arXiv subjects

Haotian Liu

Publications and source records attributed to Haotian Liu.

100 records · Page 6Linked to original sources

arXiv2021

Online Multi-agent Reinforcement Learning for Decentralized Inverter-based Volt-VAR Control

The distributed Volt/Var control (VVC) methods have been widely studied for active distribution networks(ADNs), which is based on perfect model and real-time P2P communication. However, the model is always incomplete with significant parameter errors and such P2P communication system is hard to maintain. In this paper, we propose an online multi-agent reinforcement learning and decentralized control framework (OLDC) for VVC. In this framework, the VVC problem is formulated as a constrained Markov game and we propose a novel multi-agent constrained soft actor-critic (MACSAC) reinforcement learning algorithm. MACSAC is used to train the control agents online, so the accurate ADN model is no longer needed. Then, the trained agents can realize decentralized optimal control using local measurements without real-time P2P communication. The OLDC with MACSAC has shown extraordinary flexibility, efficiency and robustness to various computing and communication conditions. Numerical simulations on IEEE test cases not only demonstrate that the proposed MACSAC outperforms the state-of-art learning algorithms, but also support the superiority of our OLDC framework in the online application.

eess.SY↗

arXiv2021

YolactEdge: Real-time Instance Segmentation on the Edge

We propose YolactEdge, the first competitive instance segmentation approach that runs on small edge devices at real-time speeds. Specifically, YolactEdge runs at up to 30.8 FPS on a Jetson AGX Xavier (and 172.7 FPS on an RTX 2080 Ti) with a ResNet-101 backbone on 550x550 resolution images. To achieve this, we make two improvements to the state-of-the-art image-based real-time method YOLACT: (1) applying TensorRT optimization while carefully trading off speed and accuracy, and (2) a novel feature warping module to exploit temporal redundancy in videos. Experiments on the YouTube VIS and MS COCO datasets demonstrate that YolactEdge produces a 3-5x speed up over existing real-time methods while producing competitive mask and box detection accuracy. We also conduct ablation studies to dissect our design choices and modules. Code and models are available at https://github.com/haotian-liu/yolact_edge.

cs.CV↗

arXiv2021

Universal time delay in static spherically symmetric spacetimes for null and timelike signals

A perturbative method to compute the total travel time of both null and lightlike rays in arbitrary static spherically symmetric spacetimes in the weak field limit is proposed. The resultant total time takes a quasi-series form of the impact parameter. The coefficient of this series at a certain order $n$ is shown to be determined by the asymptotic expansion of the metric functions to the order $n+1$. To the leading order(s), the time delay, as well as the difference between the time delays of two kinds of relativistic signals, is then shown to take a universal form for all SSS spacetimes. This universal form depends on the mass $M$ and a post-Newtonian parameter $γ$ of the spacetime. The analytical result is numerically verified using the central black hole of M87 as the gravitational lensing center.

gr-qc↗

arXiv2021

Time delay in the strong field limit for null and timelike signals and its simple interpretation

Gravitational lensing can happen not only for null signal but also timelike signals such as neutrinos and massive gravitational waves in some theories beyond GR. In this work we study the time delay between different relativistic images formed by signals with arbitrary asymptotic velocity $v$ in general static and spherically symmetric spacetimes. A perturbative method is used to calculate the total travel time in the strong field limit, which is found to be in quasi-series of the small parameter $a=1-b_c/b$ where $b$ is the impact parameter and $b_c$ is its critical value. The coefficients of the series are completely fixed by the behaviour of the metric functions near the particle sphere $r_c$ and only the first term of the series contains a weak logarithmic divergence. The time delay $Δt_{n,m}$ to the leading non-trivial order was shown to equal the particle sphere circumference divided by the local signal velocity and multiplied by the winding number and the redshift factor. By assuming the Sgr A* supermassive black hole is a Hayward one, we were able to validate the quasi-series form of the total time, and reveal the effects of the spacetime parameter $l$, the signal velocity $v$ and the source/detector coordinate difference $Δϕ_{sd}$ on the time delay. It is found that as $l$ increase from 0 to its critical value $l_c$, both $r_c$ and $Δt_{n,m}$ decrease. The variation of $Δt_{n+1,n}$ for $l$ from 0 to $l_c$ can be as large as $7.2\times 10^1$ [s], whose measurement then can be used to constrain the value of $l$. While for ultra-relativistic neutrino or gravitational wave, the variation of $Δt_{n,m}$ is too small to be resolved. The dependence of $Δt_{n,-n}$ on $Δϕ_{sd}$ shows that to temporally resolve the two sequences of images from opposite sides of the lens, $|Δϕ_{sd}-π|$ has to be larger than certain value.

gr-qc↗

arXiv2020

Constraining the spacetime spin using time delay in stationary axisymmetric spacetimes

Total travel time $t$ and time delay $Δt$ between images of gravitational lensing (GL) in the equatorial plane of stationary axisymmetric (SAS) spacetimes for null and timelike signals with arbitrary velocity are studied. Using a perturbative method in the weak field limit, $t$ in general SAS spacetimes is expressed as a quasi-series of the impact parameter $b$ with coefficients involving the source-lens distance $r_s$ and lens-detector distances $r_d$, signal velocity $v$, and asymptotic expansion coefficients of the metric functions. The time delay $Δt$ to the leading order(s) were shown to be determined by the spacetime mass $M$, spin angular momentum $a$ and post-Newtonian parameter $γ$, and kinematic variables $r_s,~r_d,~v$ and source angular position $β$. When $β\ll \sqrt{aM}/r_{s,d}$, $Δt$ is dominated by the contribution linear to spin $a$. Modeling the Sgr A* supermassive black hole as a Kerr-Newman black hole, we show that as long as $β\lesssim 1.5\times 10^{-5}$ [$^{\prime\prime}$], then $Δt$ will be able to reach the $\mathcal{O}(1)$ second level, which is well within the time resolution of current GRB, gravitational wave and neutrino observatories. Therefore measuring $Δt$ in GL of these signals will allow us to constrain the spin of the Sgr A*.

gr-qc↗

arXiv2020

Two-stage Deep Reinforcement Learning for Inverter-based Volt-VAR Control in Active Distribution Networks

Model-based Vol/VAR optimization method is widely used to eliminate voltage violations and reduce network losses. However, the parameters of active distribution networks(ADNs) are not onsite identified, so significant errors may be involved in the model and make the model-based method infeasible. To cope with this critical issue, we propose a novel two-stage deep reinforcement learning (DRL) method to improve the voltage profile by regulating inverter-based energy resources, which consists of offline stage and online stage. In the offline stage, a highly efficient adversarial reinforcement learning algorithm is developed to train an offline agent robust to the model mismatch. In the sequential online stage, we transfer the offline agent safely as the online agent to perform continuous learning and controlling online with significantly improved safety and efficiency. Numerical simulations on IEEE test cases not only demonstrate that the proposed adversarial reinforcement learning algorithm outperforms the state-of-art algorithm, but also show that our proposed two-stage method achieves much better performance than the existing DRL based methods in the online application.

eess.SY↗

arXiv2019

Using Deep Learning and Machine Learning to Detect Epileptic Seizure with Electroencephalography (EEG) Data

The prediction of epileptic seizure has always been extremely challenging in medical domain. However, as the development of computer technology, the application of machine learning introduced new ideas for seizure forecasting. Applying machine learning model onto the predication of epileptic seizure could help us obtain a better result and there have been plenty of scientists who have been doing such works so that there are sufficient medical data provided for researchers to do training of machine learning models.

cs.LG↗

arXiv2019

Time delay of timelike particles in gravitational lensing of Schwarzschild spacetime

Time delay in Schwarzschild spacetime for null and timelike signals with arbitrary velocity $v$ is studied. The total travel time $t_\mathrm{if}$ is evaluated both exactly and approximately in the weak field limit, with the result given as functions of signal velocity, source-lens and lens-observer distances, angular position of the source and lens mass. Two time delays, $Δt_v$ between signals with different velocities but coming from same side of the lens and $Δt_\mathrm{p}$ between signals from different sides of the lens, as well as the difference $Δt_{\mathrm{p}v}$ between two $Δt_\mathrm{p}$'s are calculated. These time delays are applied to the gravitational-lensed supernova neutrinos and gravitational waves (GW). It is shown that the $Δt_v$ between different mass eigenstates of supernova neutrinos can be related to the mass square difference of these eigenstates and therefore could potentially be used to discriminate neutrino mass orderings, while the difference $Δt_{\mathrm{p}v}$ between neutrino and optical signals can be correlated with the absolute mass of neutrinos. The formula for time delay in a general lens mass profile is derived and the result is applied to the singular isothermal sphere case. For GWs, it is found that the difference $Δt_{\mathrm{p}v}$ between GW and GRB can only reach $1.45\times 10^{-5}$ second for very large source distance ($2\times 10^4$ [Mpc]) and source angle (10 [as]) if $v_{GM}=(1-3\times 10^{-15})c$. This time difference is at least three order smaller than uncertainties in time measurement of the recently observed GW/GRB signals and thus calls for improvement if $Δt_{\mathrm{p}v}$ is to be used to further constrain the GW velocity.

gr-qc↗

arXiv2016

Tensor Computation: A New Framework for High-Dimensional Problems in EDA

Many critical EDA problems suffer from the curse of dimensionality, i.e. the very fast-scaling computational burden produced by large number of parameters and/or unknown variables. This phenomenon may be caused by multiple spatial or temporal factors (e.g. 3-D field solvers discretizations and multi-rate circuit simulation), nonlinearity of devices and circuits, large number of design or optimization parameters (e.g. full-chip routing/placement and circuit sizing), or extensive process variations (e.g. variability/reliability analysis and design for manufacturability). The computational challenges generated by such high dimensional problems are generally hard to handle efficiently with traditional EDA core algorithms that are based on matrix and vector computation. This paper presents "tensor computation" as an alternative general framework for the development of efficient EDA algorithms and tools. A tensor is a high-dimensional generalization of a matrix and a vector, and is a natural choice for both storing and solving efficiently high-dimensional EDA problems. This paper gives a basic tutorial on tensors, demonstrates some recent examples of EDA applications (e.g., nonlinear circuit modeling and high-dimensional uncertainty quantification), and suggests further open EDA problems where the use of tensor computation could be of advantage.

math.NA↗

arXiv2015

A Constructive Algorithm for Decomposing a Tensor into a Finite Sum of Orthonormal Rank-1 Terms

We propose a constructive algorithm that decomposes an arbitrary real tensor into a finite sum of orthonormal rank-1 outer products. The algorithm, named TTr1SVD, works by converting the tensor into a tensor-train rank-1 (TTr1) series via the singular value decomposition (SVD). TTr1SVD naturally generalizes the SVD to the tensor regime with properties such as uniqueness for a fixed order of indices, orthogonal rank-1 outer product terms, and easy truncation error quantification. Using an outer product column table it also allows, for the first time, a complete characterization of all tensors orthogonal with the original tensor. Incidentally, this leads to a strikingly simple constructive proof showing that the maximum rank of a real $2 \times 2 \times 2$ tensor over the real field is 3. We also derive a conversion of the TTr1 decomposition into a Tucker decomposition with a sparse core tensor. Numerical examples illustrate each of the favorable properties of the TTr1 decomposition.

math.NA↗