Search

Absorption is a crucial parameter in shaping wave propagation dynamics, yet achieving ultra-broadband absorption remains highly challenging, particularly in balancing low-frequency and broad bandwidth. Here, we present a metamaterial absorber (MMA) capable of achieving simultaneous spectral coverage across a seven-octave range of near-perfect absorption from 100 Hz to 12,800 Hz by engineering the quality-factor-weighted (Q-weighted) mode density. The Q-weighted mode density considers mode density, resonant frequencies, radiative loss, and intrinsic loss of multiple resonant modes, providing a comprehensive approach to govern broadband absorption properties. By optimizing the number of resonant modes and managing intrinsic losses, our approach achieves an intensive Q-weighted mode density across an ultra-wide bandwidth, enabling ultra-broadband absorption with high efficiency. These findings significantly advance the bandwidth capabilities of state-of-the-art MMAs and pave the way for the development of ultra-broadband metamaterial devices across various wave systems.

Text-to-image diffusion has attracted vast attention due to its impressive image-generation capabilities. However, when it comes to human-centric text-to-image generation, particularly in the context of faces and hands, the results often fall short of naturalness due to insufficient training priors. We alleviate the issue in this work from two perspectives. 1) From the data aspect, we carefully collect a human-centric dataset comprising over one million high-quality human-in-the-scene images and two specific sets of close-up images of faces and hands. These datasets collectively provide a rich prior knowledge base to enhance the human-centric image generation capabilities of the diffusion model. 2) On the methodological front, we propose a simple yet effective method called Mixture of Low-rank Experts (MoLE) by considering low-rank modules trained on close-up hand and face images respectively as experts. This concept draws inspiration from our observation of low-rank refinement, where a low-rank module trained by a customized close-up dataset has the potential to enhance the corresponding image part when applied at an appropriate scale. To validate the superiority of MoLE in the context of human-centric image generation compared to state-of-the-art, we construct two benchmarks and perform evaluations with diverse metrics and human studies. Datasets, model, and code are released at https://sites.google.com/view/mole4diffuser/., Comment: Published at NeurIPS 2024

We identify the leading term in the asymptotics of the quadratic Wasserstein distance between the invariant measure and empirical measures for diffusion processes on closed weighted four-dimensional Riemannian manifolds. Unlike results in lower dimensions, our analysis shows that this term depends solely on the Riemannian volume of the manifold, remaining unaffected by the potential and vector field in the diffusion generator.

Utility-scale off-grid renewable power-to-hydrogen systems (OReP2HSs) typically include photovoltaic plants, wind turbines, electrolyzers (ELs), and energy storage systems. As an island system, OReP2HS requires at least one component, generally the battery energy storage system (BESS), that operates for grid-forming control to provide frequency and voltage references and regulate them through transient power support and short-term energy balance regulation. While larger BESS capacity increases this ability, it also raises investment costs. This paper proposes a framework of layered multi-timescale energy management system (EMS) and evaluates the most cost-effective size of the grid-forming BESS in the OReP2HS. The proposed EMS covers the timescales ranging from those for power system transient behaviors to intra-day scheduling, coordinating renewable power, BESS, and ELs. Then, an iterative search procedure based on high-fidelity simulation is employed to determine the size of the BESS with minimal levelized cost of hydrogen (LCOH). Simulations over a reference year, based on the data from a planned OReP2HS project in Inner Mongolia, China, show that with the proposed EMS, the base-case optimal LCOH is 33.212 CNY/kg (4.581 USD/kg). The capital expenditure of the BESS accounts for 17.83% of the total, and the optimal BESS size accounts for 13.6% of the rated hourly energy output of power sources. Sensitivity analysis reveals that by reducing the electrolytic load adjustment time step from 90 to 5 s and increasing its ramping limit from 1% to 10% rated power per second, the BESS size decreases by 53.57%, and the LCOH decreases to 25.458 CNY/kg (3.511 USD/kg). Considering the cost of designing and manufacturing utility-scale ELs with fast load regulation capability, a load adjustment time step of 5-10 s and a ramping limit of 4-6% rated power per second are recommended.

For a class of (non-symmetric) diffusion processes on a length space, which in particular include the (reflecting) diffusion processes on a connected compact Riemannian manifold, the exact convergence rate is derived for $({\mathbb E} [{\mathbb W}_p^q(\mu_T,\mu)])^{\frac{1}{q}} (T \to \infty)$ uniformly in $(p,q)\in [1,\infty) \times (0,\infty)$, where $\mu_T$ is the empirical measure of the diffusion process, $\mu$ is the unique invariant probability measure, and ${\mathbb W}_p$ is the $p$-Wasserstein distance. Moreover, when the dimension parameter is less than $4$, we prove that ${\mathbb E} |T {\mathbb W}_2^2(\mu_T,\mu)-\Xi(T)|^q \to 0$ as $T\to\infty$ for any $q\ge 1$, where $\Xi(T)$ is explicitly given by eigenvalues and eigenfunctions for the symmetric part of the generator.

Off-grid renewable power to ammonia (ReP2A) systems present a promising pathway toward carbon neutrality in both the energy and chemical industries. However, due to chemical safety requirements, the limited flexibility of ammonia synthesis poses a challenge when attempting to align with the variable hydrogen flow produced from renewable power. This necessitates the optimal sizing of equipment capacity for effective and coordinated production across the system. Additionally, an ReP2A system may involve multiple stakeholders with varying degrees of operational flexibility, complicating the planning problem. This paper first examines the multistakeholder sizing equilibrium (MSSE) of the ReP2A system. First, we propose an MSSE model that accounts for individual planning decisions and the competing economic interests of the stakeholders of power generation, hydrogen production, and ammonia synthesis. We then construct an equivalent optimization problem based on Karush--Kuhn--Tucker (KKT) conditions to determine the equilibrium. Following this, we decompose the problem in the temporal dimension and solve it via multicut generalized Benders decomposition (GBD) to address long-term balancing issues. Case studies based on a realistic project reveal that the equilibrium does not naturally balance the interests of all stakeholders due to their heterogeneous characteristics. Our findings suggest that benefit transfer agreements ensure mutual benefits and the successful implementation of ReP2A projects.

We have implemented the 3+1 dimensional CLVisc hydrodynamics model with TRENTO-3D initial conditions to investigate the spin polarization of $\Lambda$ hyperons along the beam direction in p+Pb collisions at $\sqrt{s_{NN}} = 8.16$ TeV. Following our previous theoretical framework based on quantum kinetic theory, we consider three different scenarios: $\Lambda$ equilibrium, $s$ quark equilibrium, and iso-thermal equilibrium scenarios. We have computed the second Fourier sine coefficients of spin polarization along the beam direction, denoted as $\left\langle P_{z} \sin 2(\phi_{p} - \Psi_{2}) \right\rangle$, with $\phi_{p} - \Psi_{2}$ being the azimuthal angle relative to the second-order event plane $\Psi_{2}$, as functions of multiplicity, transverse momentum and pseudo-rapidity in the three scenarios. Additionally, we have also computed the spin polarization along the beam direction, $P_{z}$, as a function of the azimuthal angle. We find that the spin polarization induced by thermal vorticity always provides an opposite contribution compared to the shear-induced polarization in p+Pb collisions. The total spin polarization computed by the current hydrodynamic model disagrees with the data measured by LHC-CMS experiments. Our findings imply that other non-flow effects may play a crucial role in p+Pb collisions., Comment: 8 pages, 5 figures and 1 table. A new figure for local polarization as a function of pseudo-rapidity is added. submitted to PRC

In this paper, we shall show that the vector-scalar potential ($\mathbf{A}$-$\Phi$) formulation, for many problems, can be further simplified by ignoring the scalar potential contribution and setting it to zero.

In light of recent breakthroughs in large language models (LLMs) that have revolutionized natural language processing (NLP), there is an urgent need for new benchmarks to keep pace with the fast development of LLMs. In this paper, we propose CFLUE, the Chinese Financial Language Understanding Evaluation benchmark, designed to assess the capability of LLMs across various dimensions. Specifically, CFLUE provides datasets tailored for both knowledge assessment and application assessment. In knowledge assessment, it consists of 38K+ multiple-choice questions with associated solution explanations. These questions serve dual purposes: answer prediction and question reasoning. In application assessment, CFLUE features 16K+ test instances across distinct groups of NLP tasks such as text classification, machine translation, relation extraction, reading comprehension, and text generation. Upon CFLUE, we conduct a thorough evaluation of representative LLMs. The results reveal that only GPT-4 and GPT-4-turbo achieve an accuracy exceeding 60\% in answer prediction for knowledge assessment, suggesting that there is still substantial room for improvement in current LLMs. In application assessment, although GPT-4 and GPT-4-turbo are the top two performers, their considerable advantage over lightweight LLMs is noticeably diminished. The datasets and scripts associated with CFLUE are openly accessible at https://github.com/aliyun/cflue., Comment: Accepted by ACL 2024

The transformation of urban environments to accommodate growing populations has profoundly impacted public health and well-being. This paper addresses the critical challenge of estimating the impact of urban design interventions on diverse populations. Traditional approaches, reliant on questionnaires and stated preference techniques, are limited by recall bias and capturing the complex dynamics between environmental attributes and individual characteristics. To address these challenges, we integrate Virtual Reality (VR) with observational causal inference methods to estimate heterogeneous treatment effects, specifically employing Targeted Maximum Likelihood Estimation (TMLE) for its robustness against model misspecification. Our innovative approach leverages VR-based experiment to collect data that reflects perceptual and experiential factors. The result shows the heterogeneous impacts of urban design elements on public health and underscore the necessity for personalized urban design interventions. This study not only extends the application of TMLE to built environment research but also informs public health policy by illuminating the nuanced effects of urban design on mental well-being and advocating for tailored strategies that foster equitable, health-promoting urban spaces.

Self-supervised learning shows promise in harnessing extensive unlabeled data, but it also confronts significant privacy concerns, especially in vision. In this paper, we aim to perform membership inference on visual self-supervised models in a more realistic setting: self-supervised training method and details are unknown for an adversary when attacking as he usually faces a black-box system in practice. In this setting, considering that self-supervised model could be trained by completely different self-supervised paradigms, e.g., masked image modeling and contrastive learning, with complex training details, we propose a unified membership inference method called PartCrop. It is motivated by the shared part-aware capability among models and stronger part response on the training data. Specifically, PartCrop crops parts of objects in an image to query responses with the image in representation space. We conduct extensive attacks on self-supervised models with different training protocols and structures using three widely used image datasets. The results verify the effectiveness and generalization of PartCrop. Moreover, to defend against PartCrop, we evaluate two common approaches, i.e., early stop and differential privacy, and propose a tailored method called shrinking crop scale range. The defense experiments indicate that all of them are effective. Our code is available at https://github.com/JiePKU/PartCrop., Comment: Accepted by ACM CCS2024, Full version

We study the transverse momentum ($p_T$) differential anisotropic flow and flow fluctuation in Pb+Pb collisions at $\sqrt{s_{NN}}$=5.02 TeV at the LHC. A (3+1)-dimensional CLVisc hydrodynamics framework with fluctuating TRENTO (or AMPT) initial conditions is utilized to simulate the space-time evolution of the quark-gluon plasma (QGP) medium. The effects of shear viscosity and the sub-nucleon structure on anisotropic flow and flow fluctuation are analyzed. Our result shows that shear viscosity tends to suppress both flow coefficients (${v_2\{2}\}$, ${v_2\{4\}}$, ${\langle v_2\rangle}$) and flow fluctuation (${\sigma_{v_2}}$) due to its smearing effect on local density fluctuation. The flow coefficients appear to be insensitive to the sub-nucleon structure, whereas for flow fluctuation ${\sigma_{v_2}}$, it tends to be suppressed by the sub-nucleon structure in central collisions but enhanced in peripheral collisions. After taking into account the sub-nucleon structure effect, our numerical result can quantitatively describe the relative flow fluctuations (${v_2\{4\}/v_2\{2\}}$, $F({v_2})$) measured by the ALICE Collaboration at the LHC. We further investigate the effects of shear viscosity, sub-nucleon structure and initial condition model on the flow angle and flow magnitude decorrelations (${A_2^f}$, ${M_2^f}$) using the four-particle correlation method. We find that the flow decorrelation effect is typically stronger in central collisions than in peripheral collisions. The flow angle decorrelation is found to be insensitive to the shear viscosity and sub-nucleon structure, whereas the flow magnitude decorrelation shows quite different behavior when using TRENTO or AMPT initial condition model. Our study sheds light on the anisotropic flow, transport properties and initial structure of the QGP created in high-energy nuclear collisions., Comment: 11 pages, 6 figures

Two-stage pipeline is popular in speech enhancement tasks due to its superiority over traditional single-stage methods. The current two-stage approaches usually enhance the magnitude spectrum in the first stage, and further modify the complex spectrum to suppress the residual noise and recover the speech phase in the second stage. The above whole process is performed in the short-time Fourier transform (STFT) spectrum domain. In this paper, we re-implement the above second sub-process in the short-time discrete cosine transform (STDCT) spectrum domain. The reason is that we have found STDCT performs greater noise suppression capability than STFT. Additionally, the implicit phase of STDCT ensures simpler and more efficient phase recovery, which is challenging and computationally expensive in the STFT-based methods. Therefore, we propose a novel two-stage framework called the STFT-STDCT spectrum fusion network (FDFNet) for speech enhancement in cross-spectrum domain. Experimental results demonstrate that the proposed FDFNet outperforms the previous two-stage methods and also exhibits superior performance compared to other advanced systems., Comment: Accepted by ICASSP 2024

Hydrogen plays an important role in the context of global carbon reduction. For an off-grid renewable power to hydrogen system (OReP2HS), a grid-forming (GFM) source is essential to provide frequency and voltage references. Here, we take battery works as a GFM source, and the OReP2HS we focus on is comprised of solar photovoltaic, wind turbines, and alkaline electrolyzers for hydrogen generation. An elaborative full-time scale energy management strategy (EMS) covers GFM control to system scheduling (from milliseconds to hours) and is proposed to support high-precision production simulations. Loads of electrolysers are designed to track the renewable power closely to partly replace the energy balancing requirement of the battery. A continuous operation strategy during an emergency like a unit drop is also included in the presented EMS. An off-line simulation-based iterative searching algorithm is developed to find the most suitable size of the battery with the minimum levelized cost of hydrogen (LCOH). A practical OReP2H project located in Inner Mongolian, China is taken as a case study. Simulation results demonstrate the feasibility of the proposed method, and the optimized capacity of the battery (3.4MWh) only accounts for 13.6% of the total installed capacity of the sources with the proposed EMS. Sensitivity analysis shows that LOCH rises from 28.829 CYN/kg to 37.814 CYN/kg, with the time-step for fast power regulation of electrolysers changing from 4s to 1 min with a ramp rate of 0.05MW/s.

The size of deep learning models in artificial intelligence (AI) software is increasing rapidly, hindering the large-scale deployment on resource-restricted devices (e.g., smartphones). To mitigate this issue, AI software compression plays a crucial role, which aims to compress model size while keeping high performance. However, the intrinsic defects in a big model may be inherited by the compressed one. Such defects may be easily leveraged by adversaries, since a compressed model is usually deployed in a large number of devices without adequate protection. In this article, we aim to address the safe model compression problem from the perspective of safety-performance co-optimization. Specifically, inspired by the test-driven development (TDD) paradigm in software engineering, we propose a test-driven sparse training framework called SafeCompress. By simulating the attack mechanism as safety testing, SafeCompress can automatically compress a big model to a small one following the dynamic sparse training paradigm. Then, considering two kinds of representative and heterogeneous attack mechanisms, i.e., black-box membership inference attack and white-box membership inference attack, we develop two concrete instances called BMIA-SafeCompress and WMIA-SafeCompress. Further, we implement another instance called MMIA-SafeCompress by extending SafeCompress to defend against the occasion when adversaries conduct black-box and white-box membership inference attacks simultaneously. We conduct extensive experiments on five datasets for both computer vision and natural language processing tasks. The results show the effectiveness and generalizability of our framework. We also discuss how to adapt SafeCompress to other attacks besides membership inference attack, demonstrating the flexibility of SafeCompress., Comment: Accepted by IEEE Transactions on Software Engineering (TSE). Camera-ready Version. arXiv admin note: substantial text overlap with arXiv:2208.05969

Utility-scale renewable power-to-hydrogen (ReP2H) production typically uses thyristor rectifiers (TRs) to supply power to multiple electrolyzers (ELZs). They exhibit a nonlinear and non-decouplable relation between active and reactive power. The on-off scheduling and load allocation of multiple ELZs simultaneously impact energy conversion efficiency and AC-side active and reactive power flow. Improper scheduling may result in excessive reactive power demand, causing voltage violations and increased network losses, compromising safety and economy. To address these challenges, this paper first explores trade-offs between the efficiency and the reactive load of the electrolyzers. Subsequently, we propose a coordinated approach for scheduling the active and reactive power in the ReP2H system. A mixed-integer second-order cone programming (MISOCP) is established to jointly optimize active and reactive power by coordinating the ELZs, renewable energy sources, energy storage (ES), and var compensations. Case studies demonstrate that the proposed method reduces losses by 3.06% in an off-grid ReP2H system while increasing hydrogen production by 5.27% in average.

In this paper, we propose a low rank approximation method for efficiently solving stochastic partial differential equations. Specifically, our method utilizes a novel low rank approximation of the stiffness matrices, which can significantly reduce the computational load and storage requirements associated with matrix inversion without losing accuracy. To demonstrate the versatility and applicability of our method, we apply it to address two crucial uncertainty quantification problems: stochastic elliptic equations and optimal control problems governed by stochastic elliptic PDE constraints. Based on varying dimension reduction ratios, our algorithm exhibits the capability to yield a high precision numerical solution for stochastic partial differential equations, or provides a rough representation of the exact solutions as a pre-processing phase. Meanwhile, our algorithm for solving stochastic optimal control problems allows a diverse range of gradient-based unconstrained optimization methods, rendering it particularly appealing for computationally intensive large-scale problems. Numerical experiments are conducted and the results provide strong validation of the feasibility and effectiveness of our algorithm.

In speech enhancement (SE), phase estimation is important for perceptual quality, so many methods take clean speech's complex short-time Fourier transform (STFT) spectrum or the complex ideal ratio mask (cIRM) as the learning target. To predict these complex targets, the common solution is to design a complex neural network, or use a real network to separately predict the real and imaginary parts of the target. But in this paper, we propose to use a real network to estimate the magnitude mask and normalized cIRM, which not only avoids the significant increase of the model complexity caused by complex networks, but also shows better performance than previous phase estimation methods. Meanwhile, we devise a parallel sequence modeling (PSM) block to improve the RNN block in the convolutional recurrent network (CRN)-based SE model. We name our method as magnitude-and-phase-aware and PSM-based CRN (MPCRN). The experimental results illustrate that our MPCRN has superior SE performance., Comment: Accepted by ASRU 2023

The deep learning-based speech enhancement (SE) methods always take the clean speech's waveform or time-frequency spectrum feature as the learning target, and train the deep neural network (DNN) by reducing the error loss between the DNN's output and the target. This is a conventional single-task learning paradigm, which has been proven to be effective, but we find that the multi-task learning framework can improve SE performance. Specifically, we design a framework containing a SE module and a voice activity detection (VAD) module, both of which share the same encoder, and the whole network is optimized by the weighted loss of the two modules. Moreover, we design a causal spatial attention (CSA) block to promote the representation capability of DNN. Combining the VAD aided multi-task learning framework and CSA block, our SE network is named VSANet. The experimental results prove the benefits of multi-task learning and the CSA block, which give VSANet an excellent SE performance., Comment: Accepted by ASRU 2023

The High Energy Stereoscopic System (H.E.S.S) Collaboration reported the discovery of a novel radiation component from the Vela pulsar by their Cherenkov telescopes. It is of great importance that gamma rays with energies of at least 20~TeV are recorded unexpectedly. The H.E.S.S Collaboration argued that such results may challenge the state-of-the-art models for the high-energy emission of pulsars. We point out in this work that these results also provide a unique opportunity to constrain certain Lorentz invariance violation parameters, leading to the realization of studying Lorentz invariance violation by using gamma-ray pulsars. The Lorentz invariance violation scale is constrained at the level of $E_{\mathrm{LV,}1}> 1.66\times 10^{17} \rm GeV$ for the linear scenario, and $E_{\mathrm{LV,}2}>3.53\times 10^{10} \rm GeV$ for the quadratic scenario. We anticipate that digging into the detailed features of the data of the Vela pulsar and analyzing potentially more very-high-energy photon data from pulsars in the future would improve the constraints on Lorentz invariance violation., Comment: 3 pages, no figure

This paper focuses on the research of micro-expression recognition (MER) and proposes a flexible and reliable deep learning method called learning to rank onset-occurring-offset representations (LTR3O). The LTR3O method introduces a dynamic and reduced-size sequence structure known as 3O, which consists of onset, occurring, and offset frames, for representing micro-expressions (MEs). This structure facilitates the subsequent learning of ME-discriminative features. A noteworthy advantage of the 3O structure is its flexibility, as the occurring frame is randomly extracted from the original ME sequence without the need for accurate frame spotting methods. Based on the 3O structures, LTR3O generates multiple 3O representation candidates for each ME sample and incorporates well-designed modules to measure and calibrate their emotional expressiveness. This calibration process ensures that the distribution of these candidates aligns with that of macro-expressions (MaMs) over time. Consequently, the visibility of MEs can be implicitly enhanced, facilitating the reliable learning of more discriminative features for MER. Extensive experiments were conducted to evaluate the performance of LTR3O using three widely-used ME databases: CASME II, SMIC, and SAMM. The experimental results demonstrate the effectiveness and superior performance of LTR3O, particularly in terms of its flexibility and reliability, when compared to recent state-of-the-art MER methods.

The correlation of $\alpha\alpha$ was probed via measuring the transverse momentum $p_{T}$ and width $\delta p_{T}$ of one $\alpha$, for the first time, which represents the spatial and dynamical essentialities of the initial coupling state in $^{8}$Be nucleus. The weighted interaction vertex of 3$\alpha$ reflected by the magnitudes of their relative momentums and relative emission angles proves the isosceles triangle configuration for 3$\alpha$ at the high excited energy analogous Hoyle states., Comment: 8 pages, 9 figures

The enhancement of $J^{\pi}(T)$=3$^{+}$(0) state with isospin $T=0$ excited by the tensor force in the free $^{6}$Li nucleus has been observed, for the first time, relative to a shrinkable excitation in the $^{6}$Li cluster component inside its host nucleus. Comparatively, the excitation of $J^{\pi}(T)$=0$^{+}$(1) state with isospin $T=1$ for these two $^{6}$Li formations take on an approximately equal excitation strength. The mechanism of such tensor force effect was proposed due to the intensive nuclear medium role on isospin $T$=0 state., Comment: 6 pages, 4 figures

Let $(X_t)_{t \geq 0}$ be a diffusion process defined on a compact Riemannian manifold, and for $\alpha > 0$, let $$ \mu_t^{(\alpha)} = \frac{\alpha}{t^\alpha} \int_{0}^{t} \delta_{X_s} \, s^{\alpha - 1} \mathrm{d} s $$ be the associated weighted empirical measure. We investigate asymptotic behavior of $\mathbb{E}^\nu \big[ \mathrm{W}_2^2(\mu_t^{(\alpha)}, \mu) \big]$ for sufficient large $t$, where $\mathrm{W}_2$ is the quadratic Wasserstein distance and $\mu$ is the invariant measure of the process. In the particular case $\alpha = 1$, our result sharpens the limit theorem achieved in [26]. The proof is based on the PDE and mass transportation approach developed by L. Ambrosio, F. Stra and D. Trevisan.

This study investigates the causal relationship between campus walkability and the emotional walking experiences of students, with a focus on their mental well-being. Using data from 697 participants across seven Chinese tertiary education campuses, the study employs a counterfactual analysis to estimate the impact of campus walkability on students' walking experiences. The analysis reveals that students living in campuses with improved walkability are 9.75% more likely to have positive walking experiences compared to those without walkability renovations. While walking attitude is strongly correlated with walking experiences, the study emphasizes the significance of objective factors such as campus surroundings and the availability of walking spaces in influencing the walking experience. Geographical features, including campus walkability improvements, have the most substantial impact, and this effect varies across different subsets of respondents. These findings underscore the importance of considering specific subsets and geographical features when assessing the impact of walkability improvements on the walking experience. In conclusion, the study provides compelling evidence of a causal link between improved campus walkability and enhanced emotional walking experiences among students, suggesting the need for further research on mediating factors and cultural variations affecting student mental health on various Chinese campuses.

Despite significant progress in video question answering (VideoQA), existing methods fall short of questions that require causal/temporal reasoning across frames. This can be attributed to imprecise motion representations. We introduce Action Temporality Modeling (ATM) for temporality reasoning via three-fold uniqueness: (1) rethinking the optical flow and realizing that optical flow is effective in capturing the long horizon temporality reasoning; (2) training the visual-text embedding by contrastive learning in an action-centric manner, leading to better action representations in both vision and text modalities; and (3) preventing the model from answering the question given the shuffled video in the fine-tuning stage, to avoid spurious correlation between appearance and motion and hence ensure faithful temporality reasoning. In the experiments, we show that ATM outperforms previous approaches in terms of the accuracy on multiple VideoQAs and exhibits better true temporality reasoning ability.

Multimode fiber (MMF) has been proven to have good potential in imaging and optical communication because of its advantages of small diameter and large mode numbers. However, due to the mode coupling and modal dispersion, it is very sensitive to environmental changes. Minor changes in the fiber shape can lead to difficulties in information reconstruction. Here, white LED and cascaded Unet are used to achieve MMF imaging to eliminate the effect of fiber perturbations. The output speckle patterns in three different color channels of the CCD camera produced by transferring images through the MMF are concatenated and inputted into the cascaded Unet using channel stitching technology to improve the reconstruction effects. The average Pearson correlation coefficient (PCC) of the reconstructed images from the Fashion-MINIST dataset is 0.83. In order to check the flexibility of such a system, perturbation tests on the image reconstruction capability by changing the fiber shapes are conducted. The experimental results show that the MMF imaging system has good robustness properties, i. e. the average PCC remains 0.83 even after completely changing the shape of the MMF. This research potentially provides a flexible approach for the practical application of MMF imaging.

Robots assist humans in various activities, from daily living public service (e.g., airports and restaurants), and to collaborative manufacturing. However, it is risky to assume that the knowledge and strategies robots learned from one group of people can apply to other groups. The discriminatory performance of robots will undermine their service quality for some people, ignore their service requests, and even offend them. Therefore, it is critically important to mitigate bias in robot decision-making for more fair services. In this paper, we designed a self-reflective mechanism -- Fairness-Sensitive Policy Gradient Reinforcement Learning (FSPGRL), to help robots to self-identify biased behaviors during interactions with humans. FSPGRL identifies bias by examining the abnormal update along particular gradients and updates the policy network to support fair decision-making for robots. To validate FSPGRL's effectiveness, a human-centered service scenario, "A robot is serving people in a restaurant," was designed. A user study was conducted; 24 human subjects participated in generating 1,000 service demonstrations. Four commonly-seen issues "Willingness Issue," "Priority Issue," "Quality Issue," "Risk Issue" were observed from robot behaviors. By using FSPGRL to improve robot decisions, robots were proven to have a self-bias detection capability for a more fair service. We have achieved the suppression of bias and improved the quality during the process of robot learning to realize a relatively fair model.

This work reports the thermal conductivity of RbV3Sb5 and CsV3Sb5 with three-dimensional charge density wave phase transitions from 80 K to 400 K measured by pump-probe thermoreflectance techniques. The in-plane (basal plane) thermal conductivities are found moderate, i.e., 12 W/mK of RbV3Sb5 and 8.8 W/mK of CsV3Sb5 at 300 K. Low cross-plane (stacking direction) thermal conductivities are observed, with 0.72 W/mK of RbV3Sb5 and 0.49 W/mK of CsV3Sb5 at 300 K. A unique glass-like temperature dependence in the cross-plane thermal conductivity is observed, which decreases monotonically even lower than the Cahill-Pohl limit as the temperature decreases below the phase transition point TCDW. This temperature dependence is found to obey the hopping transport picture. In addition, a peak in cross-plane thermal conductivity is observed at TCDW as a fingerprint of the modulated structural distortion along the stacking direction.

Lorentz invariance is one of the foundations of modern physics; however, Lorentz violation may happen from the perspective of quantum gravity, and plenty of studies on Lorentz violation have arisen in recent years. As a good tool to explore Lorentz violation, Finsler geometry is a natural and fundamental generalization of Riemann geometry. The Finsler structure depends on both coordinates and velocities. Here, we simply introduce the mathematics of Finsler geometry. We review the connection between modified dispersion relations and Finsler geometries and discuss the physical influence from Finsler geometry. We review the connection between Finsler geometries and theories of Lorentz violation, such as the doubly special relativity, the standard-model extension, and the very special relativity., Comment: 29 pages, no figure, final version for journal publication

Finsler geometry is a natural and fundamental generalization of Riemann geometry, and is a tool to research Lorentz invariance violation. We find the connection between the most general modified dispersion relation and a pseudo-Finsler structure, and then we calculate the arrival time delay of astroparticles with different modified dispersion relations in the framework of Finsler geometry. The result suggests that the time delay is irrelevant with the exact form of the modified dispersion relation. If the modified term becomes 0 when $E=p$, there is no arrival time difference, otherwise the time delays only depend on the Lorentz violation scale and the order at which the Lorentz invariance breaks., Comment: 9 pages, no figure, version for journal publication. arXiv admin note: text overlap with arXiv:2206.07616

In this paper we are concerned with the Riesz transform on the direct product manifold ${\mathbb{H}}^n \times M$, where ${\mathbb{H}}^n$ is the $n$-dimensional real hyperbolic space and $M$ is a connected complete non-compact Riemannian manifold satisfying the volume doubling property and generalized Gaussian or sub-Gaussian upper estimates for the heat kernel. We establish its weak type $(1,1)$ property. In addition, we obtain the weak type $(1, 1)$ of the heat maximal operator in the same setting. Our arguments also work for a large class of direct product manifolds with exponential volume growth. Particularly, we provide a simpler proof of weak type $(1,1)$ boundedness of some operators considered in the work of Li, Sj\"ogren and Wu [27]., Comment: This paper has already been accepted by Science China Mathematics

Method names are crucial to program comprehension and maintenance. Recently, many approaches have been proposed to automatically recommend method names and detect inconsistent names. Despite promising, their results are still sub-optimal considering the three following drawbacks: 1) These models are mostly trained from scratch, learning two different objectives simultaneously. The misalignment between two objectives will negatively affect training efficiency and model performance. 2) The enclosing class context is not fully exploited, making it difficult to learn the abstract function of the method. 3) Current method name consistency checking methods follow a generate-then-compare process, which restricts the accuracy as they highly rely on the quality of generated names and face difficulty measuring the semantic consistency. In this paper, we propose an approach named AUMENA to AUtomate MEthod NAming tasks with context-aware prompt-tuning. Unlike existing deep learning based approaches, our model first learns the contextualized representation(i.e., class attributes) of PL and NL through the pre-training model, then fully exploits the capacity and knowledge of large language model with prompt-tuning to precisely detect inconsistent method names and recommend more accurate names. To better identify semantically consistent names, we model the method name consistency checking task as a two-class classification problem, avoiding the limitation of previous similarity-based consistency checking approaches. The experimental results reflect that AUMENA scores 68.6%, 72.0%, 73.6%, 84.7% on four datasets of method name recommendation, surpassing the state-of-the-art baseline by 8.5%, 18.4%, 11.0%, 12.0%, respectively. And our approach scores 80.8% accuracy on method name consistency checking, reaching an 5.5% outperformance. All data and trained models are publicly available., Comment: Accepted by ICPC-2023

The modified Langevin noise formalism has been proposed for the correct charaterization of quantum electromagnetic fields in the presence of finite-sized lossy dielectric objects in free space. The main modification to the original one (also known as the Green's function approach available only for bulk inhomogeneous lossy dielectric medium) was to add fluctuating sources in reaction to the radiation loss. Consequently, a resulting electric field operator is now determined by (i) boundary-assisted and (ii) medium-assisted fields on an equal footing, which are fluctuating sources due to radiation and medium losses, respectively. However, due to the lengthy mathematical manipulation and complicated concepts, the validity of the modified Langevin noise formalism has not been clearly checked yet. In this work, we propose and develop a novel numerical framework for the modified Langevin noise formalism by exploiting computational electromagnetic methods (CEM). Specifically, we utilize the finite-element method to numerically solve plane-wave-scattering and point-source-radiation problems whose solutions are boundary-assisted and medium-assisted fields, respectively. Based on the developed numerical framework, we calculate the Purcell factor of a two-level atom inside or outside a lossy dielectric slab. It is numerically proved, for the first time, that one can retrieve the conventional expression of the spontaneous emission rate, viz., the imaginary part of the Green's function. The proposed numerical framework is particularly useful for estimating the dynamics of multi-level atoms near practical plasmonic structures or metasurfaces., Comment: 14 pages, 8 figures, to be submitted to Physical Review A soon

In this paper, we are interested in understanding self-supervised pretraining through studying the capability that self-supervised representation pretraining methods learn part-aware representations. The study is mainly motivated by that random views, used in contrastive learning, and random masked (visible) patches, used in masked image modeling, are often about object parts. We explain that contrastive learning is a part-to-whole task: the projection layer hallucinates the whole object representation from the object part representation learned from the encoder, and that masked image modeling is a part-to-part task: the masked patches of the object are hallucinated from the visible patches. The explanation suggests that the self-supervised pretrained encoder is required to understand the object part. We empirically compare the off-the-shelf encoders pretrained with several representative methods on object-level recognition and part-level recognition. The results show that the fully-supervised model outperforms self-supervised models for object-level recognition, and most self-supervised contrastive learning and masked image modeling methods outperform the fully-supervised method for part-level recognition. It is observed that the combination of contrastive learning and masked image modeling further improves the performance., Comment: Accepted by TMLR

Potential-based formulation with generalized Lorenz gauge can be used in the quantization of electromagnetic fields in inhomogeneous media. However, one often faces the redundancy of modes when finding eigenmodes from potential-based formulation. In free space, this can be explained by the connection to the well-known Helmholtz decomposition. In this work, we generalize the Helmholtz decomposition to its generalized form, echoing the use of generalized Lorenz gauge in inhomogeneous media. We formulate electromagnetics eigenvalue problems using vector potential formulation which is often used in numerical quantization. The properties of the differential operators are mathematically analyzed. Orthogonality relations between the two classes of modes are proved in both continuous and discrete space. Completeness of two sets of modes and the orthogonality relations are numerically validated in inhomogeneous anisotropic media. This work serves as a foundation for numerical quantization of electromagnetic fields in inhomogeneous media with potential-based formulation.

The discovery of non-Hermitian skin effect (NHSE) has opened an exciting direction for unveiling unusual physics and phenomena in non-Hermitian system. Despite notable theoretical breakthroughs, actual observation of NHSE's whole evolvement, however, relies mainly on gain medium to provide amplified mode. It typically impedes the development of simple, robust system. Here, we show that a passive system is fully capable of supporting the observation of the complete evolution picture of NHSE, without the need of any gain medium. With a simple lattice model and acoustic ring resonators, we use complex-frequency excitation to create virtual gain effect, and experimentally demonstrate that exact NHSE can persist in a totally passive system during a quasi-stationary stage. This results in the transient NHSE: passive construction of NHSE in a short time window. Despite the general energy decay, the localization character of skin modes can still be clearly witnessed and successfully exploited. Our findings unveil the importance of excitation in realizing NHSE and paves the way towards studying the peculiar features of non-Hermitian physics with diverse passive platforms., Comment: 7 pages, 4 figures

It is postulated in Einstein's relativity that the speed of light in vacuum is a constant for all observers. However, the effect of quantum gravity could bring an energy dependence of light speed, and a series of studies on high-energy photon events from gamma-ray bursts (GRBs) and active galactic nuclei (AGNs) suggest a light speed variation $v(E)=c\left(1-E / E_{\mathrm{LV}}\right)$ with $E_{\mathrm{LV}}=3.6 \times 10^{17} ~\mathrm{GeV}$ or a bound $E_{\mathrm{LV}} \ge 3.6 \times 10^{17} ~\mathrm{GeV}$. From the newly observed gamma-ray burst GRB 221009A, we find that a 99.3~GeV photon detected by Fermi-LAT is coincident with the sharp spike in the light curves detected by both Fermi-GBM and HEBS under the above scenario of light speed variation, suggesting an option that this high-energy photon was emitted at the same time as a sharp spike of low-energy photon emission at the GRB source. Thus this highest energy photon event detected by Fermi-LAT during the prompt emission of gamma-ray bursts might be considered as an optional signal for the linear form modification of light speed in cosmological space., Comment: 5 latex pages, 2 figures, final version

In condensed matter physics, symmetry profoundly governs the fundamentals of topological matter. The emergence of new topological phase is typically linked to the enrichment of symmetries. Different parity-time symmetry relations distinguish between spinless and spinful physical systems. In spinless systems, creating pseudo-spins can realize fragile topological phase but not break the time-reversal symmetry. Therefore, growing attentions were recently focused on the strong topological phase in spinless systems. Here we break the framework of crystallographic symmetry groups by utilizing the projective symmetry gauge field, to realize the Mirror Chern Insulator in a bilayer twisted Hofstadter model. In experiments, the edge modes were unambiguously observed with odd-shaped boundaries, confirming the strong topological features.The clockwise and anti-clockwise edge states with opposite group velocities were completely separated via an energy drain. In addition, we demonstrate that MCI has robust topological whispering gallery modes. Our work establishes a strong foundation for investigating exotic topological effects arising from the interplays between artificial gauge fields and wave systems. The clockwise and anti-clockwise edge states with opposite group velocities were completely separated via an energy drain. In addition, we demonstrate that MCI has robust topological whispering gallery modes. Our work establishes a strong foundation for investigating exotic topological effects arising from the interplays between artificial gauge fields and wave systems., Comment: The manuscript will conduct modification to include more recent important progresses and reorganize the introduction

The size of deep learning models in artificial intelligence (AI) software is increasing rapidly, which hinders the large-scale deployment on resource-restricted devices (e.g., smartphones). To mitigate this issue, AI software compression plays a crucial role, which aims to compress model size while keeping high performance. However, the intrinsic defects in the big model may be inherited by the compressed one. Such defects may be easily leveraged by attackers, since the compressed models are usually deployed in a large number of devices without adequate protection. In this paper, we try to address the safe model compression problem from a safety-performance co-optimization perspective. Specifically, inspired by the test-driven development (TDD) paradigm in software engineering, we propose a test-driven sparse training framework called SafeCompress. By simulating the attack mechanism as the safety test, SafeCompress can automatically compress a big model to a small one following the dynamic sparse training paradigm. Further, considering a representative attack, i.e., membership inference attack (MIA), we develop a concrete safe model compression mechanism, called MIA-SafeCompress. Extensive experiments are conducted to evaluate MIA-SafeCompress on five datasets for both computer vision and natural language processing tasks. The results verify the effectiveness and generalization of our method. We also discuss how to adapt SafeCompress to other attacks besides MIA, demonstrating the flexibility of SafeCompress., Comment: Accepted by ASE2022. Camera-ready Version

Finsler geometry is a natural and fundamental generalization of Riemann geometry. The Finsler structure depends on both coordinates and velocities. We present the arrival time delay of astroparticles subject to Lorentz violation in the framework of Finsler geometry, and the result corresponds to that derived by Jacob and Piran in the standard model of cosmology., Comment: 7 latex pages, no figure, final version for journal publication

We propose to develop a wide-field and ultra-high-precision photometric survey mission, temporarily named "Earth 2.0 (ET)". This mission is designed to measure, for the first time, the occurrence rate and the orbital distributions of Earth-sized planets. ET consists of seven 30cm telescopes, to be launched to the Earth-Sun's L2 point. Six of these are transit telescopes with a field of view of 500 square degrees. Staring in the direction that encompasses the original Kepler field for four continuous years, this monitoring will return tens of thousands of transiting planets, including the elusive Earth twins orbiting solar-type stars. The seventh telescope is a 30cm microlensing telescope that will monitor an area of 4 square degrees toward the galactic bulge. This, combined with simultaneous ground-based KMTNet observations, will measure masses for hundreds of long-period and free-floating planets. Together, the transit and the microlensing telescopes will revolutionize our understandings of terrestrial planets across a large swath of orbital distances and free space. In addition, the survey data will also facilitate studies in the fields of asteroseismology, Galactic archeology, time-domain sciences, and black holes in binaries., Comment: 116 pages,79 figures

We present a one-dimensional coupled ring resonator lattice exhibiting a variant of the non- Hermitian skin effect (NHSE) that we call the anomalous Floquet NHSE. Unlike existing approaches to achieving the NHSE by engineering gain and loss on different ring segments, our design uses fixed on-site gain or loss in each ring. The anomalous Floquet NHSE is marked by the existence of skin modes at every value of the Floquet quasienergy, allowing for broadband asymmetric transmission. Varying the gain/loss induces a non-Hermitian topological phase transition, reversing the localization direction of the skin modes. An experimental implementation in an acoustic lattice yields good agreement with theoretical predictions, with a very broad relative bandwidth of around 40%., Comment: 7 pages, 3 figures

In the last few years, automated recommendation systems have been a major focus in the music field, where companies such as Spotify, Amazon, and Apple are competing in the ability to generate the most personalized music suggestions for their users. One of the challenges developers still fail to tackle is taking into account the psychological and emotional aspects of the music. Our goal is to find a way to integrate users' personal traits and their current emotional state into a single music recommendation system with both collaborative and content-based filtering. We seek to relate the personality and the current emotional state of the listener to the audio features in order to build an emotion-aware MRS. We compare the results both quantitatively and qualitatively to the output of the traditional MRS based on the Spotify API data to understand if our advancements make a significant impact on the quality of music recommendations.

We present a numerical framework of the new Langevin noise (LN) formalism [1,2] leveraging computational electromagnetics numerical methods to analyze quantum electromagnetic systems involving both medium and radiation losses. We then perform fully quantum-theoretic numerical simulations to retrieve quantum plasmonic Hong-Ou-Mandel (HOM) effects, demonstrated in recent experimental works [3-5], due to plasmonic interferences of two indistinguishable bosonic particles occupied in surface plasmon polariton (SPP) fields. The developed numerical framework would pave the path towards quantitative evaluations of open and dissipative quantum optics problems with the medium inhomogeneity and geometric complexity, e.g., quantum plasmonic phenomena and metasurface-based devices, useful for advancing the current quantum optics technology.

Pseudo-supervised learning methods have been shown to be effective for weakly supervised object localization tasks. However, the effectiveness depends on the powerful regularization ability of deep neural networks. Based on the assumption that the localization network should have similar location predictions on different versions of the same image, we propose a two-stage approach to learn more consistent localization. In the first stage, we propose a mask-based pseudo label generator algorithm, and use the pseudo-supervised learning method to initialize an object localization network. In the second stage, we propose a simple and effective method for evaluating the confidence of pseudo-labels based on classification discrimination, and by learning consistency from high-quality pseudo-labels, we further refine the localization network to get better localization performance. Experimental results show that our proposed approach achieves excellent performance in three benchmark datasets including CUB-200-2011, ImageNet-1k and Tiny-ImageNet, which demonstrates its effectiveness.

Federated learning is proposed by Google to safeguard data privacy through training models locally on users' devices. However, with deep learning models growing in size to achieve better results, it becomes increasingly difficult to accommodate the whole model on one single device. Thus, model parallelism is then used to divide the model weights among several devices. With this logic, the approach currently used evenly allocates weights among devices. However, in reality, a computation bottleneck may occur resulting from variant computing power of different users' devices. To address this problem, load balancing is needed to allocate the model weights based on the computational capability of the device. In this paper, we proposed Sky Computing, a load-balanced model parallelism framework to adaptively allocate the weights to devices. Sky Computing outperforms the baseline method by 55% in training time when training 160-layer BERT with 64 nodes. The source code can be found at https://github.com/hpcaitech/SkyComputing.

On-ramp merging areas are typical bottlenecks in the freeway network, since merging on-ramp vehicles may cause intensive disturbances on the mainline traffic flow and lead to various negative impacts on traffic efficiency and safety. The connected and autonomous vehicles (CAVs), with their capabilities of real-time communication and precise motion control, hold a great potential to facilitate ramp merging operation through enhanced coordination strategies. This paper presents a comprehensive review of the existing ramp merging strategies leveraging CAVs, focusing on the latest trends and developments in the research field. The review comprehensively covers 44 papers recently published in leading transportation journals. Based on the application context, control strategies are categorized into three categories: merging into sing-lane freeways with total CAVs, merging into sing-lane freeways with mixed traffic flows, and merging into multilane freeways. Relevant literature is reviewed regarding the required technologies, control decision level, applied methods, and impacts on traffic performance. More importantly, we identify the existing research gaps and provide insightful discussions on the potential and promising directions for future research based on the review, which facilitates further advancement in this research topic.

On-ramp merging areas are deemed to be typical bottlenecks for freeway networks due to the intensive disturbances induced by the frequent merging, weaving, and lane-changing behaviors. The Connected and Autonomous Vehicles (CAVs), benefited from their capabilities of real-time communication and precise motion control, hold an opportunity to promote ramp merging operation through enhanced cooperation. The existing CAV cooperation strategies are mainly designed for single-lane freeways, although multilane configurations are more prevailing in the real-world. In this paper, we present a flow-level CAV coordination strategy to facilitate merging operation in multilane freeways. The coordination integrates lane-change rules between mainstream lanes, proactive creation of large merging gaps, and platooning of ramp vehicles for enhanced benefits in traffic flow stability and efficiency. The strategy is formulated under an optimization framework, where the optimal control plan is determined based on real-time traffic conditions. The impacts of tunable model parameters on the produced control plan are discussed in detail. The efficiency of the proposed multilane strategy is demonstrated in a micro-simulation environment. The results show that the coordination can substantially improve the overall ramp merging efficiency and prevent recurrent traffic congestions, especially under high traffic volume conditions.

The release note is an essential software artifact of open-source software that documents crucial information about changes, such as new features and bug fixes. With the help of release notes, both developers and users could have a general understanding of the latest version without browsing the source code. However, it is a daunting and time-consuming job for developers to produce release notes. Although prior studies have provided some automatic approaches, they generate release notes mainly by extracting information from code changes. This will result in language-specific and not being general enough to be applicable. Therefore, helping developers produce release notes effectively remains an unsolved challenge. To address the problem, we first conduct a manual study on the release notes of 900 GitHub projects, which reveals that more than 54% of projects produce their release notes with pull requests. Based on the empirical finding, we propose a deep learning based approach named DeepRelease (Deep learning based Release notes generator) to generate release notes according to pull requests. The process of release notes generation in DeepRelease includes the change entries generation and the change category (i.e., new features or bug fixes) generation, which are formulated as a text summarization task and a multi-class classification problem, respectively. Since DeepRelease fully employs text information from pull requests to summarize changes and identify the change category, it is language-agnostic and can be used for projects in any language. We build a dataset with over 46K release notes and evaluate DeepRelease on the dataset. The experimental results indicate that DeepRelease outperforms four baselines and can generate release notes similar to those manually written ones in a fraction of the time., Comment: Accepted at APSEC'21