arXiv subjects

Zhuo Chen

Publications and source records attributed to Zhuo Chen.

At least 433 records · Page 24Linked to original sources

arXiv2018

Cracking the cocktail party problem by multi-beam deep attractor network

While recent progresses in neural network approaches to single-channel speech separation, or more generally the cocktail party problem, achieved significant improvement, their performance for complex mixtures is still not satisfactory. In this work, we propose a novel multi-channel framework for multi-talker separation. In the proposed model, an input multi-channel mixture signal is firstly converted to a set of beamformed signals using fixed beam patterns. For this beamforming, we propose to use differential beamformers as they are more suitable for speech separation. Then each beamformed signal is fed into a single-channel anchored deep attractor network to generate separated signals. And the final separation is acquired by post selecting the separating output for each beams. To evaluate the proposed system, we create a challenging dataset comprising mixtures of 2, 3 or 4 speakers. Our results show that the proposed system largely improves the state of the art in speech separation, achieving 11.5 dB, 11.76 dB and 11.02 dB average signal-to-distortion ratio improvement for 4, 3 and 2 overlapped speaker mixtures, which is comparable to the performance of a minimum variance distortionless response beamformer that uses oracle location, source, and noise information. We also run speech recognition with a clean trained acoustic model on the separated speech, achieving relative word error rate (WER) reduction of 45.76\%, 59.40\% and 62.80\% on fully overlapped speech of 4, 3 and 2 speakers, respectively. With a far talk acoustic model, the WER is further reduced.

cs.SD↗

arXiv2018

Pedestrian-Robot Interaction Experiments in an Exit Corridor

The study of human-robot interaction (HRI) has received increasing research attention for robot navigation in pedestrian crowds. In this paper, we present empirical study of pedestrian-robot interaction in an uni-directional exit corridor. We deploy a mobile robot moving in a direction perpendicular to that of the pedestrian flow, and install a pedestrian motion tracking system to record the collective motion. We analyze both individual and collective motion of pedestrians, and measure the effect of the robot motion on the overall pedestrian flow. The experimental results show the effect of passive HRI, where the pedestrians' overall speed is slowed down in the presence of the robot, and the faster the robot moves, the lower the average pedestrian velocity becomes. Experiment results show qualitative consistency of the collective HRI effect with simulation results that was previously reported. The study can be used to guide future design of robot-assisted pedestrian evacuation algorithms.

cs.RO↗

arXiv2018

Atiyah and Todd classes arising from integrable distributions

In this paper, we study the Atiyah class and Todd class of the DG manifold $(F[1],d_F)$ corresponding to an integrable distribution $F \subset T_{\mathbb{K}} M = TM \otimes_{\mathbb{R}} \mathbb{K}$, where $\mathbb{K} = \mathbb{R}$ or $\mathbb{C}$. We show that these two classes are canonically identical to those of the Lie pair $(T_{\mathbb{K}} M, F)$. As a consequence, the Atiyah class of a complex manifold $X$ is isomorphic to the Atiyah class of the corresponding DG manifold $(T^{0,1}_X[1],\bar{\partial})$. Moreover, if $X$ is a compact Kähler manifold, then the Todd class of $X$ is also isomorphic to the Todd class of the corresponding DG manifold $(T^{0,1}_X[1],\bar{\partial})$.

math.DG↗

arXiv2017

Task Scheduling for Heterogeneous Multicore Systems

In recent years, as the demand for low energy and high performance computing has steadily increased, heterogeneous computing has emerged as an important and promising solution. Because most workloads can typically run most efficiently on certain types of cores, mapping tasks on the best available resources can not only save energy but also deliver high performance. However, optimal task scheduling for performance and/or energy is yet to be solved for heterogeneous platforms. The work presented herein mathematically formulates the optimal heterogeneous system task scheduling as an optimization problem using queueing theory. We analytically solve for the common case of two processor types, e.g., CPU+GPU, and give an optimal policy (CAB). We design the GrIn heuristic to efficiently solve for near-optimal policy for any number of processor types (within 1.6% of the optimal). Both policies work for any task size distribution and processing order, and are therefore, general and practical. We extensively simulate and validate the theory, and implement the proposed policy in a CPU-GPU real platform to show the optimal throughput and energy improvement. Comparing to classic policies like load-balancing, our results range from 1.08x~2.24x better performance or 1.08x~2.26x better energy efficiency in simulations, and 2.37x~9.07x better performance in experiments.

cs.PF↗

arXiv2017

Priority-Aware Near-Optimal Scheduling for Heterogeneous Multi-Core Systems with Specialized Accelerators

To deliver high performance in power limited systems, architects have turned to using heterogeneous systems, either CPU+GPU or mixed CPU-hardware systems. However, in systems with different processor types and task affinities, scheduling tasks becomes more challenging than in homogeneous multi-core systems or systems without task affinities. The problem is even more complex when specialized accelerators and task priorities are included. In this paper, we provide a formal proof for the optimal scheduling policy for heterogeneous systems with arbitrary number of resource types, including specialized accelerators, independent of the task arrival rate, task size distribution, and resource processing order. We transform the optimal scheduling policy to a nonlinear integer optimization problem and propose a fast, near-optimal algorithm. An additional heuristic is proposed for the case of priority-aware scheduling. Our experimental results demonstrate that the proposed algorithm is only 0.3% from the optimal and superior to conventional scheduling policies.

cs.PF↗

arXiv2017

Optimal double resonant condition in metallic core-shell nanocavity for third harmonic generation

As the rapid development of nonlinear optics, the enhancement of optical third harmonic generation becomes a pop research realm in the fields of physics, chemistry, biology, materials science, information science and other fields. In this letter, theoretical analysis on double resonance situation in metallic core-shell nanostructure is performed in order to optimize the efficiency of third harmonic generation. As plenty of cavity modes ranging from visible area to near infrared can be excited efficiently in core-shell nanospheres, kinds of double resonant conditions can be formed by matching two different multipolar cavity modes. Numerical simulations show that the third harmonic generation (THG) intensity in the far field can be enhanced remarkably when the THG signal couples well with the high order cavity mode. More importantly, THG efficiency is optimum from the cavity on double resonant conditions coupling two modes with the same order. The THG intensity differs up to 3 magnitudes on this condition. Subsequent theoretical analysis indicates that changing third-order nonlinear susceptibility of the metal shell while keeping that of the core fixed has almost no effect on THG efficiency. This finding about optimal double resonance has an appreciable effect on optimizing THG in spherical cavity structures.

physics.optics↗

arXiv2017

Tunable hyperbolic metamaterial cavity towards few exciton strong coupling

We theoretically investigate coupling interaction between molecule excitons and whispering gallery modes (WGMs) that approaches the limit of single exciton strong coupling in hyperbolic metamaterial (HMM) cavity, composed of a dielectric core wrapped by several stacks of alternative layers of metal and dielectric. We demonstrate that associated with the excitation of the whispering gallery mode, the electric fields with resonance wavelengths that are much larger than the cavity size can be highly localized within a dielectric layer, leading to extremely small mode volumes. By using temporal coupled mode theory to model the interaction in the proposed WGMs-exciton system, we are able to demonstrate that the coupling between molecular excitons and hyperbolic cavity WGMs can reach the strong coupling regime. Furthermore, we also prove that changing both the thickness of inner core and outermost layer can lead to invariant resonance frequency while the threshold number of coupled excitons to fulfill the criteria for strong coupling remains ~ 4 in various sets of tunable HMM cavities.

physics.optics↗

arXiv2017

Wind-accelerated orbital evolution in binary systems with giant stars

Using 3D radiation-hydrodynamic simulations and analytic theory, we study the orbital evolution of asymptotic-giant-branch (AGB) binary systems for various initial orbital separations and mass ratios, and thus different initial accretion modes. The time evolution of binary separations and orbital periods are calculated directly from the averaged mass loss rate, accretion rate and angular momentum loss rate. We separately consider spin-orbit synchronized and zero spin AGB cases. We find that the the angular momentum carried away by the mass loss together with the mass transfer can effectively shrink the orbit when accretion occurs via wind-Roche-lobe overflow. In contrast, the larger fraction of mass lost in Bondi-Hoyle-Lyttleton accreting systems acts to enlarge the orbit. Synchronized binaries tend to experience stronger orbital period decay in close binaries. We also find that orbital period decay is faster when we account for the nonlinear evolution of the accretion mode as the binary starts to tighten. This can increase the fraction of binaries that result in common envelope, luminous red novae, Type Ia supernovae and planetary nebulae with tight central binaries. The results also imply that planets in the the habitable zone around white dwarfs are unlikely to be found.

astro-ph.SR↗

arXiv2017

Image Quality Assessment Guided Deep Neural Networks Training

For many computer vision problems, the deep neural networks are trained and validated based on the assumption that the input images are pristine (i.e., artifact-free). However, digital images are subject to a wide range of distortions in real application scenarios, while the practical issues regarding image quality in high level visual information understanding have been largely ignored. In this paper, in view of the fact that most widely deployed deep learning models are susceptible to various image distortions, the distorted images are involved for data augmentation in the deep neural network training process to learn a reliable model for practical applications. In particular, an image quality assessment based label smoothing method, which aims at regularizing the label distribution of training images, is further proposed to tune the objective functions in learning the neural network. Experimental results show that the proposed method is effective in dealing with both low and high quality images in the typical image classification task.

cs.CV↗

arXiv2017

Improving Adherence to Heart Failure Management Guidelines via Abductive Reasoning

Management of chronic diseases such as heart failure (HF) is a major public health problem. A standard approach to managing chronic diseases by medical community is to have a committee of experts develop guidelines that all physicians should follow. Due to their complexity, these guidelines are difficult to implement and are adopted slowly by the medical community at large. We have developed a physician advisory system that codes the entire set of clinical practice guidelines for managing HF using answer set programming(ASP). In this paper we show how abductive reasoning can be deployed to find missing symptoms and conditions that the patient must exhibit in order for a treatment prescribed by a physician to work effectively. Thus, if a physician does not make an appropriate recommendation or makes a non-adherent recommendation, our system will advise the physician about symptoms and conditions that must be in effect for that recommendation to apply. It is under consideration for acceptance in TPLP.

cs.AI↗

arXiv2017

Deep Clustering and Conventional Networks for Music Separation: Stronger Together

Deep clustering is the first method to handle general audio separation scenarios with multiple sources of the same type and an arbitrary number of sources, performing impressively in speaker-independent speech separation tasks. However, little is known about its effectiveness in other challenging situations such as music source separation. Contrary to conventional networks that directly estimate the source signals, deep clustering generates an embedding for each time-frequency bin, and separates sources by clustering the bins in the embedding space. We show that deep clustering outperforms conventional networks on a singing voice separation task, in both matched and mismatched conditions, even though conventional networks have the advantage of end-to-end training for best signal approximation, presumably because its more flexible objective engenders better regularization. Since the strengths of deep clustering and conventional network architectures appear complementary, we explore combining them in a single hybrid network trained via an approach akin to multi-task learning. Remarkably, the combination significantly outperforms either of its components.

stat.ML↗

arXiv2017

Mass Transfer and Disc Formation in AGB Binary Systems

We investigate mass transfer and the formation of disc in binary systems using a combination of numerical simulations and theory. We consider six models distinguished by binary separation, secondary mass and outflow mechanisms. Each system consists of an asymptotic-giant-branch (AGB) star and an accreting secondary. The AGB star loses its mass via a wind. In one of our six models, the AGB star incurs a short period of outburst. In all cases, the secondary accretes part of the ejected mass and also influences the mass-loss rate of the AGB star. The ejected mass may remain gravitationally bound to the binary system and form a circumbinary disk, or contribute to an accretion disk around the secondary. In other cases, the ejecta will escape the binary system. The accretion rate on to the secondary changes non-linearly with binary separation. In our closest binary simulations, our models exemplify the wind Roche lobe overflow while in our wide binary cases, the mass transfer exhibits Bondi-Hoyle accretion. The morphologies of the outflows in the binary systems are varied. The variety may provide clues to how the late AGB phase influences planetary nebulae shaping. We employ the adaptive-mesh-refinement code ASTROBEAR for our simulations and include ray-tracing, radiation transfer, cooling and dust formation. To attain the highest computational efficiency and the most stable results, all simulations are run in the corotating frame.

astro-ph.SR↗

arXiv2017

Deep attractor network for single-microphone speaker separation

Despite the overwhelming success of deep learning in various speech processing tasks, the problem of separating simultaneous speakers in a mixture remains challenging. Two major difficulties in such systems are the arbitrary source permutation and unknown number of sources in the mixture. We propose a novel deep learning framework for single channel speech separation by creating attractor points in high dimensional embedding space of the acoustic signals which pull together the time-frequency bins corresponding to each source. Attractor points in this study are created by finding the centroids of the sources in the embedding space, which are subsequently used to determine the similarity of each bin in the mixture to each source. The network is then trained to minimize the reconstruction error of each source by optimizing the embeddings. The proposed model is different from prior works in that it implements an end-to-end training, and it does not depend on the number of sources in the mixture. Two strategies are explored in the test time, K-means and fixed attractor points, where the latter requires no post-processing and can be implemented in real-time. We evaluated our system on Wall Street Journal dataset and show 5.49\% improvement over the previous state-of-the-art methods.

cs.SD↗

arXiv2017

End-to-End Attention based Text-Dependent Speaker Verification

A new type of End-to-End system for text-dependent speaker verification is presented in this paper. Previously, using the phonetically discriminative/speaker discriminative DNNs as feature extractors for speaker verification has shown promising results. The extracted frame-level (DNN bottleneck, posterior or d-vector) features are equally weighted and aggregated to compute an utterance-level speaker representation (d-vector or i-vector). In this work we use speaker discriminative CNNs to extract the noise-robust frame-level features. These features are smartly combined to form an utterance-level speaker vector through an attention mechanism. The proposed attention model takes the speaker discriminative information and the phonetic information to learn the weights. The whole system, including the CNN and attention model, is joint optimized using an end-to-end criterion. The training algorithm imitates exactly the evaluation process --- directly mapping a test utterance and a few target speaker utterances into a single verification score. The algorithm can automatically select the most similar impostor for each target speaker to train the network. We demonstrated the effectiveness of the proposed end-to-end system on Windows $10$ "Hey Cortana" speaker verification task.

cs.CL↗

arXiv2016

A Physician Advisory System for Chronic Heart Failure Management Based on Knowledge Patterns

Management of chronic diseases such as heart failure, diabetes, and chronic obstructive pulmonary disease (COPD) is a major problem in health care. A standard approach that the medical community has devised to manage widely prevalent chronic diseases such as chronic heart failure (CHF) is to have a committee of experts develop guidelines that all physicians should follow. These guidelines typically consist of a series of complex rules that make recommendations based on a patient's information. Due to their complexity, often the guidelines are either ignored or not complied with at all, which can result in poor medical practices. It is not even clear whether it is humanly possible to follow these guidelines due to their length and complexity. In the case of CHF management, the guidelines run nearly 80 pages. In this paper we describe a physician-advisory system for CHF management that codes the entire set of clinical practice guidelines for CHF using answer set programming. Our approach is based on developing reasoning templates (that we call knowledge patterns) and using these patterns to systemically code the clinical guidelines for CHF as ASP rules. Use of the knowledge patterns greatly facilitates the development of our system. Given a patient's medical information, our system generates a recommendation for treatment just as a human physician would, using the guidelines. Our system will work even in the presence of incomplete information. Our work makes two contributions: (i) it shows that highly complex guidelines can be successfully coded as ASP rules, and (ii) it develops a series of knowledge patterns that facilitate the coding of knowledge expressed in a natural language and that can be used for other application domains. This paper is under consideration for acceptance in TPLP.

cs.AI↗

arXiv2016

Three-dimensional hydrodynamic simulations of L2 Puppis

Recent observations of the L2 Puppis system suggest that this Mira-like variable may be in the early stages of forming a bipolar planetary nebula (PN). As one of nearest and brightest AGB stars, thought be a binary, L2 Puppis serves as a benchmark object for studying the late-stages of stellar evolution. We perform global, three-dimensional, adaptive-mesh-refinement hydrodynamic simulations of the L2 Puppis system with AstroBEAR. We use the radiative transfer code RADMC-3D to construct the broad-band spectral-energy-distribution (SED) and synthetic observational images from our simulations. Given the reported binary parameters, we are able to reproduce the current observational data if a short pulse of dense material is released from the AGB star with a velocity sufficient to escape the primary but not the binary. Such a situation could result from a thermal pulse, be induced by a periastron passage of the secondary, or could be launched if the primary ingests a planet.

astro-ph.SR↗

arXiv2016

Toda-like (0,2) mirrors to products of projective spaces

One of the open problems in understanding (0,2) mirror symmetry concerns the construction of Toda-like Landau-Ginzburg mirrors to (0,2) theories on Fano spaces. In this paper, we begin to fill this gap by making an ansatz for (0,2) Toda-like theories mirror to (0,2) supersymmetric nonlinear sigma models on products of projective spaces, with deformations of the tangent bundle, generalizing a special case previously worked out for P1xP1. We check this ansatz by matching correlation functions of the B/2-twisted Toda-like theories to correlation functions of corresponding A/2-twisted nonlinear sigma models, computed primarily using localization techniques. These (0,2) Landau-Ginzburg models admit redundancies, which can lend themselves to multiple distinct-looking representatives of the same physics, which we discuss.

hep-th↗

arXiv2016

Single-Channel Multi-Speaker Separation using Deep Clustering

Deep clustering is a recently introduced deep learning architecture that uses discriminatively trained embeddings as the basis for clustering. It was recently applied to spectrogram segmentation, resulting in impressive results on speaker-independent multi-speaker separation. In this paper we extend the baseline system with an end-to-end signal approximation objective that greatly improves performance on a challenging speech separation. We first significantly improve upon the baseline system performance by incorporating better regularization, larger temporal context, and a deeper architecture, culminating in an overall improvement in signal to distortion ratio (SDR) of 10.3 dB compared to the baseline of 6.0 dB for two-speaker separation, as well as a 7.1 dB SDR improvement for three-speaker separation. We then extend the model to incorporate an enhancement layer to refine the signal estimates, and perform end-to-end training through both the clustering and enhancement stages to maximize signal fidelity. We evaluate the results using automatic speech recognition. The new signal approximation objective, combined with end-to-end training, produces unprecedented performance, reducing the word error rate (WER) from 89.1% down to 30.8%. This represents a major advancement towards solving the cocktail party problem.

cs.LG↗