-search, such as theorem proving and more particularly automated reasoning. We use these concepts to describe the main algorithm of SAT-solvers and SMT-solvers: DPLL. We then describe the implementation of a proof-search engine called Psyche. Its architecture, based on the concept of focussing, offers a platform where smart techniques from automated reasoning (or a user interface) can safely and trustworthily be implemented via the use of an API.
https://arxiv.org/abs/1412.6781
We present an algorithm to search for the faint spectrum of a second star mixed with the spectrum of a brighter star in high resolution spectra. We model optical stellar spectra as the sum of two input spectra drawn from a vast library of stars throughout the H-R diagram. From typical spectra having resolution of R=60,000, we are able to detect companions as faint as 1% relative to the primary star in approximately the V and R bandpasses of photometry. We are also able to find evidence for triple and quadruple systems, given that any additional companions are sufficiently bright. The precise threshold percentage depends on the SNR of the spectrum and the properties of the two stars. For cases of non-detection, we place a limit on the brightness of any potential companions. This algorithm is useful for detecting both faint orbiting companions and background stars that are angularly close to a foreground target star. The size of the entrance slit to the spectrometer, 0.87 x 3 arcsec (typically), sets the angular domain within which the second star can be detected. We analyzed Keck-HIRES spectra of 1160 California Kepler Survey objects of interest (KOI) searching for the secondary spectra, with the two goals of alerting the community to two possible host stars of the transiting planet and to dilution of the light curve. We report 63 California Kepler Survey objects of interest showing spectroscopic evidence of a secondary star.
https://arxiv.org/abs/1412.5259
At this moment Autonomous cars are probably the biggest and most talked about technology in the Robotics Research Community. In spite of great technological advances over past few years a full edged autonomous car is still far from reality. This article talks about the existing system and discusses the possibility of a Computer Vision enabled driving being superior than the LiDar based system. A detailed overview of privacy violations that might arise from autonomous driving has been discussed in detail both from a technical as well as legal perspective. It has been proved through evidence and arguments that efficient and accurate estimation and efficient solution of the constraint satisfaction problem addressed in the case of autonomous cars are negatively correlated with the preserving the privacy of the user. It is a very difficult trade-off since both are very important aspects and has to be taken into account. The fact that one cannot compromise with the safety issues of the car makes it inevitable to run into serious privacy concerns that might have adverse social and political effects.
http://arxiv.org/abs/1412.5207
A comprehensive first-principles study of the energetics, electronic and magnetic properties of Co-doped GaN(0001) thin films are presented and the effect of surface structure on the magnetic coupling between Co atoms is demonstrated. It is found that Co atoms prefer to substitute the surface Ga sites in different growth conditions. In particular, a CoN/GaN interface structure with Co atoms replacing the first Ga layer is preferred under N-rich and moderately Ga-rich conditions, while CoGax/GaN interface is found to be energetically stable under extremely Ga-rich conditions. It’s worth noted that the antiferromagnetic coupling between Co atoms is favorable in clean GaN(0001) surface, but the existence of FM would be expected to occur as Co concentration increased in Ga-bilayer GaN(0001) surface. Our study provides the theoretical understanding for experimental research on Co-doped GaN films and might promise the Co:GaN system potential applications in spin injection devices.
https://arxiv.org/abs/1408.6033
Using the dielectric continuum (DC) and three-dimensional phonon (3DP) models, energy relaxation of the hot electrons in the quasi-two-dimensional channel of lattice-matched InAlN/AlN/GaN heterostructures is studied theoretically. The electron power dissipation and energy relaxation time due to both half-space and interface phonons are calculated as functions of the electron temperature $T_e$ using a variety of phonon lifetime values from experiment, and then compared with those evaluated by the 3DP model. The 3DP model yields very close results to the DC model: with no hot phonons or screening the power loss calculated from the 3DP model is 5% smaller than the DC power dissipation, whereas slightly larger 3DP power loss (by less than 4% with a phonon lifetime from 0.1 to 1 ps) is obtained throughout the electron temperature range from room temperature to 2500 K after including both the hot-phonon effect (HPE) and screening. Very close results are obtained also for energy relaxation time with the two phonon models (within a 5% of deviation). However the 3DP model is found to underestimate the HPE by 9%. The Mori-Ando sum rule is restored by which it is proved that the power dissipation values obtained from the DC and 3DP models are in general different in the pure phonon emission process, except when scattering with interface phonons is sufficiently weak, or when the degenerate modes condition is imposed, which is also consistent with Register’s scattering rate sum rule. Our calculation with both phonon models has obtained a great fall in energy relaxation time at low electron temperatures ($T_e<$ 750 K) and slow decrease at the high temperatures. The calculated temperature dependence of the relaxation time and the high-temperature relaxation time $\sim$0.09 ps are in good agreement with experimental results.
https://arxiv.org/abs/1412.3312
Salient object detection or salient region detection models, diverging from fixation prediction models, have traditionally been dealing with locating and segmenting the most salient object or region in a scene. While the notion of most salient object is sensible when multiple objects exist in a scene, current datasets for evaluation of saliency detection approaches often have scenes with only one single object. We introduce three main contributions in this paper: First, we take an indepth look at the problem of salient object detection by studying the relationship between where people look in scenes and what they choose as the most salient object when they are explicitly asked. Based on the agreement between fixations and saliency judgments, we then suggest that the most salient object is the one that attracts the highest fraction of fixations. Second, we provide two new less biased benchmark datasets containing scenes with multiple objects that challenge existing saliency models. Indeed, we observed a severe drop in performance of 8 state-of-the-art models on our datasets (40% to 70%). Third, we propose a very simple yet powerful model based on superpixels to be used as a baseline for model evaluation and comparison. While on par with the best models on MSRA-5K dataset, our model wins over other models on our data highlighting a serious drawback of existing models, which is convoluting the processes of locating the most salient object and its segmentation. We also provide a review and statistical analysis of some labeled scene datasets that can be used for evaluating salient object detection models. We believe that our work can greatly help remedy the over-fitting of models to existing biased datasets and opens new venues for future research in this fast-evolving field.
https://arxiv.org/abs/1412.5027
We present a method to perform first-pass large vocabulary continuous speech recognition using only a neural network and language model. Deep neural network acoustic models are now commonplace in HMM-based speech recognition systems, but building such systems is a complex, domain-specific task. Recent work demonstrated the feasibility of discarding the HMM sequence modeling framework by directly predicting transcript text from audio. This paper extends this approach in two ways. First, we demonstrate that a straightforward recurrent neural network architecture can achieve a high level of accuracy. Second, we propose and evaluate a modified prefix-search decoding algorithm. This approach to decoding enables first-pass speech recognition with a language model, completely unaided by the cumbersome infrastructure of HMM-based systems. Experiments on the Wall Street Journal corpus demonstrate fairly competitive word error rates, and the importance of bi-directional network recurrence.
https://arxiv.org/abs/1408.2873
The effective use of parallel computing resources to speed up algorithms in current multi-core parallel architectures remains a difficult challenge, with ease of programming playing a key role in the eventual success of various parallel architectures. In this paper we consider an alternative view of parallelism in the form of an ultra-wide word processor. We introduce the Ultra-Wide Word architecture and model, an extension of the word-RAM model that allows for constant time operations on thousands of bits in parallel. Word parallelism as exploited by the word-RAM model does not suffer from the more difficult aspects of parallel programming, namely synchronization and concurrency. For the standard word-RAM algorithms, the speedups obtained are moderate, as they are limited by the word size. We argue that a large class of word-RAM algorithms can be implemented in the Ultra-Wide Word model, obtaining speedups comparable to multi-threaded computations while keeping the simplicity of programming of the sequential RAM model. We show that this is the case by describing implementations of Ultra-Wide Word algorithms for dynamic programming and string searching. In addition, we show that the Ultra-Wide Word model can be used to implement a nonstandard memory architecture, which enables the sidestepping of lower bounds of important data structure problems such as priority queues and dynamic prefix sums. While similar ideas about operating on large words have been mentioned before in the context of multimedia processors [Thorup 2003], it is only recently that an architecture like the one we propose has become feasible and that details can be worked out.
https://arxiv.org/abs/1411.7359
Latent Dirichlet allocation (LDA) is a popular topic modeling technique in academia but less so in industry, especially in large-scale applications involving search engine and online advertising systems. A main underlying reason is that the topic models used have been too small in scale to be useful; for example, some of the largest LDA models reported in literature have up to $10^3$ topics, which cover difficultly the long-tail semantic word sets. In this paper, we show that the number of topics is a key factor that can significantly boost the utility of topic-modeling systems. In particular, we show that a “big” LDA model with at least $10^5$ topics inferred from $10^9$ search queries can achieve a significant improvement on industrial search engine and online advertising systems, both of which serving hundreds of millions of users. We develop a novel distributed system called Peacock to learn big LDA models from big data. The main features of Peacock include hierarchical distributed architecture, real-time prediction and topic de-duplication. We empirically demonstrate that the Peacock system is capable of providing significant benefits via highly scalable LDA topic models for several industrial applications.
https://arxiv.org/abs/1405.4402
Feature or interest points typically use information aggregation in 2D patches which does not remain stable at object boundaries when there is object motion against a significantly varying background. Level or iso-intensity curves are much more stable under such conditions, especially the longer ones. In this paper, we identify stable portions on long iso-curves and detect corners on them. Further, the iso-curve associated with a corner is used to discard portions from the background and improve matching. Such CoMIC (Corners on Maximally-stable Iso-intensity Curves) points yield superior results at the object boundary regions compared to state-of-the-art detectors while performing comparably at the interior regions as well. This is illustrated in exhaustive matching experiments for both boundary and non-boundary regions in applications such as stereo and point tracking for structure from motion in video sequences.
https://arxiv.org/abs/1412.1957
In this work, we investigate the use of sparsity-inducing regularizers during training of Convolution Neural Networks (CNNs). These regularizers encourage that fewer connections in the convolution and fully connected layers take non-zero values and in effect result in sparse connectivity between hidden units in the deep network. This in turn reduces the memory and runtime cost involved in deploying the learned CNNs. We show that training with such regularization can still be performed using stochastic gradient descent implying that it can be used easily in existing codebases. Experimental evaluation of our approach on MNIST, CIFAR, and ImageNet datasets shows that our regularizers can result in dramatic reductions in memory requirements. For instance, when applied on AlexNet, our method can reduce the memory consumption by a factor of four with minimal loss in accuracy.
https://arxiv.org/abs/1412.1442
We report accurate energetics of defects introduced in GaN on doping with divalent metals, focussing on the technologically important case of Mg doping, using a model which takes into consideration both the effect of hole localisation and dipolar polarisation of the host material, and includes a well-defined reference level. Defect formation and ionisation energies show that divalent dopants are counterbalanced in GaN by nitrogen vacancies and not by holes, which explains both the difficulty in achieving p-type conductivity in GaN and the associated major spectroscopic features, including the ubiquitous 3.46 eV photoluminescence line, a characteristic of all lightly divalent metal-doped GaN materials that has also been shown to occur in pure GaN samples. Our results give a comprehensive explanation for the observed behaviour of GaN doped with low concentrations of divalent metals in good agreement with relevant experiment.
https://arxiv.org/abs/1412.1694
We present microscopic calculations of the absorption spectra for GaN/Al(x)Ga(1-x)N quantum well systems. Whereas the quantum well structures with the parabolic law of dispersion exhibit the usual bleaching of an exciton resonance without shifting a spectral position, the significant red-shift of an exciton peak is found with increasing the electron-hole gas density for a wurtzite quantum well. The energy of the exciton resonance for a wurtzite quantum well is found. The obtained results can be explained by the influence of the valence band structure on quantum confinement effects. The optical gain spectrum in the Hartree-Fock approximation and the Sommerfeld enhancement are calculated. A red shift of the gain spectrum in the Hartree-Fock approximation with respect to the Hartree gain spectrum is found.
https://arxiv.org/abs/1302.2783
We propose an approach to detect flying objects such as UAVs and aircrafts when they occupy a small portion of the field of view, possibly moving against complex backgrounds, and are filmed by a camera that itself moves. Solving such a difficult problem requires combining both appearance and motion cues. To this end we propose a regression-based approach to motion stabilization of local image patches that allows us to achieve effective classification on spatio-temporal image cubes and outperform state-of-the-art techniques. As the problem is relatively new, we collected two challenging datasets for UAVs and Aircrafts, which can be used as benchmarks for flying objects detection and vision-guided collision avoidance.
https://arxiv.org/abs/1411.7715
One of the key advances in resolving the big-data problem has been the emergence of an alternative database technology. Today, classic RDBMS are complemented by a rich set of alternative Data Management Systems (DMS) specially designed to handle the volume, variety, velocity and variability ofBig Data collections; these DMS include NoSQL, NewSQL and Search-based systems. NewSQL is a class of modern relational database management systems (RDBMS) that provide the same scalable performance of NoSQL systems for online transaction processing (OLTP) read-write workloads while still maintaining the ACID guarantees of a traditional database system. This paper discusses about NewSQL data management system; and compares with NoSQL and with traditional database system. This paper covers architecture, characteristics, classification of NewSQL databases for online transaction processing (OLTP) for Big data management. It also provides the list ofpopular NoSQL as well as NewSQL databases in separate categorized tables. This paper compares SQL based RDBMS, NoSQL and NewSQL databases with set of metrics; as well as, addressed some research issues ofNoSQL and NewSQL.
https://arxiv.org/abs/1411.7343
Memories in the brain are separated in two categories: short-term and long-term memories. Long-term memories remain for a lifetime, while short-term ones exist from a few milliseconds to a few minutes. Within short-term memory studies, there is debate about what neural structure could implement it. Indeed, mechanisms responsible for long-term memories appear inadequate for the task. Instead, it has been proposed that short-term memories could be sustained by the persistent activity of a group of neurons. In this work, we explore what topology could sustain short-term memories, not by designing a model from specific hypotheses, but through Darwinian evolution in order to obtain new insights into its implementation. We evolved 10 networks capable of retaining information for a fixed duration between 2 and 11s. Our main finding has been that the evolution naturally created two functional modules in the network: one which sustains the information containing primarily excitatory neurons, while the other, which is responsible for forgetting, was composed mainly of inhibitory neurons. This demonstrates how the balance between inhibition and excitation plays an important role in cognition.
https://arxiv.org/abs/1411.6912
In this paper, we propose a design for novel and experimental cloud computing systems. The proposed system aims at enhancing computational, communicational and annalistic capabilities of road navigation services by merging several independent technologies, namely vision-based embedded navigation systems, prominent Cloud Computing Systems (CCSs) and Vehicular Ad-hoc NETwork (VANET). This work presents our initial investigations by describing the design of a global generic system. The designed system has been experimented with various scenarios of video-based road services. Moreover, the associated architecture has been implemented on a small-scale simulator of an in-vehicle embedded system. The implemented architecture has been experimented in the case of a simulated road service to aid the police agency. The goal of this service is to recognize and track searched individuals and vehicles in a real-time monitoring system remotely connected to moving cars. The presented work demonstrates the potential of our system for efficiently enhancing and diversifying real-time video services in road environments.
https://arxiv.org/abs/1412.6149
In this paper we explore the bi-directional mapping between images and their sentence-based descriptions. We propose learning this mapping using a recurrent neural network. Unlike previous approaches that map both sentences and images to a common embedding, we enable the generation of novel sentences given an image. Using the same model, we can also reconstruct the visual features associated with an image given its visual description. We use a novel recurrent visual memory that automatically learns to remember long-term visual concepts to aid in both sentence generation and visual feature reconstruction. We evaluate our approach on several tasks. These include sentence generation, sentence retrieval and image retrieval. State-of-the-art results are shown for the task of generating novel image descriptions. When compared to human generated captions, our automatically generated captions are preferred by humans over $19.8\%$ of the time. Results are better than or comparable to state-of-the-art results on the image and sentence retrieval tasks for methods using similar visual features.
在本文中，我们探讨图像和基于句子的描述之间的双向映射。我们建议使用循环神经网络来学习这种映射。与将句子和图像映射到通用嵌入的先前方法不同，我们可以在给定图像的情况下生成新的句子。使用相同的模型，我们还可以重建与图像相关的视觉特征，给出其视觉描述。我们使用一种新颖的经常性视觉记忆，自动学习记忆长期的视觉概念，以协助在句子生成和视觉特征重建。我们评估我们的方法在几个任务。这些包括句子生成，句子检索和图像检索。显示了用于生成新颖图像描述的任务的最新结果。与人类生成的字幕相比，我们自动生成的字幕是人类首选的超过$ 19.8 \％$的时间。对于使用相似视觉特征的方法的图像和语句检索任务，结果优于或相当于最新的结果。
https://arxiv.org/abs/1411.5654
Deep Convolutional Neural Networks (CNNs) have gained great success in image classification and object detection. In these fields, the outputs of all layers of CNNs are usually considered as a high dimensional feature vector extracted from an input image and the correspondence between finer level feature vectors and concepts that the input image contains is all-important. However, fewer studies focus on this deserving issue. On considering the correspondence, we propose a novel approach which generates an edited version for each original CNN feature vector by applying the maximum entropy principle to abandon particular vectors. These selected vectors correspond to the unfriendly concepts in each image category. The classifier trained from merged feature sets can significantly improve model generalization of individual categories when training data is limited. The experimental results for classification-based object detection on canonical datasets including VOC 2007 (60.1%), 2010 (56.4%) and 2012 (56.3%) show obvious improvement in mean average precision (mAP) with simple linear support vector machines.
https://arxiv.org/abs/1409.6911
Graph partitioning, a well studied problem of parallel computing has many applications in diversified fields such as distributed computing, social network analysis, data mining and many other domains. In this paper, we introduce FGPGA, an efficient genetic approach for producing feasible graph partitions. Our method takes into account the heterogeneity and capacity constraints of the partitions to ensure balanced partitioning. Such approach has various applications in mobile cloud computing that include feasible deployment of software applications on the more resourceful infrastructure in the cloud instead of mobile hand set. Our proposed approach is light weight and hence suitable for use in cloud architecture. We ensure feasibility of the partitions generated by not allowing over-sized partitions to be generated during the initialization and search. Our proposed method tested on standard benchmark datasets significantly outperforms the state-of-the-art methods in terms of quality of partitions and feasibility of the solutions.
https://arxiv.org/abs/1411.4379
The structural analysis of GaN and Al$x$Ga${1-x}$N/GaN heterostructures grown by metalorganic vapor phase epitaxy in the presence of Mn reveals how Mn affects the growth process, and in particular the incorporation of Al, the morphology of the surface, and the plastic relaxation of Al$x$Ga${1-x}$N on GaN. Moreover, the doping with Mn promotes the formation of layered Al$x$Ga${1-x}$N/GaN superlattice-like heterostructures opening wide perspective for controlling the segregation of ternary alloys during the crystal growth and for fostering the self-assembling of functional layered structures.
https://arxiv.org/abs/1411.3006
Inspired by recent advances in multimodal learning and machine translation, we introduce an encoder-decoder pipeline that learns (a): a multimodal joint embedding space with images and text and (b): a novel language model for decoding distributed representations from our space. Our pipeline effectively unifies joint image-text embedding models with multimodal neural language models. We introduce the structure-content neural language model that disentangles the structure of a sentence to its content, conditioned on representations produced by the encoder. The encoder allows one to rank images and sentences while the decoder can generate novel descriptions from scratch. Using LSTM to encode sentences, we match the state-of-the-art performance on Flickr8K and Flickr30K without using object detections. We also set new best results when using the 19-layer Oxford convolutional network. Furthermore we show that with linear encoders, the learned embedding space captures multimodal regularities in terms of vector space arithmetic e.g. image of a blue car - “blue” + “red” is near images of red cars. Sample captions generated for 800 images are made available for comparison.
受到多模式学习和机器翻译方面的最新进展的启发，我们引入了一个编码器 - 解码器流水线，它学习（a）：图像和文本的多模式联合嵌入空间和（b）：从我们的空间解码分布式表示的新型语言模型。我们的流程有效地将联合图像文本嵌入模型与多模式神经语言模型相结合。我们引入结构 - 内容神经语言模型，将句子的结构解开为其内容，以编码器产生的表示为条件。编码器允许人们排列图像和句子，而解码器可以从头开始产生新的描述。使用LSTM对句子进行编码，我们无需使用对象检测就可以匹配Flickr8K和Flickr30K上的最新性能。当使用19层牛津卷积网络时，我们也创造了新的最佳结果。此外，我们表明，利用线性编码器，学习的嵌入空间以矢量空间运算的形式捕获多模态规则。 蓝色车的图像 - “蓝色”+“红色”是红色车的图像。为800幅图像生成的样本标题可用于比较。
https://arxiv.org/abs/1411.2539
In 2013 Intel introduced the Xeon Phi, a new parallel co-processor board. The Xeon Phi is a cache-coherent many-core shared memory architecture claiming CPU-like versatility, programmability, high performance, and power efficiency. The first published micro-benchmark studies indicate that many of Intel’s claims appear to be true. The current paper is the first study on the Phi of a complex artificial intelligence application. It contains an open source MCTS application for playing tournament quality Go (an oriental board game). We report the first speedup figures for up to 240 parallel threads on a real machine, allowing a direct comparison to previous simulation studies. After a substantial amount of work, we observed that performance scales well up to 32 threads, largely confirming previous simulation results of this Go program, although the performance surprisingly deteriorates between 32 and 240 threads. Furthermore, we report (1) unexpected performance anomalies between the Xeon Phi and Xeon CPU for small problem sizes and small numbers of threads, and (2) that performance is sensitive to scheduling choices. Achieving good performance on the Xeon Phi for complex programs is not straightforward; it requires a deep understanding of (1) search patterns, (2) of scheduling, and (3) of the architecture and its many cores and caches. In practice, the Xeon Phi is less straightforward to program for than originally envisioned by Intel.
https://arxiv.org/abs/1409.4297