arXiv subjects

Sen Yang

Publications and source records attributed to Sen Yang.

At least 181 records · Page 10Linked to original sources

arXiv2021

GDP: Stabilized Neural Network Pruning via Gates with Differentiable Polarization

Model compression techniques are recently gaining explosive attention for obtaining efficient AI models for various real-time applications. Channel pruning is one important compression strategy and is widely used in slimming various DNNs. Previous gate-based or importance-based pruning methods aim to remove channels whose importance is smallest. However, it remains unclear what criteria the channel importance should be measured on, leading to various channel selection heuristics. Some other sampling-based pruning methods deploy sampling strategies to train sub-nets, which often causes the training instability and the compressed model's degraded performance. In view of the research gaps, we present a new module named Gates with Differentiable Polarization (GDP), inspired by principled optimization ideas. GDP can be plugged before convolutional layers without bells and whistles, to control the on-and-off of each channel or whole layer block. During the training process, the polarization effect will drive a subset of gates to smoothly decrease to exact zero, while other gates gradually stay away from zero by a large margin. When training terminates, those zero-gated channels can be painlessly removed, while other non-zero gates can be absorbed into the succeeding convolution kernel, causing completely no interruption to training nor damage to the trained model. Experiments conducted over CIFAR-10 and ImageNet datasets show that the proposed GDP algorithm achieves the state-of-the-art performance on various benchmark DNNs at a broad range of pruning ratios. We also apply GDP to DeepLabV3Plus-ResNet50 on the challenging Pascal VOC segmentation task, whose test performance sees no drop (even slightly improved) with over 60% FLOPs saving.

cs.CV↗

arXiv2021

TransPose: Keypoint Localization via Transformer

While CNN-based models have made remarkable progress on human pose estimation, what spatial dependencies they capture to localize keypoints remains unclear. In this work, we propose a model called \textbf{TransPose}, which introduces Transformer for human pose estimation. The attention layers built in Transformer enable our model to capture long-range relationships efficiently and also can reveal what dependencies the predicted keypoints rely on. To predict keypoint heatmaps, the last attention layer acts as an aggregator, which collects contributions from image clues and forms maximum positions of keypoints. Such a heatmap-based localization approach via Transformer conforms to the principle of Activation Maximization~\cite{erhan2009visualizing}. And the revealed dependencies are image-specific and fine-grained, which also can provide evidence of how the model handles special cases, e.g., occlusion. The experiments show that TransPose achieves 75.8 AP and 75.0 AP on COCO validation and test-dev sets, while being more lightweight and faster than mainstream CNN architectures. The TransPose model also transfers very well on MPII benchmark, achieving superior performance on the test set when fine-tuned with small training costs. Code and pre-trained models are publicly available\footnote{\url{https://github.com/yangsenius/TransPose}}.

cs.CV↗

arXiv2021

POSO: Personalized Cold Start Modules for Large-scale Recommender Systems

Recommendation for new users, also called user cold start, has been a well-recognized challenge for online recommender systems. Most existing methods view the crux as the lack of initial data. However, in this paper, we argue that there are neglected problems: 1) New users' behaviour follows much different distributions from regular users. 2) Although personalized features are involved, heavily imbalanced samples prevent the model from balancing new/regular user distributions, as if the personalized features are overwhelmed. We name the problem as the "submergence" of personalization. To tackle this problem, we propose a novel module: Personalized COld Start MOdules (POSO). Considering from a model architecture perspective, POSO personalizes existing modules by introducing multiple user-group-specialized sub-modules. Then, it fuses their outputs by personalized gates, resulting in comprehensive representations. In such way, POSO projects imbalanced features to even modules. POSO can be flexibly integrated into many existing modules and effectively improves their performance with negligible computational overheads. The proposed method shows remarkable advantage in industrial scenario. It has been deployed on the large-scale recommender system of Kwai, and improves new user Watch Time by a large margin (+7.75%). Moreover, POSO can be further generalized to regular users, inactive users and returning users (+2%-3% on Watch Time), as well as item cold start (+3.8% on Watch Time). Its effectiveness has also been verified on public dataset (MovieLens 20M). We believe such practical experience can be well generalized to other scenarios.

cs.IR↗

arXiv2021

PASTO: Strategic Parameter Optimization in Recommendation Systems -- Probabilistic is Better than Deterministic

Real-world recommendation systems often consist of two phases. In the first phase, multiple predictive models produce the probability of different immediate user actions. In the second phase, these predictions are aggregated according to a set of 'strategic parameters' to meet a diverse set of business goals, such as longer user engagement, higher revenue potential, or more community/network interactions. In addition to building accurate predictive models, it is also crucial to optimize this set of 'strategic parameters' so that primary goals are optimized while secondary guardrails are not hurt. In this setting with multiple and constrained goals, this paper discovers that a probabilistic strategic parameter regime can achieve better value compared to the standard regime of finding a single deterministic parameter. The new probabilistic regime is to learn the best distribution over strategic parameter choices and sample one strategic parameter from the distribution when each user visits the platform. To pursue the optimal probabilistic solution, we formulate the problem into a stochastic compositional optimization problem, in which the unbiased stochastic gradient is unavailable. Our approach is applied in a popular social network platform with hundreds of millions of daily users and achieves +0.22% lift of user engagement in a recommendation task and +1.7% lift in revenue in an advertising optimization scenario comparing to using the best deterministic parameter strategy.

cs.LG↗

arXiv2021

TokenPose: Learning Keypoint Tokens for Human Pose Estimation

Human pose estimation deeply relies on visual clues and anatomical constraints between parts to locate keypoints. Most existing CNN-based methods do well in visual representation, however, lacking in the ability to explicitly learn the constraint relationships between keypoints. In this paper, we propose a novel approach based on Token representation for human Pose estimation~(TokenPose). In detail, each keypoint is explicitly embedded as a token to simultaneously learn constraint relationships and appearance cues from images. Extensive experiments show that the small and large TokenPose models are on par with state-of-the-art CNN-based counterparts while being more lightweight. Specifically, our TokenPose-S and TokenPose-L achieve $72.5$ AP and $75.8$ AP on COCO validation dataset respectively, with significant reduction in parameters ($\downarrow80.6\%$; $\downarrow$ $56.8\%$) and GFLOPs ($\downarrow$ $75.3\%$; $\downarrow$ $24.7\%$). Code is publicly available.

cs.CV↗

arXiv2021

Deep Learning methods for automatic evaluation of delayed enhancement-MRI. The results of the EMIDEC challenge

A key factor for assessing the state of the heart after myocardial infarction (MI) is to measure whether the myocardium segment is viable after reperfusion or revascularization therapy. Delayed enhancement-MRI or DE-MRI, which is performed several minutes after injection of the contrast agent, provides high contrast between viable and nonviable myocardium and is therefore a method of choice to evaluate the extent of MI. To automatically assess myocardial status, the results of the EMIDEC challenge that focused on this task are presented in this paper. The challenge's main objectives were twofold. First, to evaluate if deep learning methods can distinguish between normal and pathological cases. Second, to automatically calculate the extent of myocardial infarction. The publicly available database consists of 150 exams divided into 50 cases with normal MRI after injection of a contrast agent and 100 cases with myocardial infarction (and then with a hyperenhanced area on DE-MRI), whatever their inclusion in the cardiac emergency department. Along with MRI, clinical characteristics are also provided. The obtained results issued from several works show that the automatic classification of an exam is a reachable task (the best method providing an accuracy of 0.92), and the automatic segmentation of the myocardium is possible. However, the segmentation of the diseased area needs to be improved, mainly due to the small size of these areas and the lack of contrast with the surrounding structures.

eess.IV↗

arXiv2021

Cardiac Segmentation on Late Gadolinium Enhancement MRI: A Benchmark Study from Multi-Sequence Cardiac MR Segmentation Challenge

Accurate computing, analysis and modeling of the ventricles and myocardium from medical images are important, especially in the diagnosis and treatment management for patients suffering from myocardial infarction (MI). Late gadolinium enhancement (LGE) cardiac magnetic resonance (CMR) provides an important protocol to visualize MI. However, automated segmentation of LGE CMR is still challenging, due to the indistinguishable boundaries, heterogeneous intensity distribution and complex enhancement patterns of pathological myocardium from LGE CMR. Furthermore, compared with the other sequences LGE CMR images with gold standard labels are particularly limited, which represents another obstacle for developing novel algorithms for automatic segmentation of LGE CMR. This paper presents the selective results from the Multi-Sequence Cardiac MR (MS-CMR) Segmentation challenge, in conjunction with MICCAI 2019. The challenge offered a data set of paired MS-CMR images, including auxiliary CMR sequences as well as LGE CMR, from 45 patients who underwent cardiomyopathy. It was aimed to develop new algorithms, as well as benchmark existing ones for LGE CMR segmentation and compare them objectively. In addition, the paired MS-CMR images could enable algorithms to combine the complementary information from the other sequences for the segmentation of LGE CMR. Nine representative works were selected for evaluation and comparisons, among which three methods are unsupervised methods and the other six are supervised. The results showed that the average performance of the nine methods was comparable to the inter-observer variations. The success of these methods was mainly attributed to the inclusion of the auxiliary sequences from the MS-CMR images, which provide important label information for the training of deep neural networks.

eess.IV↗

arXiv2021

Reverberation Mapping of Two Luminous Quasars: the Broad-line Region Structure and Black Hole Mass

We report the results of a multi-year spectroscopic and photometric monitoring campaign of two luminous quasars, PG~0923+201 and PG~1001+291, both located at the high-luminosity end of the broad-line region (BLR) size-luminosity relation with optical luminosities above $10^{45}~{\rm erg~s^{-1}}$. PG~0923+201 is for the first time monitored, and PG~1001+291 was previously monitored but our campaign has a much longer temporal baseline. We detect time lags of variations of the broad H$β$, H$γ$, Fe {\sc ii} lines with respect to those of the 5100~Å continuum. The velocity-resolved delay map of H$β$ in PG~0923+201 indicates a complicated structure with a mix of Keplerian disk-like motion and outflow, and the map of H$β$ in PG~1001+291 shows a signature of Keplerian disk-like motion. Assuming a virial factor of $f_{\rm BLR}=1$ and FWHM line widths, we measure the black hole mass to be $118_{-16}^{+11}\times 10^7 M_{\odot}$ for PG~0923+201 and $3.33_{-0.54}^{+0.62}\times 10^7 M_{\odot}$ for PG~1001+291. Their respective accretion rates are estimated to be $0.21_{-0.07}^{+0.06} \times L_{\rm Edd}\,c^{-2}$ and $679_{-227}^{+259}\times L_{\rm Edd}\,c^{-2}$, indicating that PG~0923+201 is a sub-Eddington accretor and PG~1001+291 is a super-Eddington accretor. While the H$β$ time lag of PG~0923+201 agrees with the size-luminosity relation, the time lag of PG~1001+291 shows a significant deviation, confirming that in high-luminosity AGN the BLR size depends on both luminosity and Eddington ratio. Black hole mass estimates from single AGN spectra will be over-estimated at high luminosities and redshifts if this effect is not taken into account.

astro-ph.GA↗

arXiv2021

Template-Based Named Entity Recognition Using BART

There is a recent interest in investigating few-shot NER, where the low-resource target domain has different label sets compared with a resource-rich source domain. Existing methods use a similarity-based metric. However, they cannot make full use of knowledge transfer in NER model parameters. To address the issue, we propose a template-based method for NER, treating NER as a language model ranking problem in a sequence-to-sequence framework, where original sentences and statement templates filled by candidate named entity span are regarded as the source sequence and the target sequence, respectively. For inference, the model is required to classify each candidate span based on the corresponding template scores. Our experiments demonstrate that the proposed method achieves 92.55% F1 score on the CoNLL03 (rich-resource task), and significantly better than fine-tuning BERT 10.88%, 15.34%, and 11.73% F1 score on the MIT Movie, the MIT Restaurant, and the ATIS (low-resource task), respectively.

cs.CL↗

arXiv2021

AGN STORM 2: I. First results: A Change in the Weather of Mrk 817

We present the first results from the ongoing, intensive, multi-wavelength monitoring program of the luminous Seyfert 1 galaxy Mrk 817. While this AGN was, in part, selected for its historically unobscured nature, we discovered that the X-ray spectrum is highly absorbed, and there are new blueshifted, broad and narrow UV absorption lines, which suggest that a dust-free, ionized obscurer located at the inner broad line region partially covers the central source. Despite the obscuration, we measure UV and optical continuum reverberation lags consistent with a centrally illuminated Shakura-Sunyaev thin accretion disk, and measure reverberation lags associated with the optical broad line region, as expected. However, in the first 55 days of the campaign, when the obscuration was becoming most extreme, we observe a de-coupling of the UV continuum and the UV broad emission line variability. The correlation recovers in the next 42 days of the campaign, as Mrk 817 enters a less obscured state. The short CIV and Ly alpha lags suggest that the accretion disk extends beyond the UV broad line region.

astro-ph.HE↗

arXiv2021

Chalcogenide perovskite BaZrS3 thin-film electronic and optoelectronic devices by low temperature processing

Owing to its superior visible light absorption and high chemical stability, chalcogenide perovskite barium zirconium sulfide has attracted significant attention in the past few years as a potential alternative to hybrid halide perovskites for optoelectronics. However, the high processing temperatures of BaZrS3 thin films at above 1000 C severely limits their potential for device applications. Herein, we report the synthesis of BaZrS3 thin films at temperatures as low as 500 C, by changing the chemical reaction pathway. The single phase BaZrS3 thin film was confirmed by X-ray diffraction and Raman spectroscopies. Atomic force microscopy and scanning electron microscopy show that crystalline size and surface roughness were consistently reduced with decreasing annealing temperature. The lower temperatures further eliminate sulfur vacancies and carbon contaminations associated with high temperature processing. The ability to synthesize chalcogenide perovskite thin films at lower temperatures removes a major hurdle for their device fabrication. The photodetectors demonstrate fast response and an on/off ratio of 80. The fabricated field effect transistors show an ambipolar behavior with electron and hole mobilities of 16.8 cm2/Vs and 2.6 cm2/Vs, respectively.

cond-mat.mtrl-sci↗

arXiv2020

In-Situ Studies of Stress Environment in Amorphous Solids Using Negatively Charged Nitrogen Vacancy Centers in Nanodiamond

Amorphous solids, which show characteristic differences from crystals, are common in daily usage. Glasses, gels, and polymers are familiar examples, and polymers are particularly important in terms of their role in construction and crafting. Previous studies have mainly focused on the bulk properties of polymeric products, and the local properties are less discussed. Here, we designed a distinctive protocol using the negatively charged nitrogen vacancy center in nanodiamond to study properties inside polymeric products in situ. Choosing the curing of poly dimethylsiloxane and the polymerization of cyanoacrylate as subjects of investigation, we measured the time dependence of local pressure and strain in the materials during the chemical processes. From the measurements, we were able to probe the local shear stress inside the two polymeric substances in situ. By regarding the surprisingly large shear stress as the internal tension, we attempted to provide a microscopic explanation for the ultimate tensile strength of a bulk solid. Our current methodology is applicable to any kind of transparent amorphous solids with the stress in the order of MPa and to the study of in situ properties in nanoscale. With better apparatus, we expect the limit can be pushed to sub-MPa scale.

cond-mat.mtrl-sci↗

arXiv2020

Pose Neural Fabrics Search

Neural Architecture Search (NAS) technologies have emerged in many domains to jointly learn the architectures and weights of the neural network. However, most existing NAS works claim they are task-specific and focus only on optimizing a single architecture to replace a human-designed neural network, in fact, their search processes are almost independent of domain knowledge of the tasks. In this paper, we propose Pose Neural Fabrics Search (PoseNFS). We explore a new solution for NAS and human pose estimation task: part-specific neural architecture search, which can be seen as a variant of multi-task learning. Firstly, we design a new neural architecture search space, Cell-based Neural Fabric (CNF), to learn micro as well as macro neural architecture using a differentiable search strategy. Then, we view locating human keypoints as multiple disentangled prediction sub-tasks, and then use prior knowledge of body structure as guidance to search for multiple part-specific neural architectures for different human parts. After search, all these part-specific CNFs have distinct micro and macro architecture parameters. The results show that such knowledge-guided NAS-based architectures have obvious performance improvements to a hand-designed part-based baseline model. The experiments on MPII and MS-COCO datasets demonstrate that PoseNFS\footnote{Code is available at \url{https://github.com/yangsenius/PoseNFS}} can achieve comparable performance to some efficient and state-of-the-art methods.

cs.CV↗

arXiv2020

Nanoscale electric-field imaging based on a quantum sensor and its charge-state control under ambient condition

Nitrogen-vacancy (NV) centers in diamond can be used as quantum sensors to image the magnetic field with nanoscale resolution. However, nanoscale electric-field mapping has not been achieved so far because of the relatively weak coupling strength between NV and electric field. Using individual shallow NVs, here we succeeded to quantitatively image the contours of electric field from a sharp tip of a qPlus-based atomic force microscope (AFM), and achieved a spatial resolution of ~10 nm. Through such local electric fields, we demonstrated electric control of NV's charge state with sub-5 nm precision. This work represents the first step towards nanoscale scanning electrometry based on a single quantum sensor and may open up new possibility of quantitatively mapping local charge, electric polarization, and dielectric response in a broad spectrum of functional materials at nanoscale.

cond-mat.mes-hall↗

arXiv2020

Making the Best Use of Review Summary for Sentiment Analysis

Sentiment analysis provides a useful overview of customer review contents. Many review websites allow a user to enter a summary in addition to a full review. Intuitively, summary information may give additional benefit for review sentiment analysis. In this paper, we conduct a study to exploit methods for better use of summary information. We start by finding out that the sentimental signal distribution of a review and that of its corresponding summary are in fact complementary to each other. We thus explore various architectures to better guide the interactions between the two and propose a hierarchically-refined review-centric attention model. Empirical results show that our review-centric model can make better use of user-written summaries for review sentiment analysis, and is also more effective compared to existing methods when the user summary is replaced with summary generated by an automatic summarization system.

cs.CL↗

arXiv2020

Optimal Quaternary Hermitian LCD codes

Linear complementary dual (LCD) codes, which is a class of linear codes introduced by Massey, have been extensively studied in literature recently. It has been shown that LCD codes can help to improve the security of the information processed by sensitive devices, especially against so-called side-channel attacks (SCA) and fault invasive attacks. In this paper, Tables are presented of good quaternary Hermitian LCD codes and there are used in the construction of puncturing, extending, shortening and combination codes. Results including tables 3 of the best-known quaternary Hermitian LCD codes of any length $ n \leq 25$ with corresponding dimension $k$ are presented. In addition, Many of these quaternary Hermitian LCD codes given in this paper are optimal which are saturating the lower or upper bound of Grassl's codetable in \cite{Grassl} and some of them are nearly optimal.

cs.IT↗

arXiv2020

Evidence for Two Distinct Broad-Line Regions from Reverberation Mapping of PG 0026+129

We report on the results of a new spectroscopic monitoring campaign of the quasar PG 0026+129 at the Calar Alto Observatory 2.2m telescope from July 2017 to February 2020. Significant variations in the fluxes of the continuum and broad-emission lines, including H$β$ and He II, were observed in the first and third years, and clear time lags between them are measured. The broad H$β$ line profile consists of two Gaussian components: an intermediate-width H$β_{\rm IC}$ with a full width at half-maximum (FWHM) of 1964$\pm$18 $\rm km~s^{-1}$ and another very broad H$β_{\rm VBC}$ with a FWHM of 7570$\pm$83 $\rm km~s^{-1}$. H$β_{\rm IC}$ has long time lags of $\sim$40--60 days in the rest frame, while H$β_{\rm VBC}$ shows nearly zero time delay with respect to the optical continuum at 5100 Å. The velocity-resolved delays show consistent results: lags of $\sim$30--50 days at the core of the broad H$β$ line and roughly zero lags at the wings. H$β_{\rm IC}$ has a redshift of $\sim$400 $\rm km~s^{-1}$ which seems to be stable for nearly 30 years by comparing with archived spectra, and may originate from an infall. The root mean square (rms) spectrum of H$β_{\rm VBC}$ shows a double-peaked profile with brighter blue peak and extended red wing in the first year, which matches the signature of a thin disk. Both the double-peaked profile and the near-zero lag suggest that H$β_{\rm VBC}$ comes from a region associated with the part of the accretion disc that emits the optical continuum. Adopting the FWHM (in the rms spectrum) and the time lag measured for the total H$β$ line, and a virial factor of 1.5, we obtain a virial mass of $2.89_{-0.69}^{+0.60} \times10^7 M_{\odot}$ for the central black hole in this quasar.

astro-ph.GA↗

arXiv2020

A universal method for depositing patterned materials in-situ

Current techniques of patterned material deposition require separate steps for patterning and material deposition. The complexity and harsh working conditions post serious limitations for fabrication. Here, we introduce a novel single-step and easy-to-adapt method that can deposit materials in-situ. Its unique methodology is based on the semiconductor nanoparticle assisted photon-induced chemical reduction and optical trapping. This universal mechanism can be used for depositing a large selection of materials including metals, insulators and magnets, with quality on par with current technologies. Patterning with several materials together with optical-diffraction-limited resolution accuracy can be achieved from macroscopic to microscopic scale. Furthermore, the setup is naturally compatible with optical microscopy based measurements, thus sample characterisation and material deposition can be realised in-situ. Various devices fabricated with this method in 2D or 3D show it is ready for deployment in practical applications. This revolutionary method will provide a distinct tool in material technology.

cond-mat.mes-hall↗