arXiv subjects

Ping Wang

Publications and source records attributed to Ping Wang.

At least 217 records · Page 12Linked to original sources

arXiv2022

Localization Distillation for Dense Object Detection

Knowledge distillation (KD) has witnessed its powerful capability in learning compact models in object detection. Previous KD methods for object detection mostly focus on imitating deep features within the imitation regions instead of mimicking classification logit due to its inefficiency in distilling localization information and trivial improvement. In this paper, by reformulating the knowledge distillation process on localization, we present a novel localization distillation (LD) method which can efficiently transfer the localization knowledge from the teacher to the student. Moreover, we also heuristically introduce the concept of valuable localization region that can aid to selectively distill the semantic and localization knowledge for a certain region. Combining these two new components, for the first time, we show that logit mimicking can outperform feature imitation and localization knowledge distillation is more important and efficient than semantic knowledge for distilling object detectors. Our distillation scheme is simple as well as effective and can be easily applied to different dense object detectors. Experiments show that our LD can boost the AP score of GFocal-ResNet-50 with a single-scale 1x training schedule from 40.1 to 42.1 on the COCO benchmark without any sacrifice on the inference speed. Our source code and trained models are publicly available at https://github.com/HikariTJU/LD

cs.CV↗

arXiv2022

An Ultra-Reliable Low-Latency Non-Binary Polar Coded SCMA Scheme

The joint transmission scheme of polar codes and sparse code multiple access (SCMA) has been regarded as a promising technology for future wireless communication systems. However, most of the existing polar-coded SCMA (PC-SCMA) systems suffer from high latency caused by the feedback iteration and list decoding. In addition, the error performance of PC-SCMA systems is unsatisfactory for ultra-reliable transmission. Inspired by the compelling benefits of non-binary polar codes, in this paper, we design a non-binary polar-coded SCMA (NB-PC-SCMA) system with a free order matching strategy to address the issues of delay and reliability. Specifically, we first formulate a joint factor graph for NB-PC-SCMA and propose a non-binary successive cancellation list (NB-SCL) and damping based joint iterative detection and decoding (NSD-JIDD) multiuser receiver to improve the BER and latency performance. Then, a lazy-search based NB-SCL (L-NB-SCL) decoding is proposed to reduce the computational complexity by simplifying the path search pattern of the list decoder. After that, we modify the update of user nodes for SCMA detection to improve the convergence error and finally propose the improved NSD-JIDD (ISD-JIDD) algorithm, which can avoid redundant operations by exploiting L-NB-SCL decoding. Simulation results show that the proposed NB-PC-SCMA system achieves better bit error rate (BER) performance and considerable latency gain when compared to its counterparts. In particular, the proposed ISD-JIDD can achieve similar BER performance of NSD-JIDD with less complexity.

cs.IT↗

arXiv2022

Insight-HXMT dedicated 33-day observation of SGR J1935+2154 I. Burst Catalog

Magnetars are neutron stars with extreme magnetic field and sometimes manifest as soft gamma-ray repeaters (SGRs). SGR J1935+2154 is one of the most prolific bursters and the first confirmed source of fast radio burst (i.e. FRB 200428). Encouraged by the discovery of the first X-ray counterpart of FRB, Insight-Hard X-ray Modulation Telescope (Insight-HXMT) implemented a dedicated 33-day long ToO observation of SGR J1935+2154 since April 28, 2020. With the HE, ME, and LE telescopes, Insight-HXMT provides a thorough monitoring of burst activity evolution of SGR J1935+2154, in a very broad energy range (1-250 keV) with high temporal resolution and high sensitivity, resulting in a unique valuable data set for detailed studies of SGR J1935+2154. In this work, we conduct a comprehensive analysis of this observation including detailed burst search, identification and temporal analyses. After carefully removing false triggers, we find a total of 75 bursts from SGR J1935+2154, out of which 70 are single-pulsed. The maximum burst rate is about 56 bursts/day. Both the burst duration and the waiting time between two successive bursts follow log-normal distributions, consistent with previous studies. We also find that bursts with longer duration (some are multi-pulsed) tend to occur during the period with relatively high burst rate. There is no correlation between the waiting time and the fluence or duration of either the former or latter burst. It also seems that there is no correlation between burst duration and hardness ratio, in contrast to some previous reports. In addition, we do not find any X-ray burst associated with any reported radio bursts except for FRB 200428.

astro-ph.HE↗

arXiv2022

A Deep Reinforcement Learning-Based Caching Strategy for IoT Networks with Transient Data

The Internet of Things (IoT) has been continuously rising in the past few years, and its potentials are now more apparent. However, transient data generation and limited energy resources are the major bottlenecks of these networks. Besides, minimum delay and other conventional quality of service measurements are still valid requirements to meet. An efficient caching policy can help meet the standard quality of service requirements while bypassing IoT networks' specific limitations. Adopting deep reinforcement learning (DRL) algorithms enables us to develop an effective caching scheme without the need for any prior knowledge or contextual information. In this work, we propose a DRL-based caching scheme that improves the cache hit rate and reduces energy consumption of the IoT networks, in the meanwhile, taking data freshness and limited lifetime of IoT data into account. To better capture the regional-different popularity distribution, we propose a hierarchical architecture to deploy edge caching nodes in IoT networks. The results of comprehensive experiments show that our proposed method outperforms the well-known conventional caching policies and an existing DRL-based solution in terms of cache hit rate and energy consumption of the IoT networks by considerable margins.

cs.NI↗

arXiv2022

Boundary-aware Information Maximization for Self-supervised Medical Image Segmentation

Unsupervised pre-training has been proven as an effective approach to boost various downstream tasks given limited labeled data. Among various methods, contrastive learning learns a discriminative representation by constructing positive and negative pairs. However, it is not trivial to build reasonable pairs for a segmentation task in an unsupervised way. In this work, we propose a novel unsupervised pre-training framework that avoids the drawback of contrastive learning. Our framework consists of two principles: unsupervised over-segmentation as a pre-train task using mutual information maximization and boundary-aware preserving learning. Experimental results on two benchmark medical segmentation datasets reveal our method's effectiveness in improving segmentation performance when few annotated images are available.

eess.IV↗

arXiv2022

Node Selection Toward Faster Convergence for Federated Learning on Non-IID Data

Federated Learning (FL) is a distributed learning paradigm that enables a large number of resource-limited nodes to collaboratively train a model without data sharing. The non-independent-and-identically-distributed (non-i.i.d.) data samples invoke discrepancies between the global and local objectives, making the FL model slow to converge. In this paper, we proposed Optimal Aggregation algorithm for better aggregation, which finds out the optimal subset of local updates of participating nodes in each global round, by identifying and excluding the adverse local updates via checking the relationship between the local gradient and the global gradient. Then, we proposed a Probabilistic Node Selection framework (FedPNS) to dynamically change the probability for each node to be selected based on the output of Optimal Aggregation. FedPNS can preferentially select nodes that propel faster model convergence. The unbiasedness of the proposed FedPNS design is illustrated and the convergence rate improvement of FedPNS over the commonly adopted Federated Averaging (FedAvg) algorithm is analyzed theoretically. Experimental results demonstrate the effectiveness of FedPNS in accelerating the FL convergence rate, as compared to FedAvg with random node selection.

cs.LG↗

arXiv2021

Randomized Response Mechanisms for Differential Privacy Data Analysis: Bounds and Applications

Randomized response, as a basic building-block for differentially private mechanism, has given rise to great interest and found various potential applications in science communities. In this work, we are concerned with three-elements randomized response (RR$_{3}$) along with relevant applications to the analysis of weighted bipartite graph upon differentially private guarantee. We develop a principled framework for estimating statistics produced by RR$_{3}$-based mechanisms, and then prove the corresponding estimations to be unbiased. At the same time, we study in detail several fundamental and significant members in RR$_{3}$ family, and derive the closed-form solutions to unbiased estimations. Next, we show potential applications of several RR$_{3}$-based mechanisms into the estimation of average degree and average weighted value on weighted bipartite graph when requiring local differential privacy guarantee. In the meantime, we determine the lower bounds for choice of relevant parameters by minimizing variance of statistics in order to design optimal RR$_{3}$-based local differential private mechanisms, with which we optimize previous protocols in the literature and put forward a version that achieves the tight bound. Last but most importantly, we observe that in the analysis of relational data such as weighted bipartite graph, a portion of privacy budget in local differential private mechanism is sometimes "consumed" by mechanism itself accidentally, resulting to a more stronger privacy guarantee than we would get by simply sequential compositions.

cs.CR↗

arXiv2021

Mean Hitting Time on Recursive Growth Tree Network

In this paper, we are concerned with mean hitting time $\langle\mathcal{H}\rangle$ for random walks on recursive growth tree networks that are built based on an arbitrary tree as the seed via implementing various primitive graphic operations, and propose a series of combinatorial techniques that are called Mapping Transformation to exactly determine the associated $\langle\mathcal{H}\rangle-$polynomial. Our formulas can be able to completely cover the previously published results in some well-studied and specific cases where a single edge or a star is often chose to serve as seed for creating recursive growth models. The techniques proposed are more convenient than the commonly-used spectral methods mainly because of getting around the operations of matrix inversion and multiplication. Accordingly, our results can be extended for both many other stochastic models including BA-scale-free tree and random uniform tree as well as graphs of great interest consisting of line graph of tree and Vicsek fractal network to derive numerical solutions of related structural parameters. And then, the closed-form solutions of two extensions of Wiener index with respect to multiplicative and additive degrees on an arbitrary tree are conveniently obtained as well. In addition, we discuss some extremal problems of random walks on tree networks and outline the related research directions in the next step.

math.CO↗

arXiv2021

OPTT: Optimal Piecewise Transformation Technique for Analyzing Numerical Data under Local Differential Privacy

Privacy preserving data analysis (PPDA) has received increasing attention due to a great variety of applications. Local differential privacy (LDP), as an emerging standard that is suitable for PPDA, has been widely deployed into various real-world scenarios to analyze massive data while protecting against many forms of privacy breach. In this study, we are mainly concerned with piecewise transformation technique (PTT) for analyzing numerical data under local differential privacy. We provide a principled framework for PTT in the context of LDP, based on which PTT is studied systematically. As a result, we show that (1) many members in PTTs are asymptotically optimal when used to obtain an unbiased estimator for mean of numerical data, and (2) for a given privacy budget, there is PTT that reaches the theoretical low bound with respect to variance. Next, we prove by studying two classes of PTTs in detail that (1) there do not exist optimal PTTs compared to the well-used technique, i.e., Duchi's scheme, in terms of the consistency noisy variance, (2) on the other hand, one has the ability to find a great number of PTTs that are consistently more optimal than the latter with regard to the worst-case noisy variance, which is never reported so far. When we are restricted to consider only the high privacy level, enough PTTs turn out to be optimal than the well-known Laplace mechanism. Lastly, we prove that for a family of PTTs, the correspondingly theoretical low bound of noisy variance follows $O(ε^{-2})$ when considering the high privacy level.

cs.CR↗

arXiv2021

The In-Flight Realtime Trigger and Localization Software of GECAM

Realtime trigger and localization of bursts are the key functions of GECAM, which is an all-sky gamma-ray monitor launched in Dec 10, 2020. We developed a multifunctional trigger and localization software operating on the CPU of the GECAM electronic box (EBOX). This onboard software has the following features: high trigger efficiency for real celestial bursts with a suppression of false triggers caused by charged particle bursts and background fluctuation, dedicated localization algorithm optimized for short and long bursts respetively, short time latency of the trigger information which is downlinked throught the BeiDou satellite navigation System (BDS). This paper presents the detailed design and deveopment of this trigger and localization software system of GECAM, including the main functions, general design, workflow and algorithms, as well as the verification and demonstration of this software, including the on-ground trigger tests with simulated gamma-ray bursts made by a dedicated X-ray tube and the in-flight performance to real gamma-ray bursts and magnetar bursts.

astro-ph.IM↗

arXiv2021

Inferring Prototypes for Multi-Label Few-Shot Image Classification with Word Vector Guided Attention

Multi-label few-shot image classification (ML-FSIC) is the task of assigning descriptive labels to previously unseen images, based on a small number of training examples. A key feature of the multi-label setting is that images often have multiple labels, which typically refer to different regions of the image. When estimating prototypes, in a metric-based setting, it is thus important to determine which regions are relevant for which labels, but the limited amount of training data makes this highly challenging. As a solution, in this paper we propose to use word embeddings as a form of prior knowledge about the meaning of the labels. In particular, visual prototypes are obtained by aggregating the local feature maps of the support images, using an attention mechanism that relies on the label embeddings. As an important advantage, our model can infer prototypes for unseen labels without the need for fine-tuning any model parameters, which demonstrates its strong generalization abilities. Experiments on COCO and PASCAL VOC furthermore show that our model substantially improves the current state-of-the-art.

cs.CV↗

arXiv2021

Random walks on stochastic uniform growth trees: Analytical formula for mean first-passage time

As known, the commonly-utilized ways to determine mean first-passage time $\overline{\mathcal{F}}$ for random walk on networks are mainly based on Laplacian spectra. However, methods of this type can become prohibitively complicated and even fail to work when the Laplacian matrix of network under consideration is difficult to describe in the first place. In this paper, we propose an effective approach to determining quantity $\overline{\mathcal{F}}$ on some widely-studied tree networks. To this end, we first build up a general formula between Wiener index $\mathcal{W}$ and $\overline{\mathcal{F}}$ on a tree. This enables us to convert issues to answer into calculation of $\mathcal{W}$ on networks in question. As opposed to most of previous work focusing on deterministic growth trees, our goal is to consider stochastic case. Towards this end, we establish a principled framework where randomness is introduced into the process of growing trees. As an immediate consequence, the previously published results upon deterministic cases are thoroughly covered by formulas established in this paper. Additionally, it is also straightforward to obtain Kirchhoff index on our tree networks using the proposed approach. Most importantly, our approach is more manageable than many other methods including spectral technique in situations considered herein.

math.PR↗

arXiv2021

Sea quark contributions to the electromagnetic form factors of $Σ$ hyperons

We study the sea quark contributions to the electromagnetic form factors of $Σ$ baryons with nonlocal chiral effective theory. Both octet and decuplet intermediate states are included in the one loop calculation. $G_{Σ^{-}}^{u}$ and $G_{Σ^{+}}^{d}$ could be priority observables for the examination of sea quark contributions to baryon structure because these quantities are much larger than the strange form factors of nucleon. It will be less difficult for lattice simulation to determine the sign of these pure sea quark contributions unambiguously. In $Σ^0$, the light sea quark form factors $G_{Σ^{0}}^{u}$ and $G_{Σ^{0}}^{d}$ are identical. Since the light sea quark form factors in proton are different, it will be more meaningful to compare lattice result of the light sea quark form factors in $Σ^0$ with that obtained from effective field theory.

hep-ph↗

arXiv2021

Quantum nonlinear spectroscopy of single nuclear spins

Nonlinear spectroscopy is widely used for studying physical systems. Conventional nonlinear optical spectroscopy and magnetic resonance spectroscopy, which use classical probes such as electromagnetic waves, can only access certain types of correlations in a quantum system. The idea of quantum nonlinear spectroscopy was recently proposed to use quantum probes such as entangled photons to achieve sensitivities and resolutions beyond the classical limits. It is shown that quantum sensing can extract arbitrary types and orders of correlations in a quantum system by first quantum-entangling a sensor and the object and then measuring the sensor. Quantum sensing has been applied to achieve nuclear magnetic resonance (NMR) of single atoms and the second-order correlation spectroscopy has been adopted to enhance the spectral resolution. However, quantum nonlinear spectroscopy (i.e., the measurement of higher-order correlations) of single nuclear spins is still elusive. Here we demonstrate the extraction of fourth-order correlations of single nuclear spins that cannot be measured in conventional nonlinear spectroscopy, using sequential weak measurement via an atomic quantum sensor, namely, a nitrogen-vacancy center in diamond. We show that the quantum nonlinear spectroscopy provides fingerprint features to identify different types of objects, such as Gaussian noises, random-phased AC fields, and quantum spins, which would be indistinguishable in second-order correlations. The measured fourth-order correlation unambiguously differentiates a single nuclear spin and a random-phased AC field. This work constitutes an initial step toward the application of higher-order correlations to quantum sensing, to examining the quantum foundation (by, e.g., higher-order Leggett-Garg inequality), and to studying quantum many-body physics.

quant-ph↗

arXiv2021

GRB 210121A: A Typical Fireball Burst Detected by Two Small Missions

The Chinese CubeSat Mission, Gamma Ray Integrated Detectors (GRID), recently detected its first gamma-ray burst, GRB 210121A, which was jointly observed by the Gravitational wave high-energy Electromagnetic Counterpart All-sky Monitor (GECAM). This burst is confirmed by several other missions, including \fermi and \textit{Insight}-HXMT. We combined multi-mission observational data and performed a comprehensive analysis of the burst's temporal and spectral properties. Our results show that the burst is relatively special in its high peak energy, thermal-like low energy indices, and large fluence. By putting it to the $E_{\rm p}$-$E_{\rmγ, iso}$ relation diagram with assumed distance, we found this burst can be constrained at the redshift range of [0.3,3.0]. The thermal spectral component is also confirmed by the direct fit of the physical models to the observed spectra. Interestingly, the physical photosphere model also constrained a redshift of $z\sim$ 0.3 for this burst, which help us to identify a host galaxy candidate at such a distance within the location error box. Assuming the host galaxy is real, we found the burst can be best explained by the photosphere emission of a typical fireball with an initial radius of $r_0\sim$ 3.2 $\times 10^7$ cm.

astro-ph.HE↗

arXiv2021

Self-Paced Contrastive Learning for Semi-supervised Medical Image Segmentation with Meta-labels

Pre-training a recognition model with contrastive learning on a large dataset of unlabeled data has shown great potential to boost the performance of a downstream task, e.g., image classification. However, in domains such as medical imaging, collecting unlabeled data can be challenging and expensive. In this work, we propose to adapt contrastive learning to work with meta-label annotations, for improving the model's performance in medical image segmentation even when no additional unlabeled data is available. Meta-labels such as the location of a 2D slice in a 3D MRI scan or the type of device used, often come for free during the acquisition process. We use the meta-labels for pre-training the image encoder as well as to regularize a semi-supervised training, in which a reduced set of annotated data is used for training. Finally, to fully exploit the weak annotations, a self-paced learning approach is used to help the learning and discriminate useful labels from noise. Results on three different medical image segmentation datasets show that our approach: i) highly boosts the performance of a model trained on a few scans, ii) outperforms previous contrastive and semi-supervised approaches, and iii) reaches close to the performance of a model trained on the full data.

cs.CV↗

arXiv2021

Delving into Deep Image Prior for Adversarial Defense: A Novel Reconstruction-based Defense Framework

Deep learning based image classification models are shown vulnerable to adversarial attacks by injecting deliberately crafted noises to clean images. To defend against adversarial attacks in a training-free and attack-agnostic manner, this work proposes a novel and effective reconstruction-based defense framework by delving into deep image prior (DIP). Fundamentally different from existing reconstruction-based defenses, the proposed method analyzes and explicitly incorporates the model decision process into our defense. Given an adversarial image, firstly we map its reconstructed images during DIP optimization to the model decision space, where cross-boundary images can be detected and on-boundary images can be further localized. Then, adversarial noise is purified by perturbing on-boundary images along the reverse direction to the adversarial image. Finally, on-manifold images are stitched to construct an image that can be correctly predicted by the victim classifier. Extensive experiments demonstrate that the proposed method outperforms existing state-of-the-art reconstruction-based methods both in defending white-box attacks and defense-aware attacks. Moreover, the proposed method can maintain a high visual quality during adversarial image reconstruction.

cs.CV↗

arXiv2021

Effects of local decoherence on quantum critical metrology

The diverging responses to parameter variations of systems at quantum critical points motivate schemes of quantum metrology that feature sub-Heisenberg scaling of the sensitivity with the system size (e.g., the number of particles). This sensitivity enhancement is fundamentally rooted in the formation of Schrödinger cat states, or macroscopic superposition states at the quantum critical points. The cat states, however, are fragile to decoherence caused by local noises on individual particles or coupling to local environments, since the local decoherence of any particle would cause the collapse of the whole cat state. Therefore, it is unclear whether the sub-Heisenberg scaling of quantum critical metrology is robust against the local decoherence. Here we study the effects of local decoherence on the quantum critical metrology, using a one-dimensional transverse-field Ising model as a representative example. Based on a previous work [Phys. Rev. Lett. 94, 047201 (2005)] on the critical behaviors of the noisy Ising model, which shows that the universality class of the quantum criticality is modified by the decoherence, we find that the standard quantum limit is recovered by the single-particle decoherence, which is equivalent to local quantum measurement conducted by the environment and destroys the many-body entanglement in the ground state at the quantum critical point. Following the renormalization group analysis [Phys. Rev. B 69, 054426 (2004)], we argue that the noise effects on quantum critical metrology should be universal. This works demonstrates the importance of protecting macroscopic quantum coherence for quantum sensing based on critical behaviors.

quant-ph↗