arXiv subjects

Xinyu Li

Publications and source records attributed to Xinyu Li.

At least 199 records · Page 11Linked to original sources

arXiv2021

A new neighborhood structure for job shop scheduling problems

Job shop scheduling problem (JSP) is a widely studied NP-complete combinatorial optimization problem. Neighborhood structures play a critical role in solving JSP. At present, there are three state-of-the-art neighborhood structures, i.e., N5, N6, and N7. Improving the upper bounds of some famous benchmarks is inseparable from the role of these neighborhood structures. However, these existing neighborhood structures only consider the movement of critical operations within a critical block. According to our experiments, it is also possible to improve the makespan of a scheduling scheme by moving a critical operation outside its critical block. According to the above finding, this paper proposes a new N8 neighborhood structure considering the movement of critical operations within a critical block and the movement of critical operations outside the critical block. Besides, a neighborhood clipping method is designed to avoid invalid movement, reducing the computational time. Tabu search (TS) is a commonly used algorithm framework combined with neighborhood structures. This paper uses this framework to compare the N8 neighborhood structure with N5, N6, and N7 neighborhood structures on four famous benchmarks. The experimental results verify that the N8 neighborhood structure is more effective and efficient in solving JSP than the other state-of-the-art neighborhood structures.

cs.AI↗

arXiv2021

Video Contrastive Learning with Global Context

Contrastive learning has revolutionized self-supervised image representation learning field, and recently been adapted to video domain. One of the greatest advantages of contrastive learning is that it allows us to flexibly define powerful loss objectives as long as we can find a reasonable way to formulate positive and negative samples to contrast. However, existing approaches rely heavily on the short-range spatiotemporal salience to form clip-level contrastive signals, thus limit themselves from using global context. In this paper, we propose a new video-level contrastive learning method based on segments to formulate positive pairs. Our formulation is able to capture global context in a video, thus robust to temporal content change. We also incorporate a temporal order regularization term to enforce the inherent sequential structure of videos. Extensive experiments show that our video-level contrastive learning framework (VCLR) is able to outperform previous state-of-the-arts on five video datasets for downstream action classification, action localization and video retrieval. Code is available at https://github.com/amazon-research/video-contrastive-learning.

cs.CV↗

arXiv2021

Selective Feature Compression for Efficient Activity Recognition Inference

Most action recognition solutions rely on dense sampling to precisely cover the informative temporal clip. Extensively searching temporal region is expensive for a real-world application. In this work, we focus on improving the inference efficiency of current action recognition backbones on trimmed videos, and illustrate that one action model can also cover then informative region by dropping non-informative features. We present Selective Feature Compression (SFC), an action recognition inference strategy that greatly increase model inference efficiency without any accuracy compromise. Differently from previous works that compress kernel sizes and decrease the channel dimension, we propose to compress feature flow at spatio-temporal dimension without changing any backbone parameters. Our experiments on Kinetics-400, UCF101 and ActivityNet show that SFC is able to reduce inference speed by 6-7x and memory usage by 5-6x compared with the commonly used 30 crops dense sampling procedure, while also slightly improving Top1 Accuracy. We thoroughly quantitatively and qualitatively evaluate SFC and all its components and show how does SFC learn to attend to important video regions and to drop temporal features that are uninformative for the task of action recognition.

cs.CV↗

arXiv2021

Neutrino Fast Flavor Conversions in Neutron-star Post-Merger Accretion Disks

A compact accretion disk may be formed in the merger of two neutron stars or of a neutron star and a stellar-mass black hole. Outflows from such accretion disks have been identified as a major site of rapid neutron-capture (r-process) nucleosynthesis and as the source of 'red' kilonova emission following the first observed neutron-star merger GW170817. We present long-term general-relativistic radiation magnetohydrodynamic simulations of a typical post-merger accretion disk at initial accretion rates of $\dot{M}\sim 1\,M_\odot\,\text{s}^{-1}$ over 400ms post-merger. We include neutrino radiation transport that accounts for effects of neutrino fast flavor conversions dynamically. We find ubiquitous flavor oscillations that result in a significantly more neutron-rich outflow, providing lanthanide and 3rd-peak r-process abundances similar to solar abundances. This provides strong evidence that post-merger accretion disks are a major production site of heavy r-process elements. A similar flavor effect may allow for increased lanthanide production in collapsars. The formalism presented here may also be used in simulations of core-collapse supernovae to explore whether fast conversions strengthen or weaken the explosion.

astro-ph.HE↗

arXiv2021

SiamMOT: Siamese Multi-Object Tracking

In this paper, we focus on improving online multi-object tracking (MOT). In particular, we introduce a region-based Siamese Multi-Object Tracking network, which we name SiamMOT. SiamMOT includes a motion model that estimates the instance's movement between two frames such that detected instances are associated. To explore how the motion modelling affects its tracking capability, we present two variants of Siamese tracker, one that implicitly models motion and one that models it explicitly. We carry out extensive quantitative experiments on three different MOT datasets: MOT17, TAO-person and Caltech Roadside Pedestrians, showing the importance of motion modelling for MOT and the ability of SiamMOT to substantially outperform the state-of-the-art. Finally, SiamMOT also outperforms the winners of ACM MM'20 HiEve Grand Challenge on HiEve dataset. Moreover, SiamMOT is efficient, and it runs at 17 FPS for 720P videos on a single modern GPU. Codes are available in \url{https://github.com/amazon-research/siam-mot}.

cs.CV↗

arXiv2021

Fast dissipation of Colliding Alfvén Waves in a Magnetically Dominated Plasma

Magnetic energy around compact objects often dominates over plasma rest mass, and its dissipation can power the object luminosity. We describe a dissipation mechanism which works faster than magnetic reconnection. The mechanism involves two strong Alfvén waves with anti-aligned magnetic fields $\boldsymbol{B}_1$ and $\boldsymbol{B}_2$ that propagate in opposite directions along background magnetic field $\boldsymbol{B}_0$ and collide. The collision forms a thin current sheet perpendicular to $\boldsymbol{B}_0$, which absorbs the incoming waves. The current sheet is sustained by electric field $\boldsymbol{E}$ breaking the magnetohydrodynamic condition $E 1/2$. It dissipates a large fraction of the wave energy, $f=(2A-1)/A^2$, reaching $100\%$ when $A=1$. The plane geometry allows one to see the dissipation process in a one-dimensional simulation. We also perform two-dimensional simulations, enabling spontaneous breaking of the plane symmetry by the tearing instability of the current sheet. At moderate $A$ of main interest the tearing instability is suppressed. Dissipation transitions to normal, slower, magnetic reconnection at $A\gg 1$. The fast dissipation described in this paper may occur in various objects with perturbed magnetic fields, including magnetars, jets from accreting black holes, and pulsar wind nebulae.

astro-ph.HE↗

arXiv2021

Multi-Label Activity Recognition using Activity-specific Features and Activity Correlations

Multi-label activity recognition is designed for recognizing multiple activities that are performed simultaneously or sequentially in each video. Most recent activity recognition networks focus on single-activities, that assume only one activity in each video. These networks extract shared features for all the activities, which are not designed for multi-label activities. We introduce an approach to multi-label activity recognition that extracts independent feature descriptors for each activity and learns activity correlations. This structure can be trained end-to-end and plugged into any existing network structures for video classification. Our method outperformed state-of-the-art approaches on four multi-label activity recognition datasets. To better understand the activity-specific features that the system generated, we visualized these activity-specific features in the Charades dataset.

cs.CV↗

arXiv2021

Imaging local discharge cascades for correlated electrons in WS2/WSe2 moiré superlattices

Transition metal dichalcogenide (TMD) moiré heterostructures provide an ideal platform to explore the extended Hubbard model1 where long-range Coulomb interactions play a critical role in determining strongly correlated electron states. This has led to experimental observations of Mott insulator states at half filling2-4 as well as a variety of extended Wigner crystal states at different fractional fillings5-9. Microscopic understanding of these emerging quantum phases, however, is still lacking. Here we describe a novel scanning tunneling microscopy (STM) technique for local sensing and manipulation of correlated electrons in a gated WS2/WSe2 moiré superlattice that enables experimental extraction of fundamental extended Hubbard model parameters. We demonstrate that the charge state of local moiré sites can be imaged by their influence on STM tunneling current, analogous to the charge-sensing mechanism in a single-electron transistor. In addition to imaging, we are also able to manipulate the charge state of correlated electrons. Discharge cascades of correlated electrons in the moiré superlattice are locally induced by ramping the STM bias, thus enabling the nearest-neighbor Coulomb interaction (UNN) to be estimated. 2D mapping of the moiré electron charge states also enables us to determine onsite energy fluctuations at different moiré sites. Our technique should be broadly applicable to many semiconductor moiré systems, offering a powerful new tool for microscopic characterization and control of strongly correlated states in moiré superlattices.

cond-mat.mes-hall↗

arXiv2020

Learning-based Prediction and Uplink Retransmission for Wireless Virtual Reality (VR) Network

Wireless Virtual Reality (VR) users are able to enjoy immersive experience from anywhere at anytime. However, providing full spherical VR video with high quality under limited VR interaction latency is challenging. If the viewpoint of the VR user can be predicted in advance, only the required viewpoint is needed to be rendered and delivered, which can reduce the VR interaction latency. Therefore, in this paper, we use offline and online learning algorithms to predict viewpoint of the VR user using real VR dataset. For the offline learning algorithm, the trained learning model is directly used to predict the viewpoint of VR users in continuous time slots. While for the online learning algorithm, based on the VR user's actual viewpoint delivered through uplink transmission, we compare it with the predicted viewpoint and update the parameters of the online learning algorithm to further improve the prediction accuracy. To guarantee the reliability of the uplink transmission, we integrate the Proactive retransmission scheme into our proposed online learning algorithm. Simulation results show that our proposed online learning algorithm for uplink wireless VR network with the proactive retransmission scheme only exhibits about 5% prediction error.

eess.SP↗

arXiv2020

NUTA: Non-uniform Temporal Aggregation for Action Recognition

In the world of action recognition research, one primary focus has been on how to construct and train networks to model the spatial-temporal volume of an input video. These methods typically uniformly sample a segment of an input clip (along the temporal dimension). However, not all parts of a video are equally important to determine the action in the clip. In this work, we focus instead on learning where to extract features, so as to focus on the most informative parts of the video. We propose a method called the non-uniform temporal aggregation (NUTA), which aggregates features only from informative temporal segments. We also introduce a synchronization method that allows our NUTA features to be temporally aligned with traditional uniformly sampled video features, so that both local and clip-level features can be combined. Our model has achieved state-of-the-art performance on four widely used large-scale action-recognition datasets (Kinetics400, Kinetics700, Something-something V2 and Charades). In addition, we have created a visualization to illustrate how the proposed NUTA method selects only the most relevant parts of a video clip.

cs.CV↗

arXiv2020

A Comprehensive Study of Deep Video Action Recognition

Video action recognition is one of the representative tasks for video understanding. Over the last decade, we have witnessed great advancements in video action recognition thanks to the emergence of deep learning. But we also encountered new challenges, including modeling long-range temporal information in videos, high computation costs, and incomparable results due to datasets and evaluation protocol variances. In this paper, we provide a comprehensive survey of over 200 existing papers on deep learning for video action recognition. We first introduce the 17 video action recognition datasets that influenced the design of models. Then we present video action recognition models in chronological order: starting with early attempts at adapting deep learning, then to the two-stream networks, followed by the adoption of 3D convolutional kernels, and finally to the recent compute-efficient models. In addition, we benchmark popular methods on several representative datasets and release code for reproducibility. In the end, we discuss open problems and shed light on opportunities for video action recognition to facilitate new research ideas.

cs.CV↗

arXiv2020

Quantifying the potential and flux landscapes for nonequilibrium multiverse, a new scenario for time arrow

We propose a new scenario of nonequilibirum multiverse. We quantified the potential landscape and the flux landscape for the Bousso-Polchinski type of multiverse. The potential landscape quantifies the weight of each universe. When the terminal vacuum with zero (flat) or negative cosmological constant (AdS) have a chance to tunnel back to the normal universes with positive cosmological constant (dS) through the bounce suggested by the recent studies, the detailed balance of the populations of the multiverse can be broken. We found that the degree of the detailed balance breaking can be quantified by the underlying average flux and associated flux landscape, which gives arise to the dynamical origin of irreversibility and the arrow of time. We also showed that the steady state of the multiverse is maintained by the thermodynamic cost quantified by the entropy production rate which is associated to the flux. This gives arise to thermodynamic origin of time irreversibility. We show that terminal vacuum universes can have dominant weights or lowest potentials giving arise to a funnel shaped potential landscape, while terminal vacuum universes together with other normal universes including ours can form dominant cycles giving arise to a funnel shaped cycle flux landscape. This indicates that even our universe may not be distinct from others based on the probability measure, it may lie in the dominant cycle(s), leading to higher chance of being found. This may provide an additional way beyond the anthropological principle for identifying our universe.

hep-th↗

arXiv2020

Don't cross the streams: caustics from Fuzzy Dark Matter

We study how tidal streams from globular clusters may be used to constrain the mass of ultra-light dark matter particles, called `fuzzy' dark matter (FDM). A general feature of FDM models is the presence of ubiquitous density fluctuations in bound, virialized dark matter structures, on the scale of the de Broglie wavelength, arising from wave interference in the evolving dark matter distribution. These time-varying fluctuations can disturb the motions of stars, leading to potentially observable signatures in cold thin tidal streams in our own Galaxy. The study of this effect has been hindered by the difficulty in simulating the FDM wavefunction in Milky Way-sized systems. We present a simple method to evolve realistic wavefunctions in nearly static potentials, that should provide an accurate estimate of this granulation effect. We quantify the impact of FDM perturbations on tidal streams, and show that initially, while stream perturbations are small in amplitude, their power spectra exhibit a sharp cutoff corresponding to the de Broglie wavelength of the FDM potential fluctuations. Eventually, when stream perturbations become nonlinear, fold caustics generically arise that lead to density fluctuations with universal behavior. This erases the signature of the de Broglie wavelength in the stream density power spectrum, but we show that it will still be possible to determine the FDM mass in this regime, by considering the fluctuations in quantities like angular momenta or actions.

astro-ph.CO↗

arXiv2020

Oscillations and Random Walk of the Soliton Core in a Fuzzy Dark Matter Halo

A Fuzzy Dark Matter (FDM) halo consists of a soliton core close to the center and an NFW-like density profile in the outer region. Previous investigations found that the soliton core exhibits temporal oscillations and random walk excursions around the halo center. Analyzing a set of numerical simulations, we show that both phenomena can be understood as the results of wave interference -- a suitable superposition of the ground (solitonic) state and excited states in a fixed potential suffices to account for the main features of these phenomena. Such an eigenmode analysis can shed light on the evolution of a satellite halo undergoing tidal disruption. As the outer halo is stripped away, reducing the amplitudes of the excited states, the ground state evolves adiabatically. This suggests diminished soliton oscillations and random walk excursions, an effect to consider in deducing constraints from stellar heating.

astro-ph.CO↗

arXiv2020

Vortices and waves in light dark matter

In a galactic halo like the Milky Way, bosonic dark matter particles lighter than about $30$ eV have a de Broglie wavelength larger than the average inter-particle separation and are therefore well described as a set of classical waves. This applies to, for instance, the QCD axion as well as to lighter axion-like particles such as fuzzy dark matter. We show that the interference of waves inside a halo inevitably leads to vortices, locations where chance destructive interference takes the density to zero. The phase of the wavefunction has non-trivial winding around these points. This can be interpreted as a non-zero velocity circulation, so that vortices are sites where the fluid velocity has a non-vanishing curl. Using analytic arguments and numerical simulations, we study the properties of vortices and show they have a number of universal features: (1) In three spatial dimensions, the generic defects take the form of vortex rings. (2) On average there is about one vortex ring per de Broglie volume and (3) generically only single winding ($\pm 1$) vortices are found in a realistic halo. (4) The density near a vortex scales as $r^2$ while the velocity goes as $1/r$, where $r$ is the distance to vortex. (5) A vortex segment moves at a velocity inversely proportional to its curvature scale so that smaller vortex rings move faster, allowing momentary motion exceeding escape velocity. We discuss observational/experimental signatures from vortices and, more broadly, wave interference. In the ultra-light regime, gravitational lensing by interference substructures leads to flux anomalies of $5-10 \%$ in strongly lensed systems. For QCD axions, vortices lead to a diminished signal in some detection experiments but not in others. We advocate the measurement of correlation functions by axion detection experiments as a way to probe and capitalize on the expected interference substructures.

astro-ph.CO↗

arXiv2020

Relativistic Alfvén Waves Entering Charge Starvation in the Magnetospheres of Neutron Stars

Instabilities in a neutron star can generate Alfvén waves in its magnetosphere. Propagation along the curved magnetic field lines strongly shears the wave, boosting its electric current $j_{\rm A}$. We derive an analytic expression for the evolution of the wave vector $\boldsymbol{k}$ and the growth of $j_{\rm A}$. In the strongly sheared regime, $j_{\rm A}$ may exceed the maximum current $j_{0}$ that can be supported by the background $e^{\pm}$ plasma. We investigate these "charge-starved" waves, first using a simplified two-fluid analytic model, then with first-principles kinetic simulations. We find that the Alfvén wave continues to propagate successfully even when $κ\equiv j_{\rm A}/j_{0} \gg 1$. It sustains $j_{\rm A}$ by compressing and advecting the plasma along the magnetic field lines with particle Lorentz factors $\sim κ^{1/2}$. The simulations show how plasma instabilities lead to gradual dissipation of the wave energy, giving a dissipation power $L_{\rm diss}\sim 10^{35}(κ/100)^{1/2} (B_w/10^{11}\,{\rm G})\,\mathrm{erg/s}$, where $B_w$ is the wave amplitude. Our results imply that dissipation due to charge starvation is not sufficient to power observed fast radio bursts (FRBs), in contrast to recent proposals.

astro-ph.HE↗

arXiv2020

Directional Temporal Modeling for Action Recognition

Many current activity recognition models use 3D convolutional neural networks (e.g. I3D, I3D-NL) to generate local spatial-temporal features. However, such features do not encode clip-level ordered temporal information. In this paper, we introduce a channel independent directional convolution (CIDC) operation, which learns to model the temporal evolution among local features. By applying multiple CIDC units we construct a light-weight network that models the clip-level temporal evolution across multiple spatial scales. Our CIDC network can be attached to any activity recognition backbone network. We evaluate our method on four popular activity recognition datasets and consistently improve upon state-of-the-art techniques. We further visualize the activation map of our CIDC network and show that it is able to focus on more meaningful, action related parts of the frame.

cs.CV↗

arXiv2020

Imaging moiré flat bands in 3D reconstructed WSe2/WS2 superlattices

Moiré superlattices in transition metal dichalcogenide (TMD) heterostructures can host novel correlated quantum phenomena due to the interplay of narrow moiré flat bands and strong, long-range Coulomb interactions1-5. However, microscopic knowledge of the atomically-reconstructed moiré superlattice and resulting flat bands is still lacking, which is critical for fundamental understanding and control of the correlated moiré phenomena. Here we quantitatively study the moiré flat bands in three-dimensional (3D) reconstructed WSe2/WS2 moiré superlattices by comparing scanning tunneling spectroscopy (STS) of high quality exfoliated TMD heterostructure devices with ab initio simulations of TMD moiré superlattices. A strong 3D buckling reconstruction accompanied by large in-plane strain redistribution is identified in our WSe2/WS2 moiré heterostructures. STS imaging demonstrates that this results in a remarkably narrow and highly localized K-point moiré flat band at the valence band edge of the heterostructure. A series of moiré flat bands are observed at different energies that exhibit varying degrees of localization. Our observations contradict previous simplified theoretical models but agree quantitatively with ab initio simulations that fully capture the 3D structural reconstruction. Here the strain redistribution and 3D buckling dominate the effective moiré potential and result in moiré flat bands at the Brillouin zone K points.

cond-mat.mes-hall↗