Searcharxiv⌕ Search

arXiv subjects

Jun Yang

Publications and source records attributed to Jun Yang.

At least 235 records · Page 13Linked to original sources
arXiv2023

Ansor: Generating High-Performance Tensor Programs for Deep Learning

High-performance tensor programs are crucial to guarantee efficient execution of deep neural networks. However, obtaining performant tensor programs for different operators on various hardware platforms is notoriously challenging. Currently, deep learning systems rely on vendor-provided kernel libraries or various search strategies to get performant tensor programs. These approaches either require significant engineering effort to develop platform-specific optimization code or fall short of finding high-performance programs due to restricted search space and ineffective exploration strategy. We present Ansor, a tensor program generation framework for deep learning applications. Compared with existing search strategies, Ansor explores many more optimization combinations by sampling programs from a hierarchical representation of the search space. Ansor then fine-tunes the sampled programs with evolutionary search and a learned cost model to identify the best programs. Ansor can find high-performance programs that are outside the search space of existing state-of-the-art approaches. In addition, Ansor utilizes a task scheduler to simultaneously optimize multiple subgraphs in deep neural networks. We show that Ansor improves the execution performance of deep neural networks relative to the state-of-the-art on the Intel CPU, ARM CPU, and NVIDIA GPU by up to $3.8\times$, $2.6\times$, and $1.7\times$, respectively.

cs.LG↗
arXiv2023

Toward Microarcsecond Astrometry for the Innermost Wobbling Jet of the BL Lacertae Object OJ 287

The BL Lacertae object OJ 287 is a very unusual quasar producing a wobbling radio jet and some double-peaked optical outbursts with a possible period of about 12 yr for more than one century. This variability is widely explained by models of binary supermassive black hole (SMBH) or precessing jet/disk from a single SMBH. To enable an independent and nearly bias-free investigation on these possible scenarios, we explored the feasibility of extremely high-precision differential astrometry on its innermost restless jet at mm-wavelengths. Through re-visiting some existing radio surveys and very long baseline interferometry (VLBI) data at frequencies from 1.4 to 15.4 GHz and performing new Very Long Baseline Array (VLBA) observations at 43.2 GHz, we find that the radio source J0854$+$1959, 7.1 arcmin apart from OJ 287 and no clearly-seen optical and infrared counterparts, could provide a nearly ideal reference point to track the complicated jet activity of OJ 287. The source J0854$+$1959 has a stable GHz-peaked radio spectrum and shows a jet structure consisting of two discrete, mas-scale-compact and steep-spectrum components and showing no proper motion over about 8 yr. The stable VLBI structure can be interpreted by an episodic, optically thin and one-sided jet. With respect to its 4.1-mJy peak feature at 43.2 GHz, we have achieved an astrometric precision at the state-of-art level, about 10 $μ$as. These results indicate that future VLBI astrometry on OJ 287 could allow us to accurately locate its jet apex and activity boundary, align its restless jet structure over decades without significant systematic bias, and probe various astrophysical scenarios.

astro-ph.HE↗
arXiv2023

Uncertainty-aware 3D Object-Level Mapping with Deep Shape Priors

3D object-level mapping is a fundamental problem in robotics, which is especially challenging when object CAD models are unavailable during inference. In this work, we propose a framework that can reconstruct high-quality object-level maps for unknown objects. Our approach takes multiple RGB-D images as input and outputs dense 3D shapes and 9-DoF poses (including 3 scale parameters) for detected objects. The core idea of our approach is to leverage a learnt generative model for shape categories as a prior and to formulate a probabilistic, uncertainty-aware optimization framework for 3D reconstruction. We derive a probabilistic formulation that propagates shape and pose uncertainty through two novel loss functions. Unlike current state-of-the-art approaches, we explicitly model the uncertainty of the object shapes and poses during our optimization, resulting in a high-quality object-level mapping system. Moreover, the resulting shape and pose uncertainties, which we demonstrate can accurately reflect the true errors of our object maps, can also be useful for downstream robotics tasks such as active vision. We perform extensive evaluations on indoor and outdoor real-world datasets, achieving achieves substantial improvements over state-of-the-art methods. Our code will be available at https://github.com/TRAILab/UncertainShapePose.

cs.CV↗
arXiv2023

Active Pose Refinement for Textureless Shiny Objects using the Structured Light Camera

6D pose estimation of textureless shiny objects has become an essential problem in many robotic applications. Many pose estimators require high-quality depth data, often measured by structured light cameras. However, when objects have shiny surfaces (e.g., metal parts), these cameras fail to sense complete depths from a single viewpoint due to the specular reflection, resulting in a significant drop in the final pose accuracy. To mitigate this issue, we present a complete active vision framework for 6D object pose refinement and next-best-view prediction. Specifically, we first develop an optimization-based pose refinement module for the structured light camera. Our system then selects the next best camera viewpoint to collect depth measurements by minimizing the predicted uncertainty of the object pose. Compared to previous approaches, we additionally predict measurement uncertainties of future viewpoints by online rendering, which significantly improves the next-best-view prediction performance. We test our approach on the challenging real-world ROBI dataset. The results demonstrate that our pose refinement method outperforms the traditional ICP-based approach when given the same input depth data, and our next-best-view strategy can achieve high object pose accuracy with significantly fewer viewpoints than the heuristic-based policies.

cs.RO↗
arXiv2023

GRB 211211A-like Events and How Gravitational Waves May Tell Their Origins

GRB 211211A is a rare burst with a genuinely long duration, yet its prominent kilonova association provides compelling evidence that this peculiar burst was the result of a compact binary merger. However, the exact nature of the merging objects, whether they were neutron star pairs, neutron star--black hole systems, or neutron star--white dwarf systems, remains unsettled. This {\it Letter} delves into the rarity of this event and the possibility of using current and next-generation gravitational wave detectors to distinguish between the various types of binary systems. Our research reveals an event rate density of $\gtrsim 5.67^{+13.04}_{-4.69} \times 10^{-3}\ \rm Gpc^{-3}\ yr^{-1}$ for GRB 211211A-like gamma-ray bursts (GRBs), which, assuming GRB 211211A is the only example of such a burst, is significantly smaller than that of typical long- and short-GRB populations. We further calculated that if the origin of GRB 211211A is a result of a neutron star--black hole merger, it would be detectable with a significant signal-to-noise ratio (S/N), given the LIGO-Virgo-KAGRA designed sensitivity. On the other hand, a neutron star--white dwarf binary would also produce a considerable S/N during the inspiral phase at decihertz and is detectable by next-generation spaceborne detectors DECIGO and the Big Bang Observer. However, to detect this type of system with millihertz spaceborne detectors like LISA, Taiji, and TianQin, the event must be very close, approximately 3 Mpc in distance or smaller.

astro-ph.HE↗
arXiv2023

Machine-learning Kohn-Sham potential from dynamics in time-dependent Kohn-Sham systems

The construction of a better exchange-correlation potential in time-dependent density functional theory (TDDFT) can improve the accuracy of TDDFT calculations and provide more accurate predictions of the properties of many-electron systems. Here, we propose a machine learning method to develop the energy functional and the Kohn-Sham potential of a time-dependent Kohn-Sham system is proposed. The method is based on the dynamics of the Kohn-Sham system and does not require any data on the exact Kohn-Sham potential for training the model. We demonstrate the results of our method with a 1D harmonic oscillator example and a 1D two-electron example. We show that the machine-learned Kohn-Sham potential matches the exact Kohn-Sham potential in the absence of memory effect. Our method can still capture the dynamics of the Kohn-Sham system in the presence of memory effects. The machine learning method developed in this article provides insight into making better approximations of the energy functional and the Kohn-Sham potential in the time-dependent Kohn-Sham system.

quant-ph↗
arXiv2023

META-SELD: Meta-Learning for Fast Adaptation to the new environment in Sound Event Localization and Detection

For learning-based sound event localization and detection (SELD) methods, different acoustic environments in the training and test sets may result in large performance differences in the validation and evaluation stages. Different environments, such as different sizes of rooms, different reverberation times, and different background noise, may be reasons for a learning-based system to fail. On the other hand, acquiring annotated spatial sound event samples, which include onset and offset time stamps, class types of sound events, and direction-of-arrival (DOA) of sound sources is very expensive. In addition, deploying a SELD system in a new environment often poses challenges due to time-consuming training and fine-tuning processes. To address these issues, we propose Meta-SELD, which applies meta-learning methods to achieve fast adaptation to new environments. More specifically, based on Model Agnostic Meta-Learning (MAML), the proposed Meta-SELD aims to find good meta-initialized parameters to adapt to new environments with only a small number of samples and parameter updating iterations. We can then quickly adapt the meta-trained SELD model to unseen environments. Our experiments compare fine-tuning methods from pre-trained SELD models with our Meta-SELD on the Sony-TAU Realistic Spatial Soundscapes 2023 (STARSSS23) dataset. The evaluation results demonstrate the effectiveness of Meta-SELD when adapting to new environments.

eess.AS↗
arXiv2023

The innermost jet in the hidden ultra-luminous X-ray source Cygnus X-3

Cygnus X-3 is a high-mass X-ray binary with a compact object accreting matter from a Wolf-Rayet donor star. Recently, it has been revealed by the Imaging X-ray Polarimetry Explorer (IXPE) as a hidden Galactic ultra-luminous X-ray (ULX) source with a luminosity above the Eddington limit along the direction of a narrow (opening angle <~32 degree) funnel. In between the IXPE observations, we observed Cyg X-3 with the European VLBI (very long baseline interferometry) Network at 22 GHz and the NICER X-ray instrument. To probe possible relations between the X-ray funnel and the potential radio jet from the ULX, we analyzed the simultaneous multi-wavelength data. Our high-resolution VLBI image reveals an elongated structure with a position angle of -3.2+/-0.4 degree, accurately perpendicular to the direction of the linear X-ray polarization. Because Cyg X-3 was in the radio quiescent state on 2022 November 10, we identify the mas-scale structure as the innermost radio jet. The finding indicates that the radio jet propagates along and within the funnel. Moreover, the jet is marginally resolved in the transverse direction. This possibly results from the strong stellar winds and the rapid orbital motion of the binary system.

astro-ph.HE↗
arXiv2023

Variational Label-Correlation Enhancement for Congestion Prediction

The physical design process of large-scale designs is a time-consuming task, often requiring hours to days to complete, with routing being the most critical and complex step. As the the complexity of Integrated Circuits (ICs) increases, there is an increased demand for accurate routing quality prediction. Accurate congestion prediction aids in identifying design flaws early on, thereby accelerating circuit design and conserving resources. Despite the advancements in current congestion prediction methodologies, an essential aspect that has been largely overlooked is the spatial label-correlation between different grids in congestion prediction. The spatial label-correlation is a fundamental characteristic of circuit design, where the congestion status of a grid is not isolated but inherently influenced by the conditions of its neighboring grids. In order to fully exploit the inherent spatial label-correlation between neighboring grids, we propose a novel approach, {\ours}, i.e., VAriational Label-Correlation Enhancement for Congestion Prediction, which considers the local label-correlation in the congestion map, associating the estimated congestion value of each grid with a local label-correlation weight influenced by its surrounding grids. {\ours} leverages variational inference techniques to estimate this weight, thereby enhancing the regression model's performance by incorporating spatial dependencies. Experiment results validate the superior effectiveness of {\ours} on the public available \texttt{ISPD2011} and \texttt{DAC2012} benchmarks using the superblue circuit line.

cs.LG↗
arXiv2023

Uncertainty-driven Trajectory Truncation for Data Augmentation in Offline Reinforcement Learning

Equipped with the trained environmental dynamics, model-based offline reinforcement learning (RL) algorithms can often successfully learn good policies from fixed-sized datasets, even some datasets with poor quality. Unfortunately, however, it can not be guaranteed that the generated samples from the trained dynamics model are reliable (e.g., some synthetic samples may lie outside of the support region of the static dataset). To address this issue, we propose Trajectory Truncation with Uncertainty (TATU), which adaptively truncates the synthetic trajectory if the accumulated uncertainty along the trajectory is too large. We theoretically show the performance bound of TATU to justify its benefits. To empirically show the advantages of TATU, we first combine it with two classical model-based offline RL algorithms, MOPO and COMBO. Furthermore, we integrate TATU with several off-the-shelf model-free offline RL algorithms, e.g., BCQ. Experimental results on the D4RL benchmark show that TATU significantly improves their performance, often by a large margin. Code is available here.

cs.LG↗
arXiv2023

Tuning Stochastic Gradient Algorithms for Statistical Inference via Large-Sample Asymptotics

The tuning of stochastic gradient algorithms (SGAs) for optimization and sampling is often based on heuristics and trial-and-error rather than generalizable theory. We address this theory--practice gap by characterizing the large-sample statistical asymptotics of SGAs via a joint step-size--sample-size scaling limit. We show that iterate averaging with a large fixed step size is robust to the choice of tuning parameters and asymptotically has covariance proportional to that of the MLE sampling distribution. We also prove a Bernstein--von Mises-like theorem to guide tuning, including for generalized posteriors that are robust to model misspecification. Numerical experiments validate our results and recommendations in realistic finite-sample regimes. Our work lays the foundation for a systematic analysis of other stochastic gradient Markov chain Monte Carlo algorithms for a wide range of models.

stat.CO↗
arXiv2023

Low-data deep quantum chemical learning for accurate MP2 and coupled-cluster correlations

Accurate ab-initio prediction of electronic energies is very expensive for macromolecules by explicitly solving post-Hartree-Fock equations. We here exploit the physically justified local correlation feature in compact basis of small molecules, and construct an expressive low-data deep neural network (dNN) model to obtain machine-learned electron correlation energies on par with MP2 and CCSD levels of theory for more complex molecules and different datasets that are not represented in the training set. We show that our dNN-powered model is data efficient and makes highly transferable prediction across alkanes of various lengths, organic molecules with non-covalent and biomolecular interactions, as well as water clusters of different sizes and morphologies. In particular, by training 800 (H$_2$O)$_8$ clusters with the local correlation descriptors, accurate MP2/cc-pVTZ correlation energies up to (H$_2$O)$_{128}$ can be predicted with a small random error within chemical accuracy from exact values, while a majority of prediction deviations are attributed to an intrinsically systematic error. Our results reveal that an extremely compact local correlation feature set, which is poor for any direct post-Hartree-Fock calculations, has however a prominent advantage in reserving important electron correlation patterns for making accurate transferable predictions across distinct molecular compositions, bond types and geometries.

physics.chem-ph↗
arXiv2023

Critical role of vertical radiative cooling contrast in triggering episodic deluges in small-domain hothouse climates

Seeley and Wordsworth (2021) showed that in small-domain cloud-resolving simulations the pattern of precipitation transforms in extremely hot climates ($\ge$ 320 K) from quasi-steady to organized episodic deluges, with outbursts of heavy rain alternating with several dry days. They proposed a mechanism for this transition involving increased water vapor absorption of solar radiation leading to net lower-tropospheric radiative heating. This heating inhibits lower-tropospheric convection and decouples the boundary layer from the upper troposphere during the dry phase, allowing lower-tropospheric moist static energy to build until it discharges, resulting in a deluge. We perform cloud-resolving simulations in polar night and show that the same transition occurs, implying that some revision of their mechanism is necessary. We show that episodic deluges can occur even if the lower-tropospheric radiative heating rate is negative, as long as the magnitude of the upper-tropospheric radiative cooling is about twice as large. We find that in the episodic deluge regime the mean precipitation can be inferred from the atmospheric column energy budget and the period can be predicted from the time for radiation and reevaporation to cool the lower atmosphere.

physics.ao-ph↗
arXiv2023

Cloud Behaviour on Tidally Locked Rocky Planets from Global High-resolution Modeling

Determining the behaviour of convection and clouds is one of the biggest challenges in our understanding of exoplanetary climates. Given the lack of in situ observations, one of the most preferable approaches is to use cloud-resolving or cloud-permitting models (CPM). Here we present CPM simulations in a quasi-global domain with high spatial resolution (4$\times$4 km grid) and explicit convection to study the cloud regime of 1 to 1 tidally locked rocky planets orbiting around low-mass stars. We show that the substellar region is covered by deep convective clouds and cloud albedo increases with increasing stellar flux. The CPM produces relatively less cloud liquid water concentration, smaller cloud coverage, lower cloud albedo, and deeper H2O spectral features than previous general circulation model (GCM) simulations employing empirical convection and cloud parameterizations. Furthermore, cloud streets--long bands of low-level clouds oriented nearly parallel to the direction of the mean boundary-layer winds--appear in the CPM and substantially affect energy balance and surface precipitation at a local level.

astro-ph.EP↗
arXiv2023

TS Cache: A Fast Cache with Timing-speculation Mechanism Under Low Supply Voltages

To mitigate the ever-worsening Power Wall problem, more and more applications need to expand their power supply to the wide-voltage range including the near-threshold region. However, the read delay distribution of the SRAM cells under the near-threshold voltage shows a more serious long-tail characteristic than that under the nominal voltage due to the process fluctuation. Such degradation of SRAM delay makes the SRAM-based cache a performance bottleneck of systems as well. To avoid the unreliable data reading, circuit-level studies use larger/more transistors in a bitcell by scarifying chip area and the static power of cache arrays. Architectural studies propose the auxiliary error correction or block disabling/remapping methods in fault-tolerant caches, which worsen both the hit latency and energy efficiency due to the complex accessing logic. This paper proposes the Timing-Speculation (TS) cache to boost the cache frequency and improve energy efficiency under low supply voltages. In the TS cache, the voltage differences of bitlines are continuously evaluated twice by a sense amplifier (SA), and the access timing error can be detected much earlier than that in prior methods. According to the measurement results from the fabricated chips, the TS L1 cache aggressively increases its frequency to 1.62X and 1.92X compared with the conventional scheme at 0.5V and 0.6V supply voltages, respectively.

cs.AR↗
arXiv2023

Metasurfaces for free-space coupling to multicore fibers

Space-division multiplexing (SDM) with multicore fibers (MCFs) is envisioned to overcome the capacity crunch in optical fiber communications. Within these systems, the coupling optics that connect single-mode fibers (SMFs) to MCFs are key components in achieving high data transfer rates. Designing a compact and scalable coupler with low loss and crosstalk is a continuing challenge. Here, we introduce a metasurface-based free-space coupler that can be designed for any input array of SMFs to a MCF with arbitrary core layout. An inverse design technique - adjoint method - optimizes the metasurface phase profiles to maximize the overlap of the output fields to the MCF modes at each core position. As proof-of-concepts, we fabricated two types of 4-mode couplers for MCFs with linear and square core arrays. The measured insertion losses were as low as 1.2 dB and the worst-case crosstalk was less than -40.1 dB in the O-band (1260-1360 nm). Owing to its foundry-compatible fabrication, this coupler design could facilitate the widespread deployment of SDM based on MCFs.

physics.optics↗
arXiv2023

Von Neumann Dimensions and Trace Formulas I: Limit Multiplicities

Given a connected semisimple Lie group $G$ and an arithmetic subgroup $Γ$, it is well-known that each irreducible representation $π$ of $G$ occurs in the discrete spectrum $L^2_{\text{disc}}(Γ\backslash G)$ of $L^2(Γ\backslash G)$ with at most a finite multiplicity $m_Γ(π)$. While $m_Γ(π)$ is unknown in general, we are interested in its limit as $Γ$ is taken to be in a tower of lattices $Γ_1\supset Γ_2\supset\dots$. For a bounded measurable subset $X$ of the unitary dual $\widehat{G}$, we let $m_{Γ_n}(X)$ be the sum of the multiplicity $m_{Γ_n}(π)$ of a representation $π$ over all $π$ in $X$. Let $H_X$ be the direct integral of the irreducible representations in $X$, which is also a module over the group von Neumann algebra $\mathcal{L}Γ_n$. We prove: \begin{center} $\lim\limits_{n\to \infty}\cfrac{m_{Γ_n}(X)}{\dim_{\mathcal{L}Γ_n}H_X}=1$, \end{center} for any bounded subset $X$ of $\widehat{G}$, when i) $Γ_n$'s are cocompact, or, ii) $G=\SL(n,\mathbb{R})$ and $\{Γ_n\}$ are principal congruence subgroups.

math.RT↗
arXiv2023

Rethinking PRL: A Multiscale Progressively Residual Learning Network for Inverse Halftoning

Image inverse halftoning is a classic image restoration task, aiming to recover continuous-tone images from halftone images with only bilevel pixels. Because the halftone images lose much of the original image content, inverse halftoning is a classic ill-problem. Although existing inverse halftoning algorithms achieve good performance, their results lose image details and features. Therefore, it is still a challenge to recover high-quality continuous-tone images. In this paper, we propose an end-to-end multiscale progressively residual learning network (MSPRL), which has a UNet architecture and takes multiscale input images. To make full use of different input image information, we design a shallow feature extraction module to capture similar features between images of different scales. We systematically study the performance of different methods and compare them with our proposed method. In addition, we employ different training strategies to optimize the model, which is important for optimizing the training process and improving performance. Extensive experiments demonstrate that our MSPRL model obtains considerable performance gains in detail restoration.

cs.CV↗