arXiv subjects

Meng Sun

Publications and source records attributed to Meng Sun.

At least 91 records · Page 5Linked to original sources

arXiv2022

Extracting Weighted Finite Automata from Recurrent Neural Networks for Natural Languages

Recurrent Neural Networks (RNNs) have achieved tremendous success in sequential data processing. However, it is quite challenging to interpret and verify RNNs' behaviors directly. To this end, many efforts have been made to extract finite automata from RNNs. Existing approaches such as exact learning are effective in extracting finite-state models to characterize the state dynamics of RNNs for formal languages, but are limited in the scalability to process natural languages. Compositional approaches that are scablable to natural languages fall short in extraction precision. In this paper, we identify the transition sparsity problem that heavily impacts the extraction precision. To address this problem, we propose a transition rule extraction approach, which is scalable to natural language processing models and effective in improving extraction precision. Specifically, we propose an empirical method to complement the missing rules in the transition diagram. In addition, we further adjust the transition matrices to enhance the context-aware ability of the extracted weighted finite automaton (WFA). Finally, we propose two data augmentation tactics to track more dynamic behaviors of the target RNN. Experiments on two popular natural language datasets show that our method can extract WFA from RNN for natural language processing with better precision than existing approaches. Our code is available at https://github.com/weizeming/Extract_WFA_from_RNN_for_NL.

cs.CL↗

arXiv2022

A New Window to Tidal Asteroseismology: Non-linearly Excited Stellar Eigenmodes and the Period Spacing Pattern in KOI-54

We revisit the Tidally Excited Oscillations (TEOs) in the A-type, main-sequence, eccentric binary KOI-54, the prototype of heartbeat stars. Although the linear tidal response of the star is a series of orbital-harmonic frequencies which are not stellar eigenfrequencies, we show that the non-linearly excited non-orbital-harmonic TEOs are eigenmodes. By carefully choosing the modes which satisfy the mode-coupling selection rules, a period spacing ($ΔP$) pattern of quadrupole gravity modes ($ΔP \approx 2520-2535$ sec) can be discerned in the Fourier spectrum, with a detection significance level of $99.9\%$. The inferred period spacing value agrees remarkably well with the theoretical $l=2,m=0$ g modes from a stellar model with the measured mass, radius and effective temperature. We also find that the two largest-amplitude TEOs at $N=90, 91$ harmonics are very close to resonance with $l=2,m=0$ eigenmodes, and likely come from different stars. Previous works on tidal oscillations primarily focus on the modeling of TEO amplitudes and phases, the high sensitivity of TEO amplitude to the frequency detuning (tidal forcing frequency minus the closest stellar eigenfrequency) requires extremely dense grids of stellar models and prevents us from constraining the stellar physical parameters easily. This work, however, opens the window of real tidal asteroseismology by using the eigenfrequencies of the star inferred from the non-linear TEOs and possibly very-close-to-resonance linear TEOs. Our seismic modeling of these identified eigen g-modes shows that the best-matching stellar models have ($M \approx 2.20, 2.35 M_{\odot}$) and super-solar metallicity, in good agreement with previous measurements.

astro-ph.SR↗

arXiv2022

POSYDON: A General-Purpose Population Synthesis Code with Detailed Binary-Evolution Simulations

Most massive stars are members of a binary or a higher-order stellar systems, where the presence of a binary companion can decisively alter their evolution via binary interactions. Interacting binaries are also important astrophysical laboratories for the study of compact objects. Binary population synthesis studies have been used extensively over the last two decades to interpret observations of compact-object binaries and to decipher the physical processes that lead to their formation. Here, we present POSYDON, a novel, binary population synthesis code that incorporates full stellar-structure and binary-evolution modeling, using the MESA code, throughout the whole evolution of the binaries. The use of POSYDON enables the self-consistent treatment of physical processes in stellar and binary evolution, including: realistic mass-transfer calculations and assessment of stability, internal angular-momentum transport and tides, stellar core sizes, mass-transfer rates and orbital periods. This paper describes the detailed methodology and implementation of POSYDON, including the assumed physics of stellar- and binary-evolution, the extensive grids of detailed single- and binary-star models, the post-processing, classification and interpolation methods we developed for use with the grids, and the treatment of evolutionary phases that are not based on pre-calculated grids. The first version of POSYDON targets binaries with massive primary stars (potential progenitors of neutron stars or black holes) at solar metallicity.

astro-ph.SR↗

arXiv2022

Bogolon-mediated light absorption in atomic condensates of different dimensionality

In the case of structureless bosons, cooled down to low temperatures, the absorption of electromagnetic waves by their Bose-Einstein condensate is usually forbidden due to the momentum and energy conservation laws: the phase velocity of the collective modes of the condensate called bogolons is sufficiently lower than the speed of light. Thus, only the light scattering processes persist. However, the situation might be different in the case of composite bosons or the bosons with an internal structure. Here, we develop a microscopic theory of electromagnetic power absorption by a Bose-Einstein condensates of cold atoms in various dimensions, utilizing the Bogoliubov model of a weakly-interacting Bose gas. Thus, we address the transitions between a collective coherent state of bosons and the discrete energy levels corresponding to excited internal degrees of freedom of non-condensed individual bosons. It is shown, that such transitions are mediated by one and two-bogolon excitations above the condensate, which demonstrate different efficiency at different frequencies and strongly depend on the condensate density, which influence varies depending on the dimensionality of the system.

cond-mat.quant-gas↗

arXiv2022

A New Adjacency Matrix Configuration in GCN-based Models for Skeleton-based Action Recognition

Human skeleton data has received increasing attention in action recognition due to its background robustness and high efficiency. In skeleton-based action recognition, graph convolutional network (GCN) has become the mainstream method. This paper analyzes the fundamental factor for GCN-based models -- the adjacency matrix. We notice that most GCN-based methods conduct their adjacency matrix based on the human natural skeleton structure. Based on our former work and analysis, we propose that the human natural skeleton structure adjacency matrix is not proper for skeleton-based action recognition. We propose a new adjacency matrix that abandons all rigid neighbor connections but lets the model adaptively learn the relationships of joints. We conduct extensive experiments and analysis with a validation model on two skeleton-based action recognition datasets (NTURGBD60 and FineGYM). Comprehensive experimental results and analysis reveals that 1) the most widely used human natural skeleton structure adjacency matrix is unsuitable in skeleton-based action recognition; 2) The proposed adjacency matrix is superior in model performance, noise robustness and transferability.

cs.CV↗

arXiv2022

TESS Giants Transiting Giants II: The hottest Jupiters orbiting evolved stars

Giant planets on short-period orbits are predicted to be inflated and eventually engulfed by their host stars. However, the detailed timescales and stages of these processes are not well known. Here we present the discovery of three hot Jupiters (P $<$ 10 d) orbiting evolved, intermediate-mass stars ($M_\star$ $\approx$ 1.5 M$_\odot$, 2 R$_\odot$ $<$ $R_\star < $ 5 R$_\odot$). By combining \tess photometry with ground-based photometry and radial velocity measurements, we report masses and radii for these three planets between 0.4 and 1.8 M$_\mathrm{J}$ and 0.8 and 1.8 R$_\mathrm{J}$. \planet has the shortest period (P=\period) of any planet discovered around a red giant star to date. Both \planettwo and \planetthree appear to be inflated, but \planet does not show any sign of inflation. The large radii and relatively low masses of \planettwo and \planetthree place them among the lowest density hot Jupiters currently known, while \planet is conversely one of the highest. All three planets have orbital eccentricities below 0.2. The large spread in radii for these systems implies that planet inflation has a complex dependence on planet mass, radius, incident flux, and orbital properties. We predict that \planet has the shortest orbital decay timescale of any planet currently known, but do not detect any orbital decay in this system. Transmission spectroscopy of \planettwo would provide a favorable opportunity for the detection of water, carbon dioxide and carbon monoxide features in the atmosphere of a planet orbiting an evolved star, and could yield new information about planet formation and atmospheric evolution.

astro-ph.EP↗

arXiv2021

A Multi-Resolution Front-End for End-to-End Speech Anti-Spoofing

The choice of an optimal time-frequency resolution is usually a difficult but important step in tasks involving speech signal classification, e.g., speech anti-spoofing. The variations of the performance with different choices of timefrequency resolutions can be as large as those with different model architectures, which makes it difficult to judge what the improvement actually comes from when a new network architecture is invented and introduced as the classifier. In this paper, we propose a multi-resolution front-end for feature extraction in an end-to-end classification framework. Optimal weighted combinations of multiple time-frequency resolutions will be learned automatically given the objective of a classification task. Features extracted with different time-frequency resolutions are weighted and concatenated as inputs to the successive networks, where the weights are predicted by a learnable neural network inspired by the weighting block in squeeze-and-excitation networks (SENet). Furthermore, the refinement of the chosen timefrequency resolutions is investigated by pruning the ones with relatively low importance, which reduces the complexity and size of the model. The proposed method is evaluated on the tasks of speech anti-spoofing in ASVSpoof 2019 and its superiority has been justified by comparing with similar baselines.

cs.SD↗

arXiv2021

2M17091769+3127589: a mass-transfer binary with an extreme mass ratio

We present the orbital solution of a peculiar double-lined spectroscopic and eclipsing binary system, 2M17091769+3127589. This solution was obtained by a simultaneous fit of both APOGEE radial velocities and TESS and ASAS-SN light curves to determine masses and radii. This system consists of an $M=0.256^{+0.010}_{-0.006}$ $M_\odot$, $R=3.961^{+0.049}_{-0.032}$ $R_{\odot}$ red giant and a hotter $M=1.518 ^{+0.057}_{-0.031}$ $M_\odot$, $R=2.608^{+0.034}_{-0.321}$ $R_{\odot}$ subgiant. Modelling with the MESA evolutionary codes indicates that the system likely formed 5.26 Gyrs ago, with a $M=1.2$ $M_\odot$ primary that is now the system's red giant and a $M=1.11$ $M_\odot$ secondary that is now a more massive subgiant. Due to Roche-lobe overflow as the primary ascends the red giant branch, the more evolved "primary" (i.e., originally the more massive star of the pair) is now only one-sixth as massive as the "secondary". Such a difference between the initial and the current mass ratio is one of the most extreme detected so far. Evolutionary modelling suggests the system is still engaged in mass transfer, at a rate of $\dot{M} \sim 10^{-9}$ $M_\odot$ yr$^{-1}$, and it provides an example of a less evolved precursor to some of the systems that consist of white dwarfs and blue stragglers.

astro-ph.SR↗

arXiv2021

Theory of BCS-like bogolon-mediated superconductivity in transition metal dichalcogenides

We report on a novel mechanism of BCS-like superconductivity, mediated by a pair of Bogoliubov quasiparticles (bogolons). It takes place in hybrid systems consisting of a two-dimensional electron gas in a transition metal dichalcogenide monolayer in the vicinity of a Bose-Einstein condensate. Taking a system of two-dimensional indirect excitons as a testing ground we show, that the bogolon-pair-mediated electron pairing mechanism is stronger than phonon-mediated and single bogolon-mediated ones. We develop a microscopic theory of bogolon-pair-mediated superconductivity, based on the Schrieffer-Wolff transformation and the Gor'kov's equations, study the temperature dependence of the superconducting gap and estimate the critical temperature of superconducting transition for various concentrations of the electron gas and the condensate densities.

cond-mat.supr-con↗

arXiv2021

Strong-coupling theory of condensate-mediated superconductivity in 2D materials

We develop a strong-coupling theory of Bose-Einstein condensate-mediated superconductivity in a hybrid system, which consists of a two-dimensional electron gas with either (i) parabolic spectrum or (ii) relativistic Dirac spectrum in the vicinity of a two-dimensional solid-state condensate of indirect excitons. The Eliashberg equations are derived and the expressions for the electron pairing self-energy due to the exchange interaction between electrons mediated by a single Bogoliubov excitation (a bogolon) and the bogolon pairs are found. Furthermore, we find the superconducting order parameter and estimate the critical temperature of the superconducting transition. The critical temperature reveals its linear dependence on the dimensionless coupling constant. It is shown, that the bogolon-pair-mediated interaction is the dominant mechanism of electron pairing in hybrid systems in both the weak and strong coupling regimes. We calculate the effective bogolon-electron interaction constant for both parabolic and linear electron dispersions and examine the dependence of the critical temperature of electron gas superconducting transition on exciton condensate density.

cond-mat.supr-con↗

arXiv2020

WOCS 5379: Detailed Analysis of the Evolution of a Post-Mass-Transfer Blue Straggler

The blue-straggler binary WOCS 5379 is a member of the old (6-7 Gyr) open cluster NGC 188. WOCS 5379 comprises a blue straggler star with a white dwarf companion in a 120-day eccentric orbit. Combined with the orbital period, this helium white dwarf is evidence of previous mass transfer by a red giant. Detailed models of the system evolution from a progenitor main-sequence binary, including mass transfer, are made using the Modules for Experiments in Stellar Astrophysics (MESA). Both of the progenitor stars are evolved in the simulation. WOCS 5379 is well reproduced with a primary star of initial mass 1.19 $M_{\odot}$, whose core becomes the white dwarf. The secondary star initially is 1.01 $M_{\odot}$. 300 Myr ago, the secondary finished receiving mass from the donor, having moved beyond the NGC 188 turnoff as a 1.20 $M_{\odot}$ blue straggler. The successful model has a mass transfer efficiency of 22\%. This non-conservative mass transfer is key to expanding the orbit fast enough to permit stable mass transfer. Even so, the mass transfer begins with a short unstable phase, during which half of the accreted mass is transferred. With increasing mass, the secondary evolves from a radiative core to a convective core. The final blue straggler interior is remarkably similar to a 2.1 Gyr-old 1.21 $M_{\odot}$ main-sequence star at the same location in the HR diagram. The white dwarf effective temperature is also reproduced, but the modeled white dwarf mass of 0.33 $M_{\odot}$ is smaller than the measured mass of 0.42 $M_{\odot}$.

astro-ph.SR↗

arXiv2020

Exciton-Polaritons in Artificial Lattices and Electron Transport in Bose-Fermi Hybrid Systems

In this thesis, we study two different aspects of many-particle physics. In the first part, we study the Bose-Einstein condensation of microcavity exciton-polaritons in different artificial lattices. Bose-Einstein condensation is a quantum phase transition, which allows the system to macroscopically occupy its ground state and develop coherence spontaneously. Often studied in microcavities, which are optical cavities that trap light at specific wavelengths, exciton-polaritons are a kind of quasiparticle arising from the strong coupling between quantum well excitons and cavity photons. By periodically aligning cavity pillars in different patterns, one can achieve different artificial lattice structures. With this setup, we apply the driven-dissipative Gross-Pitaevskii equations to investigate the different consequences of the condensation by changing the pumping schemes and the design of the trapping potentials. Topics include multivalley condensation, phase selection and intermittency of exciton-polariton condensation, flat band condensation, and exciton-polariton topological insulators. In the second part of this thesis, we focus on the electron-scattering properties of a hybrid Bose-Fermi system. We consider a system consisting of a spatially separated two-dimensional electron gas layer and an exciton gas layer that interacts via Coulomb forces. We study the temperature dependence of the system's resistivity with this interlayer electron-exciton interaction and compare the results with the electron-phonon interaction.

cond-mat.mes-hall↗

arXiv2020

When Automatic Voice Disguise Meets Automatic Speaker Verification

The technique of transforming voices in order to hide the real identity of a speaker is called voice disguise, among which automatic voice disguise (AVD) by modifying the spectral and temporal characteristics of voices with miscellaneous algorithms are easily conducted with softwares accessible to the public. AVD has posed great threat to both human listening and automatic speaker verification (ASV). In this paper, we have found that ASV is not only a victim of AVD but could be a tool to beat some simple types of AVD. Firstly, three types of AVD, pitch scaling, vocal tract length normalization (VTLN) and voice conversion (VC), are introduced as representative methods. State-of-the-art ASV methods are subsequently utilized to objectively evaluate the impact of AVD on ASV by equal error rates (EER). Moreover, an approach to restore disguised voice to its original version is proposed by minimizing a function of ASV scores w.r.t. restoration parameters. Experiments are then conducted on disguised voices from Voxceleb, a dataset recorded in real-world noisy scenario. The results have shown that, for the voice disguise by pitch scaling, the proposed approach obtains an EER around 7% comparing to the 30% EER of a recently proposed baseline using the ratio of fundamental frequencies. The proposed approach generalizes well to restore the disguise with nonlinear frequency warping in VTLN by reducing its EER from 34.3% to 18.5%. However, it is difficult to restore the source speakers in VC by our approach, where more complex forms of restoration functions or other paralinguistic cues might be necessary to restore the nonlinear transform in VC. Finally, contrastive visualization on ASV features with and without restoration illustrate the role of the proposed approach in an intuitive way.

eess.AS↗

arXiv2020

Global Robustness Verification Networks

The wide deployment of deep neural networks, though achieving great success in many domains, has severe safety and reliability concerns. Existing adversarial attack generation and automatic verification techniques cannot formally verify whether a network is globally robust, i.e., the absence or not of adversarial examples in the input space. To address this problem, we develop a global robustness verification framework with three components: 1) a novel rule-based ``back-propagation'' finding which input region is responsible for the class assignment by logic reasoning; 2) a new network architecture Sliding Door Network (SDN) enabling feasible rule-based ``back-propagation''; 3) a region-based global robustness verification (RGRV) approach. Moreover, we demonstrate the effectiveness of our approach on both synthetic and real datasets.

cs.LG↗

arXiv2020

Coherent topological polariton laser

Topological concepts have been applied to a wide range of fields in order to successfully describe the emergence of robust edge modes that are unaffected by scattering or disorder. In photonics, indications of lasing from topologically protected modes with improved overall laser characteristics were observed. Here, we study exciton-polariton microcavity traps that are arranged in a one-dimensional Su-Schrieffer-Heeger lattice and form a topological defect mode from which we unequivocally observe highly coherent polariton lasing. Additionally, we confirm the excitonic contribution to the polariton lasing by applying an external magnetic field. These systematic experimental findings of robust lasing and high temporal coherence are meticulously reproduced by a combination of a generalized Gross-Pitaevskii model and a Lindblad master equation model. Thus, by using the comparatively simple SSH geometry, we are able to describe and control the exciton-polariton topological lasing, allowing for a deeper understanding of topological effects on microlasers.

cond-mat.mes-hall↗

arXiv2020

Towards Characterizing Adversarial Defects of Deep Learning Software from the Lens of Uncertainty

Over the past decade, deep learning (DL) has been successfully applied to many industrial domain-specific tasks. However, the current state-of-the-art DL software still suffers from quality issues, which raises great concern especially in the context of safety- and security-critical scenarios. Adversarial examples (AEs) represent a typical and important type of defects needed to be urgently addressed, on which a DL software makes incorrect decisions. Such defects occur through either intentional attack or physical-world noise perceived by input sensors, potentially hindering further industry deployment. The intrinsic uncertainty nature of deep learning decisions can be a fundamental reason for its incorrect behavior. Although some testing, adversarial attack and defense techniques have been recently proposed, it still lacks a systematic study to uncover the relationship between AEs and DL uncertainty. In this paper, we conduct a large-scale study towards bridging this gap. We first investigate the capability of multiple uncertainty metrics in differentiating benign examples (BEs) and AEs, which enables to characterize the uncertainty patterns of input data. Then, we identify and categorize the uncertainty patterns of BEs and AEs, and find that while BEs and AEs generated by existing methods do follow common uncertainty patterns, some other uncertainty patterns are largely missed. Based on this, we propose an automated testing technique to generate multiple types of uncommon AEs and BEs that are largely missed by existing techniques. Our further evaluation reveals that the uncommon data generated by our method is hard to be defended by the existing defense techniques with the average defense success rate reduced by 35\%. Our results call for attention and necessity to generate more diverse data for evaluating quality assurance solutions of DL software.

cs.SE↗

arXiv2020

Partial quantum revivals of localized condensates in distorted lattices

We report on a peculiar propagation of bosons loaded by a short Laguerre-Gaussian pulse in a nearly flat band of a lattice potential. Taking a system of exciton-polaritons in a kagome lattice as an example, we show that an initially localized condensate propagates in a specific direction in space if anisotropy is taken into account. This propagation consists of quantum jumps, collapses, and revivals of the whole compact states, and it persists given any direction of anisotropy. This property reveals its signatures in the tight-binding model and, surprisingly, it is much more pronounced in a continuous model. Quantum revivals are robust to the repulsive interaction and finite lifetime of the particles. Since no magnetic field or spin-orbit interaction is required, this system provides a new kind of easily implementable optical logic.

cond-mat.mes-hall↗

arXiv2019

A Survey on Theorem Provers in Formal Methods

Mechanical reasoning is a key area of research that lies at the crossroads of mathematical logic and artificial intelligence. The main aim to develop mechanical reasoning systems (also known as theorem provers) was to enable mathematicians to prove theorems by computer programs. However, these tools evolved with time and now play vital role in the modeling and reasoning about complex and large-scale systems, especially safety-critical systems. Technically, mathematical formalisms and automated reasoning based-approaches are employed to perform inferences and to generate proofs in theorem provers. In literature, there is a shortage of comprehensive documents that can provide proper guidance about the preferences of theorem provers with respect to their designs, performances, logical frameworks, strengths, differences and their application areas. In this work, more than 40 theorem provers are studied in detail and compared to present a comprehensive analysis and evaluation of these tools. Theorem provers are investigated based on various parameters, which includes: implementation architecture, logic and calculus used, library support, level of automation, programming paradigm, programming language, differences and application areas.

cs.SE↗