arXiv subjects

Liang Zhang

Publications and source records attributed to Liang Zhang.

At least 325 records · Page 18Linked to original sources

arXiv2022

Generalizing Multimodal Pre-training into Multilingual via Language Acquisition

English-based Vision-Language Pre-training (VLP) has achieved great success in various downstream tasks. Some efforts have been taken to generalize this success to non-English languages through Multilingual Vision-Language Pre-training (M-VLP). However, due to the large number of languages, M-VLP models often require huge computing resources and cannot be flexibly extended to new languages. In this work, we propose a \textbf{M}ulti\textbf{L}ingual \textbf{A}cquisition (MLA) framework that can easily generalize a monolingual Vision-Language Pre-training model into multilingual. Specifically, we design a lightweight language acquisition encoder based on state-of-the-art monolingual VLP models. We further propose a two-stage training strategy to optimize the language acquisition encoder, namely the Native Language Transfer stage and the Language Exposure stage. With much less multilingual training data and computing resources, our model achieves state-of-the-art performance on multilingual image-text and video-text retrieval benchmarks.

cs.CL↗

arXiv2022

Variability and phase lags of the type-C quasi-periodic oscillation of MAXI J1348-630 with NICER

We study the properties of the type-C quasi-periodic oscillation (type-C QPO) of MAXI J1348-630 during its 2019 outburst and reflare with NICER. This is the first time that the evolution of the properties of type-C QPOs is studied during an outburst reflare. We found that the properties of the type-C QPO during the reflare are similar to those of type-C QPOs observed in other black-hole systems during outburst. This suggests that the physical processes responsible for type-C QPOs are the same in a reflare and in an outburst. We also found that the FWHM of a high-frequency broadband component observed during the reflare changes significantly with energy. We studied the energy-dependent fractional rms amplitude and phase lags of the type-C QPO from 0.5 keV to 12 keV. We found that the fractional rms amplitude increases up to 2-3 keV and then remains approximately constant above this energy, and the lag spectra of the type-C QPO are hard. We discuss the dependence of the fractional rms amplitude and phase lags with energy in the context of Comptonisation as the radiative mechanism driving the QPO rms and lag spectra.

astro-ph.HE↗

arXiv2022

Trigger-GNN: A Trigger-Based Graph Neural Network for Nested Named Entity Recognition

Nested named entity recognition (NER) aims to identify the entity boundaries and recognize categories of the named entities in a complex hierarchical sentence. Some works have been done using character-level, word-level, or lexicon-level based models. However, such researches ignore the role of the complementary annotations. In this paper, we propose a trigger-based graph neural network (Trigger-GNN) to leverage the nested NER. It obtains the complementary annotation embeddings through entity trigger encoding and semantic matching, and tackle nested entity utilizing an efficient graph message passing architecture, aggregation-update mode. We posit that using entity triggers as external annotations can add in complementary supervision signals on the whole sentences. It helps the model to learn and generalize more efficiently and cost-effectively. Experiments show that the Trigger-GNN consistently outperforms the baselines on four public NER datasets, and it can effectively alleviate the nested NER.

cs.CL↗

arXiv2022

Production of $ΩNN$ and $ΩΩN$ in ultra-relativistic heavy ion collisions

Even though lots of $Λ$-hypernuclei have been found and measured, multi-strangeness hypernuclei consisting of $Ω$ are not yet discovered. The studies of multi-strangeness hypernuclei help us further understand the interaction between hyperons and nucleons. Recently the $ΩN$ and $ΩΩ$ interactions as well as binding energies were calculated by the HAL-QCD's lattice Quantum Chromo-Dynamics (LQCD) simulations and production rates of $Ω$-dibaryon in Au + Au collisions at RHIC and Pb + Pb collisions at LHC energies were estimated by a coalescence model. The present work discusses the production of more exotic triple-baryons including $Ω$, namely $ΩNN$ and $ΩΩN$ as well as their decay channels. A variation method is used in calculations of bound states and binding energy of $ΩNN$ and $ΩΩN$ with the potentials from the HAL-QCD's results. The productions of $ΩNN$ and $ΩΩN$ are predicted by using a blast-wave model plus coalescence model in ultra-relativistic heavy-ion collisions at $\sqrt{s_{NN}} = 200$ GeV and $2.76$ TeV. Furthermore, plots for baryon number dependent yields of different baryons ($N$ and $Ω$), their dibaryons and hypernuclei are made and the production rate of a more exotic tetra-baryon ($ΩΩNN$) is extrapolated.

hep-ph↗

arXiv2022

Learning from Pixel-Level Noisy Label : A New Perspective for Light Field Saliency Detection

Saliency detection with light field images is becoming attractive given the abundant cues available, however, this comes at the expense of large-scale pixel level annotated data which is expensive to generate. In this paper, we propose to learn light field saliency from pixel-level noisy labels obtained from unsupervised hand crafted featured based saliency methods. Given this goal, a natural question is: can we efficiently incorporate the relationships among light field cues while identifying clean labels in a unified framework? We address this question by formulating the learning as a joint optimization of intra light field features fusion stream and inter scenes correlation stream to generate the predictions. Specially, we first introduce a pixel forgetting guided fusion module to mutually enhance the light field features and exploit pixel consistency across iterations to identify noisy pixels. Next, we introduce a cross scene noise penalty loss for better reflecting latent structures of training data and enabling the learning to be invariant to noise. Extensive experiments on multiple benchmark datasets demonstrate the superiority of our framework showing that it learns saliency prediction comparable to state-of-the-art fully supervised light field saliency methods. Our code is available at https://github.com/OLobbCode/NoiseLF.

cs.CV↗

arXiv2022

The evolving properties of the corona of GRS 1915+105: A spectral-timing perspective through variable-Comptonisation modelling

The inverse Compton process by which soft photons are up-scattered by hot electrons in a corona plays a fundamental role in shaping the X-ray spectra of black-hole (BH) low-mass X-ray binaries (LMXBs), particularly in the hard and hard-intermediate states. In these states, the power-density spectra of these sources typically show Type-C low-frequency quasi-periodic oscillations (QPOs). Although several models have been proposed to explain the dynamical origin of their frequency, only a few of those models predict the spectral-timing radiative properties of the QPOs. Here we study the physical and geometrical properties of the corona of the BH-LMXB GRS 1915+105 based on a large sample of observations available in the RXTE archive. We use a recently-developed spectral-timing Comptonisation model to fit simultaneously the energy-dependent fractional rms amplitude and phase-lag spectra of the Type-C QPO in 398 observations. For this, we include spectral information gathered from fitting a Comptonisation model to the corresponding time-averaged spectra. We analyse the dependence of the physical and geometrical properties of the corona upon the QPO frequency and spectral state of the source, the latter characterised by the hardness ratio. We find consistent trends in the evolution of the corona size, temperature, and feedback (the fraction of the corona photons that impinge back onto the disc) that persist for roughly 15~years. By correlating our observations with simultaneous radio-monitoring of the source at 15 GHz, we propose a scenario in which the disc-corona interactions connect with the launching mechanism of the radio jet in this source.

astro-ph.HE↗

arXiv2022

Landing AI on Networks: An equipment vendor viewpoint on Autonomous Driving Networks

The tremendous achievements of Artificial Intelligence (AI) in computer vision, natural language processing, games and robotics, has extended the reach of the AI hype to other fields: in telecommunication networks, the long term vision is to let AI fully manage, and autonomously drive, all aspects of network operation. In this industry vision paper, we discuss challenges and opportunities of Autonomous Driving Network (ADN) driven by AI technologies. To understand how AI can be successfully landed in current and future networks, we start by outlining challenges that are specific to the networking domain, putting them in perspective with advances that AI has achieved in other fields. We then present a system view, clarifying how AI can be fitted in the network architecture. We finally discuss current achievements as well as future promises of AI in networks, mentioning a roadmap to avoid bumps in the road that leads to true large-scale deployment of AI technologies in networks.

cs.NI↗

arXiv2022

A Roadmap for Big Model

With the rapid development of deep learning, training Big Models (BMs) for multiple downstream tasks becomes a popular paradigm. Researchers have achieved various outcomes in the construction of BMs and the BM application in many fields. At present, there is a lack of research work that sorts out the overall progress of BMs and guides the follow-up research. In this paper, we cover not only the BM technologies themselves but also the prerequisites for BM training and applications with BMs, dividing the BM review into four parts: Resource, Models, Key Technologies and Application. We introduce 16 specific BM-related topics in those four parts, they are Data, Knowledge, Computing System, Parallel Training System, Language Model, Vision Model, Multi-modal Model, Theory&Interpretability, Commonsense Reasoning, Reliability&Security, Governance, Evaluation, Machine Translation, Text Generation, Dialogue and Protein Research. In each topic, we summarize clearly the current studies and propose some future research directions. At the end of this paper, we conclude the further development of BMs in a more general view.

cs.LG↗

arXiv2022

The evolution of the high-frequency variability in the black hole candidate GRS 1915+105 as seen by RXTE

GRS 1915+105 can show type-C quasi-periodic oscillations (QPOs) in the power density spectrum. A high-frequency QPO (HFQPO) at 67 Hz has been observed in this source, albeit less often than the type-C QPOs. Besides these features, GRS 1915+105 sometimes shows a broad bump in the power spectrum at around 30-150 Hz. We study the power spectra of GRS 1915+105 with the Rossi X-ray Timing Explorer when the source was in the $χ$ class. We find that the rms amplitude of the bump depends strongly upon both the frequency of the type-C QPO and the hardness ratio, and is correlated with the corona temperature and anti-correlated with the radio flux at 15 GHz. The characteristic frequency of the bump is better correlated with a combination of the frequency of the type-C QPO and the hardness ratio than with the frequency of the type-C QPO alone. The rms amplitude of the bump generally increases with energy from ~1-2% at ~3 keV to ~10-15% at ~30 keV. We suggest that the bump and the high-frequency QPO may be the same variability component but the properties of the corona affect the coherence of this variability, leading either to a HFQPO when the spectrum is in the relatively soft $γ$ class, or to a bump when the spectrum is in the hard $χ$ class. Finally, we discuss the anti-correlation between the rms amplitude of the bump and the radio flux in the context of the relation between the corona and the jet.

astro-ph.HE↗

arXiv2022

The accretion flow geometry of MAXI J1820+070 through broadband noise research with Insight-HXMT

Here we present a detailed study of the broadband noise in the power density spectra of the black hole X-ray binary MAXI J1820+070 during the hard state of its 2018 outburst, using the Hard X-ray Modulation Telescope (Insight-HXMT) observations. The broadband noise shows two main humps, which might separately correspond to variability from a variable disk and two Comptonization regions. We fitted the two humps with multiple Lorentzian functions and studied the energy-dependent properties of each component up to 100--150 keV and their evolution with spectral changes. The lowest frequency component is considered as the sub-harmonic of QPO component and shows different energy dependence compared with other broadband noise components. We found that although the fractional rms of all the broadband noise components mainly decrease with energy, their rms spectra are different in shape. Above $\sim$ 20--30 keV, the characteristic frequencies of these components increase sharply with energy, meaning that the high-energy component is more variable on short timescales. Our results suggest that the hot inner flow in MAXI J1820+070 is likely to be inhomogeneous. We propose a geometry with a truncated accretion disk, two Comptonization regions.

astro-ph.HE↗

arXiv2022

NC-DRE: Leveraging Non-entity Clue Information for Document-level Relation Extraction

Document-level relation extraction (RE), which requires reasoning on multiple entities in different sentences to identify complex inter-sentence relations, is more challenging than sentence-level RE. To extract the complex inter-sentence relations, previous studies usually employ graph neural networks (GNN) to perform inference upon heterogeneous document-graphs. Despite their great successes, these graph-based methods, which normally only consider the words within the mentions in the process of building graphs and reasoning, tend to ignore the non-entity clue words that are not in the mentions but provide important clue information for relation reasoning. To alleviate this problem, we treat graph-based document-level RE models as an encoder-decoder framework, which typically uses a pre-trained language model as the encoder and a GNN model as the decoder, and propose a novel graph-based model NC-DRE that introduces decoder-to-encoder attention mechanism to leverage Non-entity Clue information for Document-level Relation Extraction.

cs.CL↗

arXiv2022

Birth places of extreme ultraviolet waves driven by impingement of solar jets upon coronal loops

Solar extreme ultraviolet (EUV) waves are large-scale propagating disturbances in the corona. It is generally believed that the vital key for the formation of EUV waves is the rapid expansion of the loops that overlie erupting cores in solar eruptions, such as coronal mass ejections (CMEs) and solar jets. However, the details of the interaction between the erupting cores and overlying loops are not clear, because that the overlying loops are always instantly opened after the energetic eruptions. Here, we present three typical jet-driven EUV waves without CME to study the interaction between the jets and the overlying loops that remained closed during the events. All three jets emanated from magnetic flux cancelation sites in source regions. Interestingly, after the interactions between jets and overlying loops, three EUV waves respectively formed ahead of the top, the near end (close to the jet source), and the far (another) end of the overlying loops. According to the magnetic field distribution of the loops extrapolated from Potential Field Source Surface method, it is confirmed that the birth places of three jet-driven EUV waves were around the weakest magnetic field strength part of the overlying loops. We suggest that the jet-driven EUV waves preferentially occur at the weakest part of the overlying loops, and the location can be subject to the magnetic field intensity around the ends of the loops.

astro-ph.SR↗

arXiv2022

Twin extreme ultraviolet waves in the solar corona

Solar extreme ultraviolet (EUV) waves are spectacular propagating disturbances with EUV enhancements in annular shapes in the solar corona. These EUV waves carry critical information about the coronal magnetised plasma that can shed light on the elusive physical parameters (e.g. the magnetic field strength) by global solar coronal magneto-seismology. EUV waves are closely associated with a wide range of solar atmospheric eruptions, from violent flares and coronal mass ejections (CMEs) to less energetic plasma jets or mini-filament eruptions. However, the physical nature and driving mechanism of EUV waves is still controversial. Here, we report the unique discovery of twin EUV waves (TEWs) that were formed in a single eruption with observations from two different perspectives. In all earlier studies, a single eruption was associated at most with a single EUV wave. The newly found TEWs urge to re-visit our theoretical understanding about the underlying formation mechanism(s) of coronal EUV waves. Two distinct scenarios of TEWs were found. In the first scenario, the two waves were separately associated with a filament eruption and a precursor jet, while in another scenario the two waves were successively associated with a filament eruption. Hence, we label these distinguished scenarios as "fraternal TEWs" and "identical TEWs", respectively. Further, we also suggest that impulsive lateral expansions of two distinct groups of coronal loops are critical to the formation of TEWs in a single eruption.

astro-ph.SR↗

arXiv2022

A Densely Connected Criss-Cross Attention Network for Document-level Relation Extraction

Document-level relation extraction (RE) aims to identify relations between two entities in a given document. Compared with its sentence-level counterpart, document-level RE requires complex reasoning. Previous research normally completed reasoning through information propagation on the mention-level or entity-level document-graph, but rarely considered reasoning at the entity-pair-level.In this paper, we propose a novel model, called Densely Connected Criss-Cross Attention Network (Dense-CCNet), for document-level RE, which can complete logical reasoning at the entity-pair-level. Specifically, the Dense-CCNet performs entity-pair-level logical reasoning through the Criss-Cross Attention (CCA), which can collect contextual information in horizontal and vertical directions on the entity-pair matrix to enhance the corresponding entity-pair representation. In addition, we densely connect multiple layers of the CCA to simultaneously capture the features of single-hop and multi-hop logical reasoning.We evaluate our Dense-CCNet model on three public document-level RE datasets, DocRED, CDR, and GDA. Experimental results demonstrate that our model achieves state-of-the-art performance on these three datasets.

cs.CL↗

arXiv2022

Algorithm for Optimized mRNA Design Improves Stability and Immunogenicity

Messenger RNA (mRNA) vaccines are being used for COVID-19, but still suffer from the critical issue of mRNA instability and degradation, which is a major obstacle in the storage, distribution, and efficacy of the vaccine. Previous work showed that optimizing secondary structure stability lengthens mRNA half-life, which, together with optimal codons, increases protein expression. Therefore, a principled mRNA design algorithm must optimize both structural stability and codon usage to improve mRNA efficiency. However, due to synonymous codons, the mRNA design space is prohibitively large, e.g., there are $\sim\!10^{632}$ mRNAs for the SARS-CoV-2 Spike protein, which poses insurmountable challenges to previous methods. Here we provide a surprisingly simple solution to this hard problem by reducing it to a classical problem in computational linguistics, where finding the optimal mRNA is akin to finding the most likely sentence among similar sounding alternatives. Our algorithm, named LinearDesign, takes only 11 minutes for the Spike protein, and can jointly optimize stability and codon usage. Experimentally, without chemical modification, our designs substantially improve mRNA half-life and protein expression in vitro, and dramatically increase antibody response by up to 23$\times$ in vivo, compared to the codon-optimized benchmark. Our work enables the exploration of highly stable and efficient designs that are previously unreachable and is a timely tool not only for vaccines but also for mRNA medicine encoding all therapeutic proteins (e.g., monoclonal antibodies and anti-cancer drugs).

q-bio.BM↗

arXiv2022

The evolution of the corona in MAXI J1535-571 through type-C quasi-periodic oscillations with Insight-HXMT

Type-C quasi-periodic oscillations (QPOs) in black hole X-ray transients can appear when the source is in the low-hard and hard-intermediate states. The spectral-timing evolution of the type-C QPO in MAXI J1535-571 has been recently studied with Insight-HXMT. Here we fit simultaneously the time-averaged energy spectrum, using a relativistic reflection model, and the fractional rms and phase-lag spectra of the type-C QPOs, using a recently developed time-dependent Comptonization model when the source was in the intermediate state. We show, for the first time, that the time-dependent Comptonization model can successfully explain the X-ray data up to 100 keV. We find that in the hard-intermediate state the frequency of the type-C QPO decreases from 2.6 Hz to 2.1 Hz, then increases to 3.3 Hz, and finally increases to ~ 9 Hz. Simultaneously with this, the evolution of corona size and the feedback fraction (the fraction of photons up-scattered in the corona that return to the disc) indicates the change of the morphology of the corona. Comparing with contemporaneous radio observations, this evolution suggests a possible connection between the corona and the jet when the system is in the hard-intermediate state and about to transit into the soft-intermediate state.

astro-ph.HE↗

arXiv2022

Coupling between the accreting corona and the relativistic jet in the micro quasar GRS 1915+105

GRS 1915+105 was the first stellar-mass black-hole in our Galaxy to display a superluminal radio jet, similar to those observed in active galactic nuclei with a supermassive black hole at the centre. It has been proposed that the radio emission in GRS 1915+105 is fed by instabilities in the accretion disc by which the inner parts of the accretion flow is ejected in the jet. Here we show that there is a significant correlation between: (i) the radio flux, coming from the jet, and the flux of the iron emission line, coming from the disc and, (ii) the temperature of the corona that produces the high-energy part of the X-ray spectrum via inverse Compton scattering and the amplitude of a high-frequency variability component coming from the innermost part of the accretion flow. At the same time, the radio flux and the flux of the iron line are strongly anti-correlated with the temperature of the X-ray corona and the amplitude of the high-frequency variability component. These correlations persist over ~10 years, despite the highly variable X-ray and radio properties of the source in that period. Our findings provide, for the first time, incontrovertible evidence that the energy that powers this black-hole system can be directed either to the X-ray corona or the jet. When this energy is used to power the corona, raising its temperature, there is less energy left to fuel the jet and the radio flux drops, and vice versa. These facts, plus the modelling of the variability in this source show conclusively that in GRS 1915+105 the X-ray corona morphs into the jet.

astro-ph.HE↗

arXiv2022

Radiation hardness study on a CMOS pixel sensor for charged particle tracking

A CMOS pixel sensor, named Supix-1, is developed for a pixelated silicon tracker for the Circular Electron-Positron Collider (CEPC) project. The sensor, consisted of nine sectors varying in pixel sizes, diode sizes and geometries, is fabricated with a 180 nm CMOS Image Sensor (CIS) process to study the particle detection performance of enlarged pixels. In this work, the radiation-induced effects on the charge collection of the sensor under the fluence of 1 $\times$ 10^13 1 MeV neq/cm^2 are studied by the measurements with the radioactive source of Fe-55 and the Technology Computer Aided Design (TCAD) simulations, since the radiation hardness of 6.8 $\times$ 10^12 1 MeV neq/cm^2 per year for Non-Ionizing Energy Loss (NIEL) effects is required. In measurements, the sensor gain has been calibrated using the k-$α$ peak of Fe-55 before and after irradiation. The pixel-wise equivalent noise charge (ENC), charge collection efficiency (CCE) and signal-to-noise ratio (SNR) were evaluated. The radiation-induced effects on cluster properties are studied through a self-developed reconstruction algorithm. In TCAD simulations, charge collections in 5 $\times$ 5 pixel matrixes for two typical impinging cases of incident particles were simulated with and without irradiation. Both measurements and simulations indicate that enlarged pixels with area of 21 $μ$m $\times$ 84 $μ$m, though suffering greater loss on sensor performance than small pixels do, still have satisfactory noise and charge collection performance after irradiation for particle tracking in the upcoming collider detectors.

physics.ins-det↗