Searcharxiv⌕ Search

arXiv subjects

Yihan Wang

Publications and source records attributed to Yihan Wang.

124 records · Page 7Linked to original sources
arXiv2022

Close Encounters of Stars with Stellar-mass Black Hole Binaries

Many astrophysical environments, from star clusters and globular clusters to the disks of Active Galactic Nuclei, are characterized by frequent interactions between stars and the compact objects that they leave behind. Here, using a suite of $3-D$ hydrodynamics simulations, we explore the outcome of close interactions between $1M_{\odot}$ stars and binary black holes (BBHs) in the gravitational wave regime, resulting in a tidal disruption event (TDE) or a pure scattering, focusing on the accretion rates, the back reaction on the BH binary orbital parameters and the increase in the binary BH effective spin. We find that TDEs can make a significant impact on the binary orbit, which is often different from that of pure scattering. Binaries experiencing a prograde (retrograde) TDE tend to be widened (hardened) by up to $\simeq 20\%$. Initially circular binaries become more eccentric by $\lesssim 10\%$ by a prograde or retrograde TDE, whereas the eccentricity of initially eccentric binaries increases (decreases) by a retrograde (prograde) TDE by $\lesssim 5\%$. Overall a single TDE can generally result in changes of the gravitational wave-driven merger time scale by order unity. The accretion rates of both black holes are very highly super-Eddington, showing modulations (preferentially for retrograde TDEs) on a time scale of the orbital period, which can be a characteristic feature of BBH-driven TDEs. Prograde TDEs result in the effective spin parameter $χ$ to vary by $\lesssim 0.02$ while $χ\gtrsim -0.005$ for retrograde TDEs.

astro-ph.HE↗
arXiv2022

Lite Pose: Efficient Architecture Design for 2D Human Pose Estimation

Pose estimation plays a critical role in human-centered vision applications. However, it is difficult to deploy state-of-the-art HRNet-based pose estimation models on resource-constrained edge devices due to the high computational cost (more than 150 GMACs per frame). In this paper, we study efficient architecture design for real-time multi-person pose estimation on edge. We reveal that HRNet's high-resolution branches are redundant for models at the low-computation region via our gradual shrinking experiments. Removing them improves both efficiency and performance. Inspired by this finding, we design LitePose, an efficient single-branch architecture for pose estimation, and introduce two simple approaches to enhance the capacity of LitePose, including Fusion Deconv Head and Large Kernel Convs. Fusion Deconv Head removes the redundancy in high-resolution branches, allowing scale-aware feature fusion with low overhead. Large Kernel Convs significantly improve the model's capacity and receptive field while maintaining a low computational cost. With only 25% computation increment, 7x7 kernels achieve +14.0 mAP better than 3x3 kernels on the CrowdPose dataset. On mobile platforms, LitePose reduces the latency by up to 5.0x without sacrificing performance, compared with prior state-of-the-art efficient pose estimation models, pushing the frontier of real-time multi-person pose estimation on edge. Our code and pre-trained models are released at https://github.com/mit-han-lab/litepose.

cs.CV↗
arXiv2022

The emergence of diffused Gamma-Ray Burst afterglows from the disks of Active Galactic Nuclei

The disks of Active Galactic Nuclei (AGNs) have emerged as rich environments for the production and capture of stars and the compact objects that they leave behind. These stars produce long Gamma-Ray Bursts (LGRBs) at their deaths, while frequent interactions among compact objects form binary neutron stars and neutron star-black hole binaries, leading to short GRBs (SGRBs) upon their merger. Predicting the properties of these transients as they emerge from the dense environments of AGN disks is key to their proper identification and to better constrain the star and compact object population in AGN disks. Some of these transients would appear unusual because they take place in much higher densities than the interstellar medium. Others, which are the subject of this paper, would additionally be modified by radiation diffusion since they are generated within optically thick regions of the accretion disks. Here we compute the GRB afterglow light curves for diffused GRB sources for a representative variety of central black-hole masses and disk locations. We find that the radiation from radio to UV and soft X-rays can be strongly suppressed by synchrotron self-absorption in the dense medium of the AGN disk. In addition, photon diffusion can significantly delay the emergence of the emission peak, turning a beamed, fast transient into a slow, isotropic, and dimmer one. These would appear as broadband-correlated AGN variability with dominance at the higher frequencies. Their properties can constrain both the stellar populations within AGN disks as well as the disk structure.

astro-ph.HE↗
arXiv2022

Adversarial Parameter Attack on Deep Neural Networks

In this paper, a new parameter perturbation attack on DNNs, called adversarial parameter attack, is proposed, in which small perturbations to the parameters of the DNN are made such that the accuracy of the attacked DNN does not decrease much, but its robustness becomes much lower. The adversarial parameter attack is stronger than previous parameter perturbation attacks in that the attack is more difficult to be recognized by users and the attacked DNN gives a wrong label for any modified sample input with high probability. The existence of adversarial parameters is proved. For a DNN $F_Θ$ with the parameter set $Θ$ satisfying certain conditions, it is shown that if the depth of the DNN is sufficiently large, then there exists an adversarial parameter set $Θ_a$ for $Θ$ such that the accuracy of $F_{Θ_a}$ is equal to that of $F_Θ$, but the robustness measure of $F_{Θ_a}$ is smaller than any given bound. An effective training algorithm is given to compute adversarial parameters and numerical experiments are used to demonstrate that the algorithms are effective to produce high quality adversarial parameters.

cs.LG↗
arXiv2022

On the Convergence of Certified Robust Training with Interval Bound Propagation

Interval Bound Propagation (IBP) is so far the base of state-of-the-art methods for training neural networks with certifiable robustness guarantees when potential adversarial perturbations present, while the convergence of IBP training remains unknown in existing literature. In this paper, we present a theoretical analysis on the convergence of IBP training. With an overparameterized assumption, we analyze the convergence of IBP robust training. We show that when using IBP training to train a randomly initialized two-layer ReLU neural network with logistic loss, gradient descent can linearly converge to zero robust training error with a high probability if we have sufficiently small perturbation radius and large network width.

cs.LG↗
arXiv2021

Fast Certified Robust Training with Short Warmup

Recently, bound propagation based certified robust training methods have been proposed for training neural networks with certifiable robustness guarantees. Despite that state-of-the-art (SOTA) methods including interval bound propagation (IBP) and CROWN-IBP have per-batch training complexity similar to standard neural network training, they usually use a long warmup schedule with hundreds or thousands epochs to reach SOTA performance and are thus still costly. In this paper, we identify two important issues in existing methods, namely exploded bounds at initialization, and the imbalance in ReLU activation states and improve IBP training. These two issues make certified training difficult and unstable, and thereby long warmup schedules were needed in prior works. To mitigate these issues and conduct faster certified training with shorter warmup, we propose three improvements based on IBP training: 1) We derive a new weight initialization method for IBP training; 2) We propose to fully add Batch Normalization (BN) to each layer in the model, since we find BN can reduce the imbalance in ReLU activation states; 3) We also design regularization to explicitly tighten certified bounds and balance ReLU activation states during wamrup. We are able to obtain 65.03% verified error on CIFAR-10 ($ε=\frac{8}{255}$) and 82.36% verified error on TinyImageNet ($ε=\frac{1}{255}$) using very short training schedules (160 and 80 total epochs, respectively), outperforming literature SOTA trained with hundreds or thousands epochs under the same network architecture. The code is available at https://github.com/shizhouxing/Fast-Certified-Robust-Training.

cs.LG↗
arXiv2021

Symmetry Breaking in Dynamical Encounters in the Disks of Active Galactic Nuclei

Active galactic nucleus (AGN) disks may be important sites of binary black hole (BBH) mergers. Here we show via numerical experiments with the high-accuracy, high precision code {\tt SpaceHub} that broken symmetry in dynamical encounters in AGN disks can lead to an asymmetry between prograde and retrograde BBH mergers. The direction of the hardening asymmetry depends on the initial binary semi-major axis. An asymmetric distribution of mass-weighted projected spin $χ_{\rm eff}$ should therefore be expected in LIGO-Virgo detections of BBH mergers from AGN disks. This channel further predicts that negative $χ_{\rm eff}$ BBH mergers are most likely for massive binaries.

astro-ph.HE↗
arXiv2021

Hot Jupiter formation in dense clusters: secular chaos in multi-planetary systems

Exoplanetary observations reveal that the occurrence rate of hot Jupiters is correlated with star clustering. In star clusters, interactions between planetary systems and close fly-by stars can significantly change the architecture of primordially coplanar, circular planetary systems. Flybys in dense clusters have a significant impact on hot Jupiter formation via activation of high eccentricity excitation mechanisms such as the Zeipel-Lidov-Kozai (ZLK) effect and planet-planet scattering. Previous studies have shown that if there are two giant planets in the planetary system, close flybys can efficiently activate the ZLK mechanism, thus triggering high eccentricity tidal migration and ultimately form hot Jupiters in star clusters. Here we extend our previous study with a multi-planet (triple) system. We perform high precision, high-accuracy few-body simulations of stellar flybys and subsequent planetary migration within the perturbed planetary systems using the code {\tt SpaceHub}. Our simulations demonstrate that a single close flyby on a multi-planet system in a cluster can activate secular chaos and ultimately lead to hot Jupiter formation via high eccentricity migration. We find that the hot Jupiter formation rate per system increases with both the size of the planetary system as well as with the mass of the outer planet, and we quantify the relative formation fractions for a range of parameters. Hot Jupiters formed via secular chaos are expected to be accompanied by massive companions with very long periods. Our study further shows that this flyby-induced secular chaos is preferred in low-density clusters where multi-planet systems are more likely to survive, and that it contributes a significant fraction of the hot Jupiter formation in star clusters compared to the flyby-induced ZLK mechanism.

astro-ph.EP↗
arXiv2021

Fast and Complete: Enabling Complete Neural Network Verification with Rapid and Massively Parallel Incomplete Verifiers

Formal verification of neural networks (NNs) is a challenging and important problem. Existing efficient complete solvers typically require the branch-and-bound (BaB) process, which splits the problem domain into sub-domains and solves each sub-domain using faster but weaker incomplete verifiers, such as Linear Programming (LP) on linearly relaxed sub-domains. In this paper, we propose to use the backward mode linear relaxation based perturbation analysis (LiRPA) to replace LP during the BaB process, which can be efficiently implemented on the typical machine learning accelerators such as GPUs and TPUs. However, unlike LP, LiRPA when applied naively can produce much weaker bounds and even cannot check certain conflicts of sub-domains during splitting, making the entire procedure incomplete after BaB. To address these challenges, we apply a fast gradient based bound tightening procedure combined with batch splits and the design of minimal usage of LP bound procedure, enabling us to effectively use LiRPA on the accelerator hardware for the challenging complete NN verification problem and significantly outperform LP-based approaches. On a single GPU, we demonstrate an order of magnitude speedup compared to existing LP-based approaches.

cs.AI↗
arXiv2020

Automatic Perturbation Analysis for Scalable Certified Robustness and Beyond

Linear relaxation based perturbation analysis (LiRPA) for neural networks, which computes provable linear bounds of output neurons given a certain amount of input perturbation, has become a core component in robustness verification and certified defense. The majority of LiRPA-based methods focus on simple feed-forward networks and need particular manual derivations and implementations when extended to other architectures. In this paper, we develop an automatic framework to enable perturbation analysis on any neural network structures, by generalizing existing LiRPA algorithms such as CROWN to operate on general computational graphs. The flexibility, differentiability and ease of use of our framework allow us to obtain state-of-the-art results on LiRPA based certified defense on fairly complicated networks like DenseNet, ResNeXt and Transformer that are not supported by prior works. Our framework also enables loss fusion, a technique that significantly reduces the computational complexity of LiRPA for certified defense. For the first time, we demonstrate LiRPA based certified defense on Tiny ImageNet and Downscaled ImageNet where previous approaches cannot scale to due to the relatively large number of classes. Our work also yields an open-source library for the community to apply LiRPA to areas beyond certified defense without much LiRPA expertise, e.g., we create a neural network with a probably flat optimization landscape by applying LiRPA to network parameters. Our opensource library is available at https://github.com/KaidiXu/auto_LiRPA.

cs.LG↗
arXiv2020

Off-Policy Multi-Agent Decomposed Policy Gradients

Multi-agent policy gradient (MAPG) methods recently witness vigorous progress. However, there is a significant performance discrepancy between MAPG methods and state-of-the-art multi-agent value-based approaches. In this paper, we investigate causes that hinder the performance of MAPG algorithms and present a multi-agent decomposed policy gradient method (DOP). This method introduces the idea of value function decomposition into the multi-agent actor-critic framework. Based on this idea, DOP supports efficient off-policy learning and addresses the issue of centralized-decentralized mismatch and credit assignment in both discrete and continuous action spaces. We formally show that DOP critics have sufficient representational capability to guarantee convergence. In addition, empirical evaluations on the StarCraft II micromanagement benchmark and multi-agent particle environments demonstrate that DOP significantly outperforms both state-of-the-art value-based and policy-based multi-agent reinforcement learning algorithms. Demonstrative videos are available at https://sites.google.com/view/dop-mapg/.

cs.LG↗
arXiv2020

On $\ell_p$-norm Robustness of Ensemble Stumps and Trees

Recent papers have demonstrated that ensemble stumps and trees could be vulnerable to small input perturbations, so robustness verification and defense for those models have become an important research problem. However, due to the structure of decision trees, where each node makes decision purely based on one feature value, all the previous works only consider the $\ell_\infty$ norm perturbation. To study robustness with respect to a general $\ell_p$ norm perturbation, one has to consider the correlation between perturbations on different features, which has not been handled by previous algorithms. In this paper, we study the problem of robustness verification and certified defense with respect to general $\ell_p$ norm perturbations for ensemble decision stumps and trees. For robustness verification of ensemble stumps, we prove that complete verification is NP-complete for $p\in(0, \infty)$ while polynomial time algorithms exist for $p=0$ or $\infty$. For $p\in(0, \infty)$ we develop an efficient dynamic programming based algorithm for sound verification of ensemble stumps. For ensemble trees, we generalize the previous multi-level robustness verification algorithm to $\ell_p$ norm. We demonstrate the first certified defense method for training ensemble stumps and trees with respect to $\ell_p$ norm perturbations, and verify its effectiveness empirically on real datasets.

cs.LG↗
arXiv2019

Multi-Messenger Astronomy with Extremely Large Telescopes

The field of time-domain astrophysics has entered the era of Multi-messenger Astronomy (MMA). One key science goal for the next decade (and beyond) will be to characterize gravitational wave (GW) and neutrino sources using the next generation of Extremely Large Telescopes (ELTs). These studies will have a broad impact across astrophysics, informing our knowledge of the production and enrichment history of the heaviest chemical elements, constrain the dense matter equation of state, provide independent constraints on cosmology, increase our understanding of particle acceleration in shocks and jets, and study the lives of black holes in the universe. Future GW detectors will greatly improve their sensitivity during the coming decade, as will near-infrared telescopes capable of independently finding kilonovae from neutron star mergers. However, the electromagnetic counterparts to high-frequency (LIGO/Virgo band) GW sources will be distant and faint and thus demand ELT capabilities for characterization. ELTs will be important and necessary contributors to an advanced and complete multi-messenger network.

astro-ph.HE↗
arXiv2018

Fractal Dimensions Derived from Spatial Allometric Scaling of Urban Form

The improved city clustering algorithm can be used to identify urban boundaries on a digital map, and the results are a set of isolines. The relationships between the urban measurements within the variable boundaries follow allometric scaling law, which indicates spatial allometry of cities. This paper is devoted to exploring the fractal dimension proceeding from urban spatial allometry. By theoretical reasoning and empirical analysis of urban traffic network, we can derive a set of fractal dimension from the spatial allometry and reveal the basic property of the new fractal parameters. The findings are as follows. First, the fractal dimension values of traffic lines are higher than those of traffic nodes. Second, the fractal dimension values based on variable boundaries are lower than those based on the concentric circles. Conclusions can be reached that the fractal dimensions coming from spatial allometry are a type of correlation dimension rather than capacity dimension, and the relative growth rate of traffic points is greater than that of traffic nodes. This study provides new way of understanding allometry, fractals, and scaling in urban systems.

physics.soc-ph↗
arXiv2017

Localization of highest weight modules of a class of Extended Affine Lie Algebras

In 2006, Gao and Zeng \cite{GZ} gave the free field realizations of highest weight modules over a class of extended affine Lie algebras. In the present paper, applying the technique of localization to those free field realizations, we construct a class of new weight modules over the extended affine Lie algebras. We give necessary and sufficient conditions for these modules to be irreducible. In this way, we construct free field realizations for a class of simple weight modules with infinite weight multiplicities over the extended affine Lie algebras.

math.RT↗