arXiv subjects

Haochen Wang

Publications and source records attributed to Haochen Wang.

107 records · Page 6Linked to original sources

arXiv2023

CHIME/FRB Discovery of 25 Repeating Fast Radio Burst Sources

We present the discovery of 25 new repeating fast radio burst (FRB) sources found among CHIME/FRB events detected between 2019 September 30 and 2021 May 1. The sources were found using a new clustering algorithm that looks for multiple events co-located on the sky having similar dispersion measures (DMs). The new repeaters have DMs ranging from $\sim$220 pc cm$^{-3}$ to $\sim$1700 pc cm$^{-3}$, and include sources having exhibited as few as two bursts to as many as twelve. We report a statistically significant difference in both the DM and extragalactic DM (eDM) distributions between repeating and apparently nonrepeating sources, with repeaters having lower mean DM and eDM, and we discuss the implications. We find no clear bimodality between the repetition rates of repeaters and upper limits on repetition from apparently nonrepeating sources after correcting for sensitivity and exposure effects, although some active repeating sources stand out as anomalous. We measure the repeater fraction over time and find that it tends to an equilibrium of $2.6_{-2.6}^{+2.9}$% over our total time-on-sky thus far. We also report on 14 more sources which are promising repeating FRB candidates and which merit follow-up observations for confirmation.

astro-ph.HE↗

arXiv2023

The First CHIME/FRB Fast Radio Burst Catalog

We present a catalog of 536 fast radio bursts (FRBs) detected by the Canadian Hydrogen Intensity Mapping Experiment Fast Radio Burst (CHIME/FRB) Project between 400 and 800 MHz from 2018 July 25 to 2019 July 1, including 62 bursts from 18 previously reported repeating sources. The catalog represents the first large sample, including bursts from repeaters and non-repeaters, observed in a single survey with uniform selection effects. This facilitates comparative and absolute studies of the FRB population. We show that repeaters and apparent non-repeaters have sky locations and dispersion measures (DMs) that are consistent with being drawn from the same distribution. However, bursts from repeating sources differ from apparent non-repeaters in intrinsic temporal width and spectral bandwidth. Through injection of simulated events into our detection pipeline, we perform an absolute calibration of selection effects to account for systematic biases. We find evidence for a population of FRBs - comprising a large fraction of the overall population - with a scattering time at 600 MHz in excess of 10 ms, of which only a small fraction are observed by CHIME/FRB. We infer a power-law index for the cumulative fluence distribution of $α=-1.40\pm0.11(\textrm{stat.})^{+0.06}_{-0.09}(\textrm{sys.})$, consistent with the $-3/2$ expectation for a non-evolving population in Euclidean space. We find $α$ is steeper for high-DM events and shallower for low-DM events, which is what would be expected when DM is correlated with distance. We infer a sky rate of $[525\pm30(\textrm{stat.})^{+140}_{-130}({\textrm{sys.}})]/\textrm{sky}/\textrm{day}$ above a fluence of 5 Jy ms at 600 MHz, with scattering time at $600$ MHz under 10 ms, and DM above 100 pc cm$^{-3}$.

astro-ph.HE↗

arXiv2022

Score Jacobian Chaining: Lifting Pretrained 2D Diffusion Models for 3D Generation

A diffusion model learns to predict a vector field of gradients. We propose to apply chain rule on the learned gradients, and back-propagate the score of a diffusion model through the Jacobian of a differentiable renderer, which we instantiate to be a voxel radiance field. This setup aggregates 2D scores at multiple camera viewpoints into a 3D score, and repurposes a pretrained 2D model for 3D data generation. We identify a technical challenge of distribution mismatch that arises in this application, and propose a novel estimation mechanism to resolve it. We run our algorithm on several off-the-shelf diffusion image generative models, including the recently released Stable Diffusion trained on the large-scale LAION dataset.

cs.CV↗

arXiv2022

Characterization of the John A. Galt telescope for radio holography with CHIME

The Canadian Hydrogen Intensity Mapping Experiment (CHIME) will measure the 21 cm emission of astrophysical neutral hydrogen to probe large scale structure at redshifts z=0.8-2.5. However, detecting the 21 cm signal beneath substantially brighter foregrounds remains a key challenge. Due to the high dynamic range between 21 cm and foreground emission, an exquisite calibration of instrument systematics, notably the telescope beam, is required to successfully filter out the foregrounds. One technique being used to achieve a high fidelity measurement of the CHIME beam is radio holography, wherein signals from each of CHIME's analog inputs are correlated with the signal from a co-located reference antenna, the 26 m John A. Galt telescope, as the 26 m Galt telescope tracks a bright point source transiting over CHIME. In this work we present an analysis of several of the Galt telescope's properties. We employ driftscan measurements of several bright sources, along with background estimates derived from the 408 MHz Haslam map, to estimate the Galt system temperature. To determine the Galt telescope's beam shape, we perform and analyze a raster scan of the bright radio source Cassiopeia A. Finally, we use early holographic measurements to measure the Galt telescope's geometry with respect to CHIME for the holographic analysis of the CHIME and Galt interferometric data set.

astro-ph.IM↗

arXiv2022

Learning from Future: A Novel Self-Training Framework for Semantic Segmentation

Self-training has shown great potential in semi-supervised learning. Its core idea is to use the model learned on labeled data to generate pseudo-labels for unlabeled samples, and in turn teach itself. To obtain valid supervision, active attempts typically employ a momentum teacher for pseudo-label prediction yet observe the confirmation bias issue, where the incorrect predictions may provide wrong supervision signals and get accumulated in the training process. The primary cause of such a drawback is that the prevailing self-training framework acts as guiding the current state with previous knowledge, because the teacher is updated with the past student only. To alleviate this problem, we propose a novel self-training strategy, which allows the model to learn from the future. Concretely, at each training step, we first virtually optimize the student (i.e., caching the gradients without applying them to the model weights), then update the teacher with the virtual future student, and finally ask the teacher to produce pseudo-labels for the current student as the guidance. In this way, we manage to improve the quality of pseudo-labels and thus boost the performance. We also develop two variants of our future-self-training (FST) framework through peeping at the future both deeply (FST-D) and widely (FST-W). Taking the tasks of unsupervised domain adaptive semantic segmentation and semi-supervised semantic segmentation as the instances, we experimentally demonstrate the effectiveness and superiority of our approach under a wide range of settings. Code will be made publicly available.

cs.CV↗

arXiv2022

An Overview of CHIME, the Canadian Hydrogen Intensity Mapping Experiment

The Canadian Hydrogen Intensity Mapping Experiment (CHIME) is a drift scan radio telescope operating across the 400-800 MHz band. CHIME is located at the Dominion Radio Astrophysical Observatory near Penticton, BC Canada. The instrument is designed to map neutral hydrogen over the redshift range 0.8 to 2.5 to constrain the expansion history of the Universe. This goal drives the design features of the instrument. CHIME consists of four parallel cylindrical reflectors, oriented north-south, each 100 m $\times$ 20 m and outfitted with a 256 element dual-polarization linear feed array. CHIME observes a two degree wide stripe covering the entire meridian at any given moment, observing 3/4 of the sky every day due to Earth rotation. An FX correlator utilizes FPGAs and GPUs to digitize and correlate the signals, with different correlation products generated for cosmological, fast radio burst, pulsar, VLBI, and 21 cm absorber backends. For the cosmology backend, the $N_\mathrm{feed}^2$ correlation matrix is formed for 1024 frequency channels across the band every 31 ms. A data receiver system applies calibration and flagging and, for our primary cosmological data product, stacks redundant baselines and integrates for 10 s. We present an overview of the instrument, its performance metrics based on the first three years of science data, and we describe the current progress in characterizing CHIME's primary beam response. We also present maps of the sky derived from CHIME data; we are using versions of these maps for a cosmological stacking analysis as well as for investigation of Galactic foregrounds.

astro-ph.IM↗

arXiv2022

Using the Sun to Measure the Primary Beam Response of the Canadian Hydrogen Intensity Mapping Experiment

We present a beam pattern measurement of the Canadian Hydrogen Intensity Mapping Experiment (CHIME) made using the Sun as a calibration source. As CHIME is a pure drift scan instrument, we rely on the seasonal North-South motion of the Sun to probe the beam at different elevations. This semiannual range in elevation, combined with the radio brightness of the Sun, enables a beam measurement which spans ~7,200 square degrees on the sky without the need to move the telescope. We take advantage of observations made near solar minimum to minimize the impact of solar variability, which is observed to be <10% in intensity over the observation period. The resulting data set is highly complementary to other CHIME beam measurements -- both in terms of angular coverage and systematics -- and plays an important role in the ongoing program to characterize the CHIME primary beam.

astro-ph.IM↗

arXiv2022

NFormer: Robust Person Re-identification with Neighbor Transformer

Person re-identification aims to retrieve persons in highly varying settings across different cameras and scenarios, in which robust and discriminative representation learning is crucial. Most research considers learning representations from single images, ignoring any potential interactions between them. However, due to the high intra-identity variations, ignoring such interactions typically leads to outlier features. To tackle this issue, we propose a Neighbor Transformer Network, or NFormer, which explicitly models interactions across all input images, thus suppressing outlier features and leading to more robust representations overall. As modelling interactions between enormous amount of images is a massive task with lots of distractors, NFormer introduces two novel modules, the Landmark Agent Attention, and the Reciprocal Neighbor Softmax. Specifically, the Landmark Agent Attention efficiently models the relation map between images by a low-rank factorization with a few landmarks in feature space. Moreover, the Reciprocal Neighbor Softmax achieves sparse attention to relevant -- rather than all -- neighbors only, which alleviates interference of irrelevant representations and further relieves the computational burden. In experiments on four large-scale datasets, NFormer achieves a new state-of-the-art. The code is released at \url{https://github.com/haochenheheda/NFormer}.

cs.CV↗

arXiv2022

Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels

The crux of semi-supervised semantic segmentation is to assign adequate pseudo-labels to the pixels of unlabeled images. A common practice is to select the highly confident predictions as the pseudo ground-truth, but it leads to a problem that most pixels may be left unused due to their unreliability. We argue that every pixel matters to the model training, even its prediction is ambiguous. Intuitively, an unreliable prediction may get confused among the top classes (i.e., those with the highest probabilities), however, it should be confident about the pixel not belonging to the remaining classes. Hence, such a pixel can be convincingly treated as a negative sample to those most unlikely categories. Based on this insight, we develop an effective pipeline to make sufficient use of unlabeled data. Concretely, we separate reliable and unreliable pixels via the entropy of predictions, push each unreliable pixel to a category-wise queue that consists of negative samples, and manage to train the model with all candidate pixels. Considering the training evolution, where the prediction becomes more and more accurate, we adaptively adjust the threshold for the reliable-unreliable partition. Experimental results on various benchmarks and training settings demonstrate the superiority of our approach over the state-of-the-art alternatives.

cs.CV↗

arXiv2022

Removing systematics-induced 21-cm foreground residuals by cross-correlating filtered data

Observations of the redshifted 21-cm signal emitted by neutral hydrogen represent a promising probe of large-scale structure in the universe. However, cosmological 21-cm signal is challenging to observe due to astrophysical foregrounds which are several orders of magnitude brighter. Traditional linear foreground removal methods can optimally remove foregrounds for a known telescope response but are sensitive to telescope systematic errors such as antenna gain and delay errors, leaving foreground contamination in the recovered signal. Non-linear methods such as principal component analysis, on the other hand, have been used successfully for foreground removal, but they lead to signal loss that is difficult to characterize and requires careful analysis. In this paper, we present a systematics-robust foreground removal technique which combines both linear and non-linear methods. We first obtain signal and foreground estimates using a linear filter. Under the assumption that the signal estimate is contaminated by foreground residuals induced by parameterizable systematic effects, we infer the systematics-induced contamination by cross-correlating the initial signal and foreground estimates. Correcting for the inferred error, we are able to subtract foreground contamination from the linearly filtered signal up to the first order in the amplitude of the telescope systematics. In simulations of an interferometric 21-cm survey, our algorithm removes foreground leakage induced by complex gain errors by one to two orders of magnitude in the power spectrum. Our technique thus eases the requirements on telescope characterization for modern and next-generation 21-cm cosmology experiments.

astro-ph.CO↗

arXiv2022

Decoupled IoU Regression for Object Detection

Non-maximum suppression (NMS) is widely used in object detection pipelines for removing duplicated bounding boxes. The inconsistency between the confidence for NMS and the real localization confidence seriously affects detection performance. Prior works propose to predict Intersection-over-Union (IoU) between bounding boxes and corresponding ground-truths to improve NMS, while accurately predicting IoU is still a challenging problem. We argue that the complex definition of IoU and feature misalignment make it difficult to predict IoU accurately. In this paper, we propose a novel Decoupled IoU Regression (DIR) model to handle these problems. The proposed DIR decouples the traditional localization confidence metric IoU into two new metrics, Purity and Integrity. Purity reflects the proportion of the object area in the detected bounding box, and Integrity refers to the completeness of the detected object area. Separately predicting Purity and Integrity can divide the complex mapping between the bounding box and its IoU into two clearer mappings and model them independently. In addition, a simple but effective feature realignment approach is also introduced to make the IoU regressor work in a hindsight manner, which can make the target mapping more stable. The proposed DIR can be conveniently integrated with existing two-stage detectors and significantly improve their performance. Through a simple implementation of DIR with HTC, we obtain 51.3% AP on MS COCO benchmark, which outperforms previous methods and achieves state-of-the-art.

cs.CV↗

arXiv2022

Detection of Cosmological 21 cm Emission with the Canadian Hydrogen Intensity Mapping Experiment

We present a detection of 21-cm emission from large-scale structure (LSS) between redshift 0.78 and 1.43 made with the Canadian Hydrogen Intensity Mapping Experiment (CHIME). Radio observations acquired over 102 nights are used to construct maps which are foreground filtered and stacked on the angular and spectral locations of luminous red galaxies (LRG), emission line galaxies (ELG), and quasars (QSO) from the eBOSS clustering catalogs. We find decisive evidence for a detection when stacking on all three tracers of LSS, with the logarithm of the Bayes Factor equal to 18.9 (LRG), 10.8 (ELG), and 56.3 (QSO). An alternative frequentist interpretation, based on the likelihood-ratio test, yields a detection significance of $7.1σ$ (LRG), $5.7σ$ (ELG), and $11.1σ$ (QSO). These are the first 21-cm intensity mapping measurements made with an interferometer. We constrain the effective clustering amplitude of neutral hydrogen (HI), defined as $\mathcal{A}_{\rm HI}\equiv 10^{3}\,Ω_\mathrm{HI}\left(b_\mathrm{HI}+\langle\,fμ^{2}\rangle\right)$, where $Ω_\mathrm{HI}$ is the cosmic abundance of HI, $b_\mathrm{HI}$ is the linear bias of HI, and $\langle\,fμ^{2}\rangle=0.552$ encodes the effect of redshift-space distortions at linear order. We find $\mathcal{A}_\mathrm{HI}=1.51^{+3.60}_{-0.97}$ for LRGs $(z=0.84)$, $\mathcal{A}_\mathrm{HI}=6.76^{+9.04}_{-3.79}$ for ELGs $(z=0.96)$, and $\mathcal{A}_\mathrm{HI}=1.68^{+1.10}_{-0.67}$ for QSOs $(z=1.20)$, with constraints limited by modeling uncertainties at nonlinear scales. We are also sensitive to bias in the spectroscopic redshifts of each tracer, and find a non-zero bias $Δ\,v= -66 \pm 20 \mathrm{km/s}$ for the QSOs. We split the QSO catalog into three redshift bins and have a decisive detection in each, with the upper bin at $z=1.30$ producing the highest redshift 21-cm intensity mapping measurement thus far.

astro-ph.CO↗

arXiv2021

Piezoelectric effect and polarization switching in Al$_{1-x}$Sc$_x$N

Aluminum nitride is piezoelectric and exhibits spontaneous polarization along the $c$-axis, but the polarization cannot be switched by applying an electric field. Adding Sc to AlN enhances the piezoelectric properties, and can make the alloy ferroelectric. We perform a detailed first-principles analysis of spontaneous and piezoelectric polarization. Comparisons between explicit supercell calculations show that the virtual crystal approximation produces accurate results for polarization, but falls short in describing the phase stability of the alloy. We relate the behavior of the piezoelectric constant $e_{33}$ to the microscopic behavior of the internal displacement parameter $u$, finding that the internal strain contribution dominates in the Sc-induced enhancement. The value of $u$ increases with scandium concentration, bringing the alloy locally closer to a layered hexagonal structure. Our approach allows us to calculate the ferroelectric switching barrier, which we analyze as a function of Sc concentration and temperature based on Ginzburg-Landau theory.

cond-mat.mtrl-sci↗

arXiv2021

SwiftNet: Real-time Video Object Segmentation

In this work we present SwiftNet for real-time semisupervised video object segmentation (one-shot VOS), which reports 77.8% J &F and 70 FPS on DAVIS 2017 validation dataset, leading all present solutions in overall accuracy and speed performance. We achieve this by elaborately compressing spatiotemporal redundancy in matching-based VOS via Pixel-Adaptive Memory (PAM). Temporally, PAM adaptively triggers memory updates on frames where objects display noteworthy inter-frame variations. Spatially, PAM selectively performs memory update and match on dynamic pixels while ignoring the static ones, significantly reducing redundant computations wasted on segmentation-irrelevant pixels. To promote efficient reference encoding, light-aggregation encoder is also introduced in SwiftNet deploying reversed sub-pixel. We hope SwiftNet could set a strong and efficient baseline for real-time VOS and facilitate its application in mobile vision. The source code of SwiftNet can be found at https://github.com/haochenheheda/SwiftNet.

cs.CV↗

arXiv2020

Pixel Consensus Voting for Panoptic Segmentation

The core of our approach, Pixel Consensus Voting, is a framework for instance segmentation based on the Generalized Hough transform. Pixels cast discretized, probabilistic votes for the likely regions that contain instance centroids. At the detected peaks that emerge in the voting heatmap, backprojection is applied to collect pixels and produce instance masks. Unlike a sliding window detector that densely enumerates object proposals, our method detects instances as a result of the consensus among pixel-wise votes. We implement vote aggregation and backprojection using native operators of a convolutional neural network. The discretization of centroid voting reduces the training of instance segmentation to pixel labeling, analogous and complementary to FCN-style semantic segmentation, leading to an efficient and unified architecture that jointly models things and stuff. We demonstrate the effectiveness of our pipeline on COCO and Cityscapes Panoptic Segmentation and obtain competitive results. Code will be open-sourced.

cs.CV↗

arXiv2019

DIODE: A Dense Indoor and Outdoor DEpth Dataset

We introduce DIODE, a dataset that contains thousands of diverse high resolution color images with accurate, dense, long-range depth measurements. DIODE (Dense Indoor/Outdoor DEpth) is the first public dataset to include RGBD images of indoor and outdoor scenes obtained with one sensor suite. This is in contrast to existing datasets that focus on just one domain/scene type and employ different sensors, making generalization across domains difficult. The dataset is available for download at http://diode-dataset.org

cs.CV↗

arXiv2019

Bayesian cross validation for gravitational-wave searches in pulsar-timing array data

Gravitational-wave data analysis demands sophisticated statistical noise models in a bid to extract highly obscured signals from data. In Bayesian model comparison, we choose among a landscape of models by comparing their marginal likelihoods. However, this computation is numerically fraught and can be sensitive to arbitrary choices in the specification of parameter priors. In Bayesian cross validation, we characterize the fit and predictive power of a model by computing the Bayesian posterior of its parameters in a training dataset, and then use that posterior to compute the averaged likelihood of a different testing dataset. The resulting cross-validation scores are straightforward to compute; they are insensitive to prior tuning; and they penalize unnecessarily complex models that overfit the training data at the expense of predictive performance. In this article, we discuss cross validation in the context of pulsar-timing-array data analysis, and we exemplify its application to simulated pulsar data (where it successfully selects the correct spectral index of a stochastic gravitational-wave background), and to a pulsar dataset from the NANOGrav 11-year release (where it convincingly favors a model that represents a transient feature in the interstellar medium). We argue that cross validation offers a promising alternative to Bayesian model comparison, and we discuss its use for gravitational-wave detection, by selecting or refuting models that include a gravitational-wave component.

astro-ph.IM↗