arXiv subjects

Ang Li

Publications and source records attributed to Ang Li.

At least 595 records · Page 33Linked to original sources

arXiv2018

SymmNet: A Symmetric Convolutional Neural Network for Occlusion Detection

Detecting the occlusion from stereo images or video frames is important to many computer vision applications. Previous efforts focus on bundling it with the computation of disparity or optical flow, leading to a chicken-and-egg problem. In this paper, we leverage convolutional neural network to liberate the occlusion detection task from the interleaved, traditional calculation framework. We propose a Symmetric Network (SymmNet) to directly exploit information from an image pair, without estimating disparity or motion in advance. The proposed network is structurally left-right symmetric to learn the binocular occlusion simultaneously, aimed at jointly improving both results. The comprehensive experiments show that our model achieves state-of-the-art results on detecting the stereo and motion occlusion.

cs.CV↗

arXiv2018

Josephson effect in a few-hole quantum dot

We use a Ge-Si core-shell nanowire to realise a Josephson field-effect transistor with highly transparent contacts to superconducting leads. By changing the electric field we gain access to two distinct regimes not combined before in a single device: In the accumulation mode the device is highly transparent and the supercurrent is carried by multiple subbands, while near depletion supercurrent is carried by single-particle levels of a strongly coupled quantum dot operating in the few-hole regime. These results establish Ge-Si nanowires as an important platform for hybrid superconductor-semiconductor physics and Majorana fermions.

cond-mat.mes-hall↗

arXiv2018

C-WSL: Count-guided Weakly Supervised Localization

We introduce count-guided weakly supervised localization (C-WSL), an approach that uses per-class object count as a new form of supervision to improve weakly supervised localization (WSL). C-WSL uses a simple count-based region selection algorithm to select high-quality regions, each of which covers a single object instance during training, and improves existing WSL methods by training with the selected regions. To demonstrate the effectiveness of C-WSL, we integrate it into two WSL architectures and conduct extensive experiments on VOC2007 and VOC2012. Experimental results show that C-WSL leads to large improvements in WSL and that the proposed approach significantly outperforms the state-of-the-art methods. The results of annotation experiments on VOC2007 suggest that a modest extra time is needed to obtain per-class object counts compared to labeling only object categories in an image. Furthermore, we reduce the annotation time by more than $2\times$ and $38\times$ compared to center-click and bounding-box annotations.

cs.CV↗

arXiv2018

Neutron star equation of state from the quark level in the light of GW170817

Matter state inside neutron stars is an exciting problem in astrophysics, nuclear physics and particle physics. The equation of state (EOS) of neutron stars plays a crucial role in the present multimessenger astronomy, especially after the event of GW170817. We propose a new neutron star EOS "QMF18" from the quark level, which describes well robust observational constraints from free-space nucleon, nuclear matter saturation, heavy pulsar measurements and the tidal deformability of the very recent GW170817 observation. For this purpose, we employ the quark-mean-field (QMF) model, allowing one to tune the density dependence of the symmetry energy and study effectively its correlations with the Love number and the tidal deformability. We provide tabulated data for the new EOS and compare it with other recent EOSs from various many-body frameworks.

nucl-th↗

arXiv2018

Outflows from black hole hyperaccretion systems: short and long-short gamma-ray bursts and "quasi-supernovae"

The detections of some long gamma-ray bursts (LGRBs) relevant to mergers of neutron star (NS)-NS or black hole (BH)-NS, as well as some short gamma-ray bursts (SGRBs) probably produced by collapsars, muddle the boundary of two categories of gamma-ray bursts (GRBs). In both cases, a plausible candidate of central engine is a BH surrounded by a hyperaccretion disc with strong outflows, launching relativistic jets driven by Blandford-Znajek mechanism. In the framework of compact binary mergers, we test the applicability of the BH hyperaccretion inflow-outflow model on powering observed GRBs. We find that, for a low outflow ratio, $\sim 50\%$, postmerger hyperaccretion processes could power not only all SGRBs but also most of LGRBs. Some LGRBs might do originate from merger events in the BH hyperaccretion scenario, at least on the energy requirement. Moreover, kilonovae might be produced by neutron-rich outflows, and their luminosities and timescales significantly depend on the outflow strengths. GRBs and their associated kilonovae are competitive with each other on the disc mass and total energy budgets. The stronger the outflow, the more similar the characteristics of kilonovae to supernovae (SNe). This kind of `nova' might be called `quasi-SN'.

astro-ph.HE↗

arXiv2018

Dynamic Zoom-in Network for Fast Object Detection in Large Images

We introduce a generic framework that reduces the computational cost of object detection while retaining accuracy for scenarios where objects with varied sizes appear in high resolution images. Detection progresses in a coarse-to-fine manner, first on a down-sampled version of the image and then on a sequence of higher resolution regions identified as likely to improve the detection accuracy. Built upon reinforcement learning, our approach consists of a model (R-net) that uses coarse detection results to predict the potential accuracy gain for analyzing a region at a higher resolution and another model (Q-net) that sequentially selects regions to zoom in. Experiments on the Caltech Pedestrians dataset show that our approach reduces the number of processed pixels by over 50% without a drop in detection accuracy. The merits of our approach become more significant on a high resolution test set collected from YFCC100M dataset, where our approach maintains high detection performance while reducing the number of processed pixels by about 70% and the detection time by over 50%.

cs.CV↗

arXiv2018

Constraints on interquark interaction parameters with GW170817 in a binary strange star scenario

The LIGO/VIRGO detection of the gravitational waves from a binary merger system, GW170817, has put a clean and strong constraint on the tidal deformability of the merging objects. From this constraint, deep insights can be obtained in compact star equation of states, which has been one of the most puzzling problems for nuclear physicists and astrophysicists. Employing one of the most widely-used quark star EOS model, we characterize the star properties by the strange quark mass ($m_s$), an effective bag constant ($B_{\rm eff}$), the perturbative QCD correction ($a_4$), as well as the gap parameter ($Δ$) when considering quark pairing, and investigate the dependences of the tidal deformablity on them. We find that the tidal deformability is dominated by $B_{\rm eff}$, and insensitive to $m_s$, $a_4$. We discuss the correlation between the tidal deformability and the maximum mass ($M_\mathrm{TOV}$) of a static quark star, which allows the model possibility to rule out the existence of quark stars with future gravitational wave observations and mass measurements. The current tidal deformability measurement implies $M_\mathrm{TOV} \le2.18\,M_\odot$ ($2.32\,M_\odot$ when pairing is considered) for quark stars. Combining with two-solar-mass pulsar observations, we also make constraints on the poorly known gap parameter $Δ$ for color-flavor-locked quark matter.

astro-ph.HE↗

arXiv2018

NISP: Pruning Networks using Neuron Importance Score Propagation

To reduce the significant redundancy in deep Convolutional Neural Networks (CNNs), most existing methods prune neurons by only considering statistics of an individual layer or two consecutive layers (e.g., prune one layer to minimize the reconstruction error of the next layer), ignoring the effect of error propagation in deep networks. In contrast, we argue that it is essential to prune neurons in the entire neuron network jointly based on a unified goal: minimizing the reconstruction error of important responses in the "final response layer" (FRL), which is the second-to-last layer before classification, for a pruned network to retrain its predictive power. Specifically, we apply feature ranking techniques to measure the importance of each neuron in the FRL, and formulate network pruning as a binary integer optimization problem and derive a closed-form solution to it for pruning neurons in earlier layers. Based on our theoretical analysis, we propose the Neuron Importance Score Propagation (NISP) algorithm to propagate the importance scores of final responses to every neuron in the network. The CNN is pruned by removing neurons with least importance, and then fine-tuned to retain its predictive power. NISP is evaluated on several datasets with multiple CNN models and demonstrated to achieve significant acceleration and compression with negligible accuracy loss.

cs.CV↗

arXiv2018

Nucleon radius effects on neutron stars in quark mean field model

We study the effects of free space nucleon radius on nuclear matter and neutron stars within the framework of quark mean field model. The nucleon radius is treated self-consistently with this model, where quark confinement is adjusted to fit different values of nucleon radius. Corrections due to center-of-mass motion, quark-pion coupling, and one gluon exchange are included to obtain the nucleon mass in vacuum. The meson coupling constants that describe the behavior of many-body nucleonic system are newly-constructed by reproducing the empirical saturation properties of nuclear matter, including the recent determinations of symmetry energy parameters. Our results show that the nucleon radius in free space have negligible effects on nuclear matter equation of state and neutron star mass-radius relations, which is different from the conclusion drawn in previous studies. We further explore that the sensitivity of star radius on the nucleon radius found in earlier publications is actually from the symmetry energy and its slope.

nucl-th↗

arXiv2018

The allowed parameter space of a long-lived neutron star as the merger remnant of GW170817

Limited by the sensitivities of the current gravitational wave (GW) detectors, the central remnant of the binary neutron star (NS) merger associated with GW170817 remains an open question. Considering the relatively large total mass, it is generally proposed that the merger of GW170817 would lead to a shortly lived hypermassive NS or directly produce a black hole (BH). There is no clear evidence to support or rule out a long-lived NS as the merger remnant. Here we utilize the GW and electromagnetic (EM) signals to comprehensively investigate the parameter space that allows a long-lived NS to survive as the merger remnant of GW170817. We find that for some stiff equations of state, the merger of GW170817 could, in principle, lead to a massive NS, which has a millisecond spin period. The post-merger GW signal could hardly constrain the ellipticity of the NS. If the ellipticity reaches $10^{-3}$, in order to be compatible with the multi-band EM observations, the dipole magnetic field of the NS ($B_p$) is constrained to the magnetar level of $\sim10^{14}$ G. If the ellipticity is smaller than $10^{-4}$, $B_p$ is constrained to the level of $\sim10^{10}-10^{12}\,$G. These conclusions weakly depend on the adoption of equations of state.

astro-ph.HE↗

arXiv2018

SuperNeurons: Dynamic GPU Memory Management for Training Deep Neural Networks

Going deeper and wider in neural architectures improves the accuracy, while the limited GPU DRAM places an undesired restriction on the network design domain. Deep Learning (DL) practitioners either need change to less desired network architectures, or nontrivially dissect a network across multiGPUs. These distract DL practitioners from concentrating on their original machine learning tasks. We present SuperNeurons: a dynamic GPU memory scheduling runtime to enable the network training far beyond the GPU DRAM capacity. SuperNeurons features 3 memory optimizations, \textit{Liveness Analysis}, \textit{Unified Tensor Pool}, and \textit{Cost-Aware Recomputation}, all together they effectively reduce the network-wide peak memory usage down to the maximal memory usage among layers. We also address the performance issues in those memory saving techniques. Given the limited GPU DRAM, SuperNeurons not only provisions the necessary memory for the training, but also dynamically allocates the memory for convolution workspaces to achieve the high performance. Evaluations against Caffe, Torch, MXNet and TensorFlow have demonstrated that SuperNeurons trains at least 3.2432 deeper network than current ones with the leading performance. Particularly, SuperNeurons can train ResNet2500 that has $10^4$ basic network layers on a 12GB K40c.

cs.DC↗

arXiv2017

Interference Exploitation Precoding Made Practical: Closed-Form Solutions with Optimal Performance

In this paper, we propose closed-form precoding schemes with optimal performance for constructive interference (CI) exploitation in the multiuser multiple-input single-output (MU-MISO) downlink. We first consider an optimization where we maximize the distance between the constructive region and the detection thresholds. The cases of both strict and non-strict phase rotation are considered and can further be formulated as convex optimization problems. For optimization with strict phase rotation, we mathematically derive the optimal beamforming structure with Lagrangian and Karush-Kuhn-Tucker (KKT) conditions. By formulating its dual problem, the optimization problem is further shown to be equivalent to a quadratic programming (QP) over a simplex, which can be solved more efficiently. We then extend our analysis to the case of non-strict phase rotation, where it is mathematically shown that a K-dimensional optimization for non-strict phase rotation is equivalent to a 2K-dimensional optimization for strict phase rotation in terms of the problem formulation. The connection with the conventional zero-forcing (ZF) precoding is also discussed. Based on the above analysis, we further propose an iterative closed-form scheme to obtain the optimal beamforming matrix, where within each iteration a closed-form solution can be obtained. Numerical results validate our analysis and the optimality of the proposed iterative scheme, and further show that the proposed closed-form scheme is more efficient than the conventional QP algorithms with interior-point methods, which motivates the use of CI beamforming in practical wireless systems.

cs.IT↗

arXiv2017

Towards Dual-functional Radar-Communication Systems: Optimal Waveform Design

We focus on a dual-functional multi-input-multi-output (MIMO) radar-communication (RadCom) system, where a single transmitter communicates with downlink cellular users and detects radar targets simultaneously. Several design criteria are considered for minimizing the downlink multi-user interference. First, we consider both the omnidirectional and directional beampattern design problems, where the closed-form globally optimal solutions are obtained. Based on these waveforms, we further consider a weighted optimization to enable a flexible trade-off between radar and communications performance and introduce a low-complexity algorithm. The computational costs of the above three designs are shown to be similar to the conventional zero-forcing (ZF) precoding. Moreover, to address the more practical constant modulus waveform design problem, we propose a branch-and-bound algorithm that obtains a globally optimal solution and derive its worst-case complexity as a function of the maximum iteration number. Finally, we assess the effectiveness of the proposed waveform design approaches by numerical results.

cs.IT↗

arXiv2017

Massive MIMO 1-Bit DAC Transmission: A Low-Complexity Symbol Scaling Approach

We study multi-user massive multiple-input single-output (MISO) systems and focus on downlink transmission, where the base station (BS) employs a large antenna array with low-cost 1-bit digital-to-analog converters (DACs). The direct combination of existing beamforming schemes with 1-bit DACs is shown to lead to an error floor at medium-to-high SNR regime, due to the coarse quantization of the DACs with limited precision. In this paper, based on the constructive interference we consider both a quantized linear beamforming scheme where we analytically obtain the optimal beamforming matrix, and a non-linear mapping scheme where we directly design the transmit signal vector. Due to the 1-bit quantization, the formulated optimization for the non-linear mapping scheme is shown to be non-convex. To solve this problem, the non-convex constraints of the 1-bit DACs are firstly relaxed, followed by an element-wise normalization to satisfy the 1-bit DAC transmission. We further propose a low-complexity symbol scaling scheme that consists of three stages, in which the quantized transmit signal on each antenna element is selected sequentially. Numerical results show that the proposed symbol scaling scheme achieves a comparable performance to the optimization-based non-linear mapping approach, while its corresponding complexity is negligible compared to that of the non-linear scheme.

eess.SP↗

arXiv2017

Comparison of gravitational waves from central engines of gamma-ray bursts: neutrino-dominated accretion flows, Blandford-Znajek mechanisms, and millisecond magnetars

Neutrino-dominated accretion flow (NDAF) around a rotating stellar-mass black hole (BH) is one of the plausible candidates for the central engines of gamma-ray bursts (GRBs). Because the time-variant and anisotropic emission of neutrinos from NDAFs leads to GRB variability, NDAFs can be regarded as the sources of the strong gravitational waves (GWs). We calculate the dependences of the GW strains on both the BH spin and the accretion rate. We demonstrate that for typical GRBs with either single pulse or multiple pulses, the GWs from NDAFs might be detected at a distance of $\sim 100$ kpc/$\sim 1$ Mpc by advanced LIGO/Einstein Telescope with a typical frequency of $\sim 10-100$ Hz. Besides NDAFs, the other two competitive candidates for GRB central engine are Blandford-Znajek (BZ) mechanism and millisecond magnetars. We explore the GW signals from these two as well, and compare the corresponding results with NDAFs'. We find that for a certain GRB, the possible detected distance from NDAFs is about two orders of magnitude higher than that from BZ mechanism, but at least two orders of magnitude lower than that from magnetars. The typical GW frequency for BZ mechanism is the same with that of NDAFs, $\sim 10-100$ Hz, while the typical frequency for magnetars is $\sim 2000$ Hz. Therefore, the GWs released by the central engines of adjacent GRBs might be used to determine whether there is an NDAF, a BZ jet or a magnetar in GRB center.

astro-ph.HE↗

arXiv2017

On the Sensitivity of k-Uniform Hypergraph Properties

In this paper we present a graph property with sensitivity $Θ(\sqrt{n})$, where $n={v\choose2}$ is the number of variables, and generalize it to a $k$-uniform hypergraph property with sensitivity $Θ(\sqrt{n})$, where $n={v\choose k}$ is again the number of variables. This yields the smallest sensitivity yet achieved for a $k$-uniform hypergraph property. We then show that, for even $k$, there is a $k$-uniform hypergraph property that demonstrates a quadratic gap between sensitivity and block sensitivity. This matches the largest known gap found by Ambainis and Sun (2011) for Boolean functions in general, and is the first known example of such a gap for a graph or hypergraph property.

cs.CC↗

arXiv2017

Multiple testing with the structure adaptive Benjamini-Hochberg algorithm

In multiple testing problems, where a large number of hypotheses are tested simultaneously, false discovery rate (FDR) control can be achieved with the well-known Benjamini-Hochberg procedure, which adapts to the amount of signal present in the data. Many modifications of this procedure have been proposed to improve power in scenarios where the hypotheses are organized into groups or into a hierarchy, as well as other structured settings. Here we introduce SABHA, the "structure-adaptive Benjamini-Hochberg algorithm", as a generalization of these adaptive testing methods. SABHA incorporates prior information about any pre-determined type of structure in the pattern of locations of the signals and nulls within the list of hypotheses, to reweight the p-values in a data-adaptive way. This raises the power by making more discoveries in regions where signals appear to be more common. Our main theoretical result proves that SABHA controls FDR at a level that is at most slightly higher than the target FDR level, as long as the adaptive weights are constrained sufficiently so as not to overfit too much to the data-interestingly, the excess FDR can be related to the Rademacher complexity or Gaussian width of the class from which we choose our data-adaptive weights. We apply this general framework to various structured settings, including ordered, grouped, and low total variation structures, and get the bounds on FDR for each specific setting. We also examine the empirical performance of SABHA on fMRI activity data and on gene/drug response data, as well as on simulated data.

stat.ME↗

arXiv2017

Learning Visual N-Grams from Web Data

Real-world image recognition systems need to recognize tens of thousands of classes that constitute a plethora of visual concepts. The traditional approach of annotating thousands of images per class for training is infeasible in such a scenario, prompting the use of webly supervised data. This paper explores the training of image-recognition systems on large numbers of images and associated user comments. In particular, we develop visual n-gram models that can predict arbitrary phrases that are relevant to the content of an image. Our visual n-gram models are feed-forward convolutional networks trained using new loss functions that are inspired by n-gram models commonly used in language modeling. We demonstrate the merits of our models in phrase prediction, phrase-based image retrieval, relating images and captions, and zero-shot transfer.

cs.CV↗