arXiv subjects
Publications and source records attributed to Yu Cao.
We prove the exponential convergence to the equilibrium, quantified by Rényi divergence, of the solution of the Fokker-Planck equation with drift given by the gradient of a strictly convex potential. This extends the classical exponential decay result on the relative entropy for the same equation.
In this paper, we extend the dynamical low-rank approximation method to the space of finite signed measures. Under this framework, we derive stochastic low-rank dynamics for stochastic differential equations (SDEs) coming from classical stochastic dynamics or unraveling of Lindblad quantum master equations. We justify the proposed method by error analysis and also numerical examples for applications in solving high-dimensional SDE, stochastic Burgers' equation, and high-dimensional Lindblad equation.
Neuromorphic engineering (NE) encompasses a diverse range of approaches to information processing that are inspired by neurobiological systems, and this feature distinguishes neuromorphic systems from conventional computing systems. The brain has evolved over billions of years to solve difficult engineering problems by using efficient, parallel, low-power computation. The goal of NE is to design systems capable of brain-like computation. Numerous large-scale neuromorphic projects have emerged recently. This interdisciplinary field was listed among the top 10 technology breakthroughs of 2014 by the MIT Technology Review and among the top 10 emerging technologies of 2015 by the World Economic Forum. NE has two-way goals: one, a scientific goal to understand the computational properties of biological neural systems by using models implemented in integrated circuits (ICs); second, an engineering goal to exploit the known properties of biological systems to design and implement efficient devices for engineering applications. Building hardware neural emulators can be extremely useful for simulating large-scale neural models to explain how intelligent behavior arises in the brain. The principle advantages of neuromorphic emulators are that they are highly energy efficient, parallel and distributed, and require a small silicon area. Thus, compared to conventional CPUs, these neuromorphic emulators are beneficial in many engineering applications such as for the porting of deep learning algorithms for various recognitions tasks. In this review article, we describe some of the most significant neuromorphic spiking emulators, compare the different architectures and approaches used by them, illustrate their advantages and drawbacks, and highlight the capabilities that each can deliver to neural modelers.
This paper introduces a deep learning enabled generative sensing framework which integrates low-end sensors with computational intelligence to attain a high recognition accuracy on par with that attained with high-end sensors. The proposed generative sensing framework aims at transforming low-end, low-quality sensor data into higher quality sensor data in terms of achieved classification accuracy. The low-end data can be transformed into higher quality data of the same modality or into data of another modality. Different from existing methods for image generation, the proposed framework is based on discriminative models and targets to maximize the recognition accuracy rather than a similarity measure. This is achieved through the introduction of selective feature regeneration in a deep neural network (DNN). The proposed generative sensing will essentially transform low-quality sensor data into high-quality information for robust perception. Results are presented to illustrate the performance of the proposed framework.
Electrolyte gating is widely used to induce large carrier density modulation on solid surfaces to explore various properties. Most of past works have attributed the charge modulation to electrostatic field effect. However, some recent reports have argued that the electrolyte gating effect in VO2, TiO2 and SrTiO3 originated from field-induced oxygen vacancy formation. This gives rise to a controversy about the gating mechanism, and it is therefore vital to reveal the relationship between the role of electrolyte gating and the intrinsic properties of materials. Here, we report entirely different mechanisms of electrolyte gating on two high-Tc cuprates, NdBa2Cu3O7-δ (NBCO) and Pr2-xCexCuO4 (PCCO), with different crystal structures. We show that field-induced oxygen vacancy formation in CuO chains of NBCO plays the dominant role while it is mainly an electrostatic field effect in the case of PCCO. The possible reason is that NBCO has mobile oxygen in CuO chains while PCCO does not. Our study helps clarify the controversy relating to the mechanism of electrolyte gating, leading to a better understanding of the role of oxygen electro migration which is very material specific.
We present a new back propagation based training algorithm for discrete-time spiking neural networks (SNN). Inspired by recent deep learning algorithms on binarized neural networks, binary activation with a straight-through gradient estimator is used to model the leaky integrate-fire spiking neuron, overcoming the difficulty in training SNNs using back propagation. Two SNN training algorithms are proposed: (1) SNN with discontinuous integration, which is suitable for rate-coded input spikes, and (2) SNN with continuous integration, which is more general and can handle input spikes with temporal information. Neuromorphic hardware designed in 40nm CMOS exploits the spike sparsity and demonstrates high classification accuracy (>98% on MNIST) and low energy (48.4-773 nJ/image).
For multi-level open quantum system, the interaction between different levels could pose challenge to understand the quantum system both analytically and numerically. In this work, we study the approximation of the dynamics of the Anderson-Holstein model, as a model of multi-level open quantum system, by Redfield and Lindblad equations. Both equations have a desirable property that if the density operators for different levels is diagonal initially, they remain to be diagonal for any time. Thanks to this nice property, the semi-classical limit of both Redfield and Lindblad equations could be derived explicitly; the resulting classical master equations share similar structures of transport and hopping terms. The Redfield and Lindblad equations are also compared from the angle of time dependent perturbation theory.
In recent years, Deep Learning (DL) has found great success in domains such as multimedia understanding. However, the complex nature of multimedia data makes it difficult to develop DL-based software. The state-of-the art tools, such as Caffe, TensorFlow, Torch7, and CNTK, while are successful in their applicable domains, are programming libraries with fixed user interface, internal representation, and execution environment. This makes it difficult to implement portable and customized DL applications. In this paper, we present DeepDSL, a domain specific language (DSL) embedded in Scala, that compiles deep networks written in DeepDSL to Java source code. Deep DSL provides (1) intuitive constructs to support compact encoding of deep networks; (2) symbolic gradient derivation of the networks; (3) static analysis for memory consumption and error detection; and (4) DSL-level optimization to improve memory and runtime efficiency. DeepDSL programs are compiled into compact, efficient, customizable, and portable Java source code, which operates the CUDA and CUDNN interfaces running on Nvidia GPU via a Java Native Interface (JNI) library. We evaluated DeepDSL with a number of popular DL networks. Our experiments show that the compiled programs have very competitive runtime performance and memory efficiency compared to the existing libraries.
Worldwide, in 2014, more than 1.9 billion adults, 18 years and older, were overweight. Of these, over 600 million were obese. Accurately documenting dietary caloric intake is crucial to manage weight loss, but also presents challenges because most of the current methods for dietary assessment must rely on memory to recall foods eaten. The ultimate goal of our research is to develop computer-aided technical solutions to enhance and improve the accuracy of current measurements of dietary intake. Our proposed system in this paper aims to improve the accuracy of dietary assessment by analyzing the food images captured by mobile devices (e.g., smartphone). The key technique innovation in this paper is the deep learning-based food image recognition algorithms. Substantial research has demonstrated that digital imaging accurately estimates dietary intake in many environments and it has many advantages over other methods. However, how to derive the food information (e.g., food type and portion size) from food image effectively and efficiently remains a challenging and open research problem. We propose a new Convolutional Neural Network (CNN)-based food image recognition algorithm to address this problem. We applied our proposed approach to two real-world food image data sets (UEC-256 and Food-101) and achieved impressive results. To the best of our knowledge, these results outperformed all other reported work using these two data sets. Our experiments have demonstrated that the proposed approach is a promising solution for addressing the food image recognition problem. Our future work includes further improving the performance of the algorithms and integrating our system into a real-world mobile and cloud computing-based system to enhance the accuracy of current measurements of dietary intake.
Deep neural networks are typically represented by a much larger number of parameters than shallow models, making them prohibitive for small footprint devices. Recent research shows that there is considerable redundancy in the parameter space of deep neural networks. In this paper, we propose a method to compress deep neural networks by using the Fisher Information metric, which we estimate through a stochastic optimization method that keeps track of second-order information in the network. We first remove unimportant parameters and then use non-uniform fixed point quantization to assign more bits to parameters with higher Fisher Information estimates. We evaluate our method on a classification task with a convolutional neural network trained on the MNIST data set. Experimental results show that our method outperforms existing methods for both network pruning and quantization.
We consider the noise-induced transitions in the randomly perturbed discrete logistic map from a linearly stable periodic orbit consisting of T periodic points. The traditional large deviation theory and asymptotic analysis for small noise limit as well as the derived quasi-potential can not distinguish the quantitative difference in noise-induced stochastic instabilities of these T periodic points. We generalize the transition path theory to the discrete-time continuous-space stochastic process to attack this problem. As a first criterion of quantifying the relative instability among T periodic points, we compare the distribution of the last passage locations in the transitions from the whole periodic orbit to a prescribed set far away. This distribution is related to the contributions to the transition rate from each periodic points. The second criterion is based on the competency of the transition paths associated with each periodic point. Both criteria utilise the reactive probability current in the transition path theory. Our numerical results for the logistic map reveal the transition mechanism of escaping from the stable periodic orbit and identify which peri- odic point is more prone to lose stability so as to make successful transitions under random perturbations.
We propose an application-layer forward error correction (AL-FEC) code rate allocation scheme to maximize the quality of experience (QoE) of a video multicast. The allocation dynamically assigns multicast clients to the quality layers of a scalable video bitstream, based on their heterogeneous channel qualities and video playback capabilities. Normalized mean opinion score (NMOS) is employed to value the client's quality of experience across various possible adaptations of a multilayer video, coded using mixed spatial-temporal-amplitude scalability. The scheme provides assurance of reception of the video layers using fountain coding and effectively allocates coding rates across the layers to maximize a multicast utility measure. An advantageous feature of the proposed scheme is that the complexity of the optimization is independent of the number of clients. Additionally, a convex formulation is proposed that attains close to the best performance and offers a reliable alternative when further reduction in computational complexity is desired. The optimization is extended to perform suppression of QoE fluctuations for clients with marginal channel qualities. The scheme offers a means to trade-off service utility for the entire multicast group and clients with the worst channels. According to the simulation results, the proposed optimization framework is robust against source rate variations and limited amount of client feedback.
This paper introduces a new method to solve the cross-domain recognition problem. Different from the traditional domain adaption methods which rely on a global domain shift for all classes between source and target domain, the proposed method is more flexible to capture individual class variations across domains. By adopting a natural and widely used assumption -- "the data samples from the same class should lay on a low-dimensional subspace, even if they come from different domains", the proposed method circumvents the limitation of the global domain shift, and solves the cross-domain recognition by finding the compact joint subspaces of source and target domain. Specifically, given labeled samples in source domain, we construct subspaces for each of the classes. Then we construct subspaces in the target domain, called anchor subspaces, by collecting unlabeled samples that are close to each other and highly likely all fall into the same class. The corresponding class label is then assigned by minimizing a cost function which reflects the overlap and topological structure consistency between subspaces across source and target domains, and within anchor subspaces, respectively.We further combine the anchor subspaces to corresponding source subspaces to construct the compact joint subspaces. Subsequently, one-vs-rest SVM classifiers are trained in the compact joint subspaces and applied to unlabeled data in the target domain. We evaluate the proposed method on two widely used datasets: object recognition dataset for computer vision tasks, and sentiment classification dataset for natural language processing tasks. Comparison results demonstrate that the proposed method outperforms the comparison methods on both datasets.
Rateless codes have been shown to be able to provide greater flexibility and efficiency than fixed-rate codes for multicast applications. In the following, we optimize rateless codes for unequal error protection (UEP) for multimedia multicasting to a set of heterogeneous users. The proposed designs have the objectives of providing either guaranteed or best-effort quality of service (QoS). A randomly interleaved rateless encoder is proposed whereby users only need to decode symbols up to their own QoS level. The proposed coder is optimized based on measured transmission properties of standardized raptor codes over wireless channels. It is shown that a guaranteed QoS problem formulation can be transformed into a convex optimization problem, yielding a globally optimal solution. Numerical results demonstrate that the proposed optimized random interleaved UEP rateless coder's performance compares favorably with that of other recently proposed UEP rateless codes.
We experimentally demonstrate energy exchange between a delay-tuned femtosecond beam and two delay-fixed ones as they spatiotemporally overlapped and experienced filamentation in air. The energy exchange process in the relative time delay is dramatically elongated up to 40 ps in the presence of plasma grating, indicating that filamentary beams coupling may be an effective method for filament control.
Analytic functions represent the state-of-the-art way of performing complex data analysis within a single SQL statement. In particular, an important class of analytic functions that has been frequently used in commercial systems to support OLAP and decision support applications is the class of window functions. A window function returns for each input tuple a value derived from applying a function over a window of neighboring tuples. However, existing window function evaluation approaches are based on a naive sorting scheme. In this paper, we study the problem of optimizing the evaluation of window functions. We propose several efficient techniques, and identify optimization opportunities that allow us to optimize the evaluation of a set of window functions. We have integrated our scheme into PostgreSQL. Our comprehensive experimental study on the TPC-DS datasets as well as synthetic datasets and queries demonstrate significant speedup over existing approaches.
The role of spontaneous and piezoelectric polarization in III-V nitride heterostructure devices is discussed. Problems as well as opportunities in incorporating polarization in abrupt and graded heterojunctions composed of binary, ternary, and quaternary nitrides are outlined.
An atlas of the Galactic plane (-4.7 deg < b < 4.7 deg) plus the molecular clouds in Orion, Rho Oph, and Taurus-Auriga has been produced at 60 and 100 micron from IRAS data. The Atlas consists of resolution-enhanced coadded images having 1 arcmin -- 2 arcmin resolution as well as coadded images at the native IRAS resolution. The IRAS Galaxy Atlas, together with the DRAO HI line / 21 cm continuum and FCRAO CO (1-0) line Galactic plane surveys, both with similar (approx. 1 arcmin) resolution, provide a powerful venue for studying the interstellar medium, star formation and large scale structure in our Galaxy. This paper documents the production and characteristics of the Atlas.