Search

Abstract Aiming at the data authenticity and storage problems in the current coordinated scheduling of virtual power plants, as well as the opaque information and high transaction costs, a dual blockchains security mechanism is proposed to solve above problems. In the process of security scheduling, a hybrid attribute proxy re-encryption algorithm based on ciphertext strategy is designed. The algorithm is composed of an identity encryption algorithm and an attribute proxy re-encryption algorithm with ciphertext strategy. Combining blockchain, tamper-proof smart metering equipment can effectively solve data authenticity and confidentiality in the information transmission process of distributed energy. In the research process of the trading mechanism, a continuous double auction mechanism based on reputation is proposed. In order to create a favorable trading atmosphere, reputation-based market segmentation mechanisms are integrated, and participants are divided according to reputation value. Depending on the properties of the stored information, they are divided into the private blockchain (with coordination scheduling information) and the consortium blockchain (with transaction information). The system analysis shows the reliability of the dual blockchains architecture. The communication and calculation costs of the proxy re-encryption algorithm verify the practicability of the proposed scheme. The case analysis of the auction mechanism declares that the mechanism can operate effective in the electricity trading market.

The two-wheeled robot is a nonlinear system of multi-variables, higherorder and strong coupling. This paper presented a PID Controller with Double Loops (PCDL) to control the tilt angle and velocity of a two-wheeled robot. The angle controller is the regular negative feedback, while the velocity control is the positive feedback. The Double Loops work cooperatively to endow the system with strong anti-interference ability. The stability of the whole system is analyzed and the criterion of the system stability is developed. The simulation and experiments showed that the two-wheeled robot can self-balance and move at an expected velocity and the system has strong anti-interference ability.

Computer-Aided-Diagnosis (CADx) systems assist radiologists with identifyingand classifying potentially malignant pulmonary nodules on chest CT scans usingmorphology and texture-based (radiomic) features. However, radiomic featuresare sensitive to differences in acquisitions due to variations in dose levelsand slice thickness. This study investigates the feasibility of generating anormalized scan from heterogeneous CT scans as input. We obtained projectiondata from 40 low-dose chest CT scans, simulating acquisitions at 10%, 25% and50% dose and reconstructing the scans at 1.0mm and 2.0mm slice thickness. A 3Dgenerative adversarial network (GAN) was used to simultaneously normalizereduced dose, thick slice (2.0mm) images to normal dose (100%), thinner slice(1.0mm) images. We evaluated the normalized image quality using peaksignal-to-noise ratio (PSNR), structural similarity index (SSIM) and LearnedPerceptual Image Patch Similarity (LPIPS). Our GAN improved perceptualsimilarity by 35%, compared to a baseline CNN method. Our analysis also showsthat the GAN-based approach led to a significantly smaller error (p-value <0.05) in nine studied radiomic features. These results indicated that GANscould be used to normalize heterogeneous CT images and reduce the variabilityin radiomic feature values.

Most of the existing works in video synthesis focus on generating videosusing adversarial learning. Despite their success, these methods often requireinput reference frame or fail to generate diverse videos from the given datadistribution, with little to no uniformity in the quality of videos that can begenerated. Different from these methods, we focus on the problem of generatingvideos from latent noise vectors, without any reference input frames. To thisend, we develop a novel approach that jointly optimizes the input latent space,the weights of a recurrent neural network and a generator throughnon-adversarial learning. Optimizing for the input latent space along with thenetwork weights allows us to generate videos in a controlled environment, i.e.,we can faithfully generate all videos the model has seen during the learningprocess as well as new unseen videos. Extensive experiments on threechallenging and diverse datasets well demonstrate that our approach generatessuperior quality videos compared to the existing state-of-the-art methods.

Automatic music generation is an interdisciplinary research topic thatcombines computational creativity and semantic analysis of music to createautomatic machine improvisations. An important property of such a system isallowing the user to specify conditions and desired properties of the generatedmusic. In this paper we designed a model for composing melodies given a userspecified symbolic scenario combined with a previous music context. We addmanual labeled vectors denoting external music quality in terms of chordfunction that provides a low dimensional representation of the harmonic tensionand resolution. Our model is capable of generating long melodies by regarding8-beat note sequences as basic units, and shares consistent rhythm patternstructure with another specific song. The model contains two stages andrequires separate training where the first stage adopts a ConditionalVariational Autoencoder (C-VAE) to build a bijection between note sequences andtheir latent representations, and the second stage adopts long short-termmemory networks (LSTM) with structural conditions to continue writing futuremelodies. We further exploit the disentanglement technique via C-VAE to allowmelody generation based on pitch contour information separately fromconditioning on rhythm patterns. Finally, we evaluate the proposed model usingquantitative analysis of rhythm and the subjective listening study. Resultsshow that the music generated by our model tends to have salient repetitionstructures, rich motives, and stable rhythm patterns. The ability to generatelonger and more structural phrases from disentangled representations combinedwith semantic scenario specification conditions shows a broad application ofour model.

We study a wireless ad-hoc sensor network (WASN) where $N$ sensors gatherdata from the surrounding environment and transmit their sensed information to$M$ fusion centers (FCs) via multi-hop wireless communications. This nodedeployment problem is formulated as an optimization problem to make a trade-offbetween the sensing uncertainty and energy consumption of the network. Ourprimary goal is to find an optimal deployment of sensors and FCs to minimize aLagrange combination of the sensing uncertainty and energy consumption. Tosupport arbitrary routing protocols in WASNs, the routing-dependent necessaryconditions for the optimal deployment are explored. Based on these necessaryconditions, we propose a routing-aware Lloyd algorithm to optimize nodedeployment. Simulation results show that, on average, the proposed algorithmoutperforms the existing deployment algorithms.

We propose a novel analytical framework for evaluating the coverageperformance of a millimeter wave (mmWave) cellular network where idle userequipments (UEs) act as relays. In this network, the base station (BS) adoptseither the direct mode to transmit to the destination UE, or the relay mode ifthe direct mode fails, where the BS transmits to the relay UE and then therelay UE transmits to the destination UE. To address the drastic rotationalmovements of destination UEs in practice, we propose to adopt selectioncombining at destination UEs. New expression is derived for thesignal-to-interference-plus-noise ratio (SINR) coverage probability of thenetwork. Using numerical results, we first demonstrate the accuracy of our newexpression. Then we show that ignoring spatial correlation, which has beencommonly adopted in the literature, leads to severe overestimation of the SINRcoverage probability. Furthermore, we show that introducing relays into ammWave cellular network vastly improves the coverage performance. In addition,we show that the optimal BS density maximizing the SINR coverage probabilitycan be determined by using our analysis.

Automatic synthesis of realistic eye images with pre- scribed gaze direction is important for multiple application domains. We introduce EyeGAN, an algorithm to generate eye images in the style of a desired target domain, that in- herit annotations available in images from a source domain. EyeGAN takes in input ternary masks, which are used as domain-independent proxies for gaze direction. We eval- uate EyeGAN against competing eye image synthesis al- gorithms by measuring a specific gaze consistency index. In addition, we present results from multiple experiments (involving eye region segmentation, pupil localization, and gaze direction estimation) showing that the use of EyeGAN- generated images with inherited annotations for network training leads to superior performances compared to other domain transfer algorithms.

Spike-based communication between biological neurons is sparse andunreliable. This enables the brain to process visual information from the eyesefficiently. Taking inspiration from biology, artificial spiking neuralnetworks coupled with silicon retinas attempt to model these computations.Recent findings in machine learning allowed the derivation of a family ofpowerful synaptic plasticity rules approximating backpropagation for spikingnetworks. Are these rules capable of processing real-world visual sensory data?In this paper, we evaluate the performance of Event-Driven RandomBack-Propagation (eRBP) at learning representations from event streams providedby a Dynamic Vision Sensor (DVS). First, we show that eRBP matchesstate-of-the-art performance on the DvsGesture dataset with the addition of asimple covert attention mechanism. By remapping visual receptive fieldsrelatively to the center of the motion, this attention mechanism providestranslation invariance at low computational cost compared to convolutions.Second, we successfully integrate eRBP in a real robotic setup, where a roboticarm grasps objects according to detected visual affordances. In this setup,visual information is actively sensed by a DVS mounted on a robotic headperforming microsaccadic eye movements. We show that our method classifiesaffordances within 100ms after microsaccade onset, which is comparable to humanperformance reported in behavioral study. Our results suggest that advances inneuromorphic technology and plasticity rules enable the development ofautonomous robots operating at high speed and low energy consumption.

Public transit stations and hubs can be diffi- cult to navigate by people who are blind or have low vision, and by people with cognitive impairment. We are building a system, named RouteMe2, that provides microrouting and guidance in these environments. A critical component of this system is self-localization. In this paper, we present a system for self-localization in outdoor places (such as a train station), in which GPS signal is available but, due to shading from nearby buildings, often unreliable. We propose a new approach that uses a small number of Blue- tooth low energy (BLE) beacons to increase self-localization accuracy by means of statistical fusion with data from GPS, paired with a Bayes discrete filter tracker. A number of experiments were conducted at San Jose Diridon light rail station to quantitatively assess the performance of the proposed system.

We present a model for capturing musical features and creating novelsequences of music, called the Convolutional Variational Recurrent NeuralNetwork. To generate sequential data, the model uses an encoder-decoderarchitecture with latent probabilistic connections to capture the hiddenstructure of music. Using the sequence-to-sequence model, our generative modelcan exploit samples from a prior distribution and generate a longer sequence ofmusic. We compare the performance of our proposed model with other types ofNeural Networks using the criteria of Information Rate that is implemented byVariable Markov Oracle, a method that allows statistical characterization ofmusical information dynamics and detection of motifs in a song. Our resultssuggest that the proposed model has a better statistical resemblance to themusical structure of the training data, which improves the creation of newsequences of music in the style of the originals.

Deep learning is a promising tool to determine the physical model thatdescribes our universe. To handle the considerable computational cost of thisproblem, we present CosmoFlow: a highly scalable deep learning applicationbuilt on top of the TensorFlow framework. CosmoFlow uses efficientimplementations of 3D convolution and pooling primitives, together withimprovements in threading for many element-wise operations, to improve trainingperformance on Intel(C) Xeon Phi(TM) processors. We also utilize the Cray PEMachine Learning Plugin for efficient scaling to multiple nodes. We demonstratefully synchronous data-parallel training on 8192 nodes of Cori with 77%parallel efficiency, achieving 3.5 Pflop/s sustained performance. To ourknowledge, this is the first large-scale science application of the TensorFlowframework at supercomputer scale with fully-synchronous training. Theseenhancements enable us to process large 3D dark matter distribution and predictthe cosmological parameters $\Omega_M$, $\sigma_8$ and n$_s$ with unprecedentedaccuracy.

We propose a new order preserving bilinear framework that exploitslow-resolution video for person detection in a multi-modal setting using deepneural networks. In this setting cameras are strategically placed such thatless robust sensors, e.g. geophones that monitor seismic activity, are locatedwithin the field of views (FOVs) of cameras. The primary challenge is beingable to leverage sufficient information from videos where there are less than40 pixels on targets, while also taking advantage of less discriminativeinformation from other modalities, e.g. seismic. Unlike state-of-the-artmethods, our bilinear framework retains spatio-temporal order when computingthe vector outer products between pairs of features. Despite the highdimensionality of these outer products, we demonstrate that our orderpreserving bilinear framework yields better performance than recent orderlessbilinear models and alternative fusion methods.

We propose a new technique for the accurate segmenta- tion of text strokes from an image. The algorithm takes in a cropped image containing a word. It first performs a coarse segmentation using a Fully Convolutional Network (FCN). While not accurate, this initial segmentation can usually identify most of the text stroke content even in difficult situ- ations, with uneven lighting and non-uniform background. The segmentation is then refined using a fully connected Conditional Random Field (CRF) with a novel kernel defini- tion that includes stroke width information. In order to train the network, we created a new synthetic data set with 100K text images. Tested against standard benchmarks with pixel- level annotation (ICDAR 2003, ICDAR 2011, and SVT) our algorithm outperforms the state of the art by a noticeable margin.

We present a novel algorithm for geometry and camera pose reconstruction from image sequences that is specialized for indoor Manhattan scenes. Unlike general-purpose SfM/SLAM, our system represents geometric primitives in terms of canonically oriented planes. The algorithm starts by computing multi-planar segmentation and motion estimation from image pairs using constrained homographies. It then proceeds to recover the relative scale at each frame and to determine chains of match clusters, where each cluster is associated with a plane in the scene. Motion and scene geometry (expressed in terms of planar models) are then optimized using a novel formulation of Bundle Adjustment. Compared with other state-of-the-art SfM/SLAM algorithms, our technique is shown to produce superior and realistic surface reconstruction for a monocular indoor scene.

In this paper, we focus on activating only a few sensors, among manyavailable, to estimate the state of a stochastic process of interest. Thisproblem is important in applications such as target tracking and simultaneouslocalization and mapping (SLAM). It is challenging since it involves stochasticsystems whose evolution is largely unknown, sensors with nonlinearmeasurements, and limited operational resources that constrain the number ofactive sensors at each measurement step. We provide an algorithm applicable togeneral stochastic processes and nonlinear measurements whose time complexityis linear in the planning horizon and whose performance is a multiplicativefactor 1/2 away from the optimal performance. This is notable because thealgorithm offers a significant computational advantage over the polynomial-timealgorithm that achieves the best approximation factor 1/e. In addition, forimportant classes of Gaussian processes and nonlinear measurements corruptedwith Gaussian noise, our algorithm enjoys the same time complexity as even thestate-of-the-art algorithms for linear systems and measurements. We achieve ourresults by proving two properties for the entropy of the batch state vectorconditioned on the measurements: a) it is supermodular in the choice of thesensors; b) it has a sparsity pattern (involves block tri-diagonal matrices)that facilitates its evaluation at each sensor set.