arXiv subjects

Mi Zhang

Publications and source records attributed to Mi Zhang.

119 records · Page 7Linked to original sources

arXiv2018

NestDNN: Resource-Aware Multi-Tenant On-Device Deep Learning for Continuous Mobile Vision

Mobile vision systems such as smartphones, drones, and augmented-reality headsets are revolutionizing our lives. These systems usually run multiple applications concurrently and their available resources at runtime are dynamic due to events such as starting new applications, closing existing applications, and application priority changes. In this paper, we present NestDNN, a framework that takes the dynamics of runtime resources into account to enable resource-aware multi-tenant on-device deep learning for mobile vision systems. NestDNN enables each deep learning model to offer flexible resource-accuracy trade-offs. At runtime, it dynamically selects the optimal resource-accuracy trade-off for each deep learning model to fit the model's resource demand to the system's available runtime resources. In doing so, NestDNN efficiently utilizes the limited resources in mobile vision systems to jointly maximize the performance of all the concurrently running applications. Our experiments show that compared to the resource-agnostic status quo approach, NestDNN achieves as much as 4.2% increase in inference accuracy, 2.0x increase in video frame processing rate and 1.7x reduction on energy consumption.

cs.CV↗

arXiv2018

Quantized Nonlinear Gaussian-Beam Dynamics $-$ Tailoring Multimode Squeezed-Light Generation

We present a general, second quantization procedure for multi-transverse-spatial mode Gaussian beam dynamics in nonlinear interactions. Previous treatments have focused on the spectral density and angular distribution of spatial modes. Here we go a layer deeper by investigating the complex transverse-spatial mode in each angular-spatial mode. Furthermore, to implement the theory, we simulate four-wave mixing and parametric down-conversion schemes, showing how one can elucidate and tailor the underlying multi-transverse-spatial mode structure along with it's quantum properties.

quant-ph↗

arXiv2018

DeepASL: Enabling Ubiquitous and Non-Intrusive Word and Sentence-Level Sign Language Translation

There is an undeniable communication barrier between deaf people and people with normal hearing ability. Although innovations in sign language translation technology aim to tear down this communication barrier, the majority of existing sign language translation systems are either intrusive or constrained by resolution or ambient lighting conditions. Moreover, these existing systems can only perform single-sign ASL translation rather than sentence-level translation, making them much less useful in daily-life communication scenarios. In this work, we fill this critical gap by presenting DeepASL, a transformative deep learning-based sign language translation technology that enables ubiquitous and non-intrusive American Sign Language (ASL) translation at both word and sentence levels. DeepASL uses infrared light as its sensing mechanism to non-intrusively capture the ASL signs. It incorporates a novel hierarchical bidirectional deep recurrent neural network (HB-RNN) and a probabilistic framework based on Connectionist Temporal Classification (CTC) for word-level and sentence-level ASL translation respectively. To evaluate its performance, we have collected 7,306 samples from 11 participants, covering 56 commonly used ASL words and 100 ASL sentences. DeepASL achieves an average 94.5% word-level translation accuracy and an average 8.2% word error rate on translating unseen ASL sentences. Given its promising performance, we believe DeepASL represents a significant step towards breaking the communication barrier between deaf people and hearing majority, and thus has the significant potential to fundamentally change deaf people's lives.

cs.CV↗

arXiv2018

An analytically simple and computationally efficient Gaussian beam mode-decomposition approach to classical diffraction theory

We present a method of Gaussian-beam-mode decomposition to calculate classical diffraction of optical beams by apertures. This method offers a entirely different approach to examine the classic problem. Although our method is based on a very straightforward setup, it is surprisingly effective. We validate our method by comparing its results with those of Kirchhoff's full diffraction formula. Not only does our method have a simple and organized analytical framework, it also offers significant computational advantage.

physics.optics↗

arXiv2018

Theoretical Analysis of Image-to-Image Translation with Adversarial Learning

Recently, a unified model for image-to-image translation tasks within adversarial learning framework has aroused widespread research interests in computer vision practitioners. Their reported empirical success however lacks solid theoretical interpretations for its inherent mechanism. In this paper, we reformulate their model from a brand-new geometrical perspective and have eventually reached a full interpretation on some interesting but unclear empirical phenomenons from their experiments. Furthermore, by extending the definition of generalization for generative adversarial nets to a broader sense, we have derived a condition to control the generalization capability of their model. According to our derived condition, several practical suggestions have also been proposed on model design and dataset construction as a guidance for further empirical researches.

stat.ML↗

arXiv2017

Optimal Repair Layering for Erasure-Coded Data Centers: From Theory to Practice

Repair performance in hierarchical data centers is often bottlenecked by cross-rack network transfer. Recent theoretical results show that the cross-rack repair traffic can be minimized through repair layering, whose idea is to partition a repair operation into inner-rack and cross-rack layers. However, how repair layering should be implemented and deployed in practice remains an open issue. In this paper, we address this issue by proposing a practical repair layering framework called DoubleR. We design two families of practical double regenerating codes (DRC), which not only minimize the cross-rack repair traffic, but also have several practical properties that improve state-of-the-art regenerating codes. We implement and deploy DoubleR atop Hadoop Distributed File System (HDFS), and show that DoubleR maintains the theoretical guarantees of DRC and improves the repair performance of regenerating codes in both node recovery and degraded read operations.

cs.DC↗

arXiv2017

Multi-pass configuration for Improved Squeezed Vacuum Generation in Hot Rb Vapor

We study a squeezed vacuum field generated in hot Rb vapor via the polarization self-rotation effect. Our previous experiments showed that the amount of observed squeezing may be limited by the contamination of the squeezed vacuum output with higher-order spatial modes, also generated inside the cell. Here, we demonstrate that the squeezing can be improved by making the light interact several times with a less dense atomic ensemble. With optimization of some parameters we can achieve up to -2.6 dB of squeezing in the multi-pass case, which is 0.6 dB improvement compared to the single-pass experimental configuration. Our results show that other than the optical depth of the medium, the spatial mode structure and cell configuration also affect the squeezing level.

physics.atom-ph↗

arXiv2017

Gaussian Beam-Propagation Theory for Nonlinear Optics - Featuring an Exact Treatment of Orbital Angular Momentum Transfer

We present a general, Gaussian spatial mode propagation formalism for describing the generation of higher order multi-spatial-mode beams generated during nonlinear interactions. Furthermore, to implement the theory, we simulate optical angular momentum transfer interactions, and show how one can optimize the interaction to reduce the undesired modes. Past theoretical treatments of this problem have often been phenomenological, at best. Here we present an exact solution for the single-pass no-cavity regime, in which the the nonlinear interaction is not overly strong. We apply our theory to two experiments, with very good agreement, and give examples of several more configurations, easily tested in the laboratory.

physics.optics↗

arXiv2017

Why a hole is like a beam splitter--a general diffraction theory for multimode quantum states of light

Within the second-quantization framework, we develop a formalism for describing a spatially multimode optical field diffracted through a spatial mask and show that this process can be described as an effective interaction between various spatial modes. We demonstrate a method to calculate the quantum state in the diffracted optical field for any given quantum state in the incident field. Using numerical simulations, we also show that with single-mode squeezed-vacuum state input, the prediction of our theory is in qualitative agreement with our experimental data. We also give several additional examples of how the theory works, for various quantum input states, which may be easily tested in the lab; including two single-mode squeezed vacuums, single- and two-photon inputs, where we show the diffraction process produces two-mode squeezed vacuum, number-path entanglement and a Hong-Ou-Mandel-like effect--analogous to a beam splitter.

quant-ph↗

arXiv2016

Spatial Multi-Mode Structure of Atom-Generated Squeezed Light

We investigated the spatial distribution of quantum fluctuations in a squeezed vacuum field, generated via polarization self-rotation (PSR) interaction of an ensemble of Rb atoms and a strong near-resonant linearly polarized laser field. We found that the noise suppression is greatly effected by the transverse profile of a spatial mask, placed in both the squeezed field and the local oscillator, as well as its position along the focused beam near the focal point. These observations indicate the spatial multi-mode structure of the squeezed vacuum field. We have developed a theoretical model that describes the generation of higher-order Laguerre-Gauss modes as a result of PSR light-atom interaction. The prediction of this model are in a good qualitative agreement with the experimental measurements.

physics.optics↗

arXiv2013

Generating squeezed vacuum field with non-zero orbital angular momentum with atomic ensembles

We demonstrated that by using a pump field with non-zero orbital angular momentum (OAM) in the polarization self-rotation squeezing process it is possible to generate a squeezed vacuum optical field with the matching OAM. We found a similar level of maximum quantum noise reduction for a first-order Laguerre-Gaussian pump beam and a regular Gaussian pump beam, even though the optimal operational conditions differed in these two cases. Also, we investigated the effect of self-defocusing on the level of the vacuum squeezing by simultaneously monitoring the minimum quantum noise level and the output beam transverse profile at various pump laser powers and atomic densities, and found no direct correlations between the increased beam size and the degree of measured squeezing.

quant-ph↗