arXiv subjects

Bin Ren

Publications and source records attributed to Bin Ren.

At least 145 records · Page 8Linked to original sources

arXiv2020

Early High-contrast Imaging Results with Keck/NIRC2-PWFS: The SR 21 Disk

High-contrast imaging of exoplanets and protoplanetary disks depends on wavefront sensing and correction made by adaptive optics instruments. Classically, wavefront sensing has been conducted at optical wavelengths, which made high-contrast imaging of red targets such as M-type stars or extincted T Tauri stars challenging. Keck/NIRC2 has combined near-infrared (NIR) detector technology with the pyramid wavefront sensor (PWFS). With this new module we observed SR~21, a young star that is brighter at NIR wavelengths than at optical wavelengths. Compared with the archival data of SR~21 taken with the optical wavefront sensing we achieved $\sim$20\% better Strehl ratio in similar natural seeing conditions. Further post-processing utilizing angular differential imaging and reference-star differential imaging confirmed the spiral feature reported by the VLT/SPHERE polarimetric observation, which is the first detection of the SR~21 spiral in total intensity at $L^\prime$ band. We also compared the contrast limit of our result ($10^{-4}$ at $0\farcs4$ and $2\times10^{-5}$ at $1\farcs0$) with the archival data that were taken with optical wavefront sensing and confirmed the improvement, particularly at $\leq0\farcs5$. Our observation demonstrates that the NIR PWFS improves AO performance and will provide more opportunities for red targets in the future.

astro-ph.EP↗

arXiv2020

Real-Time Execution of Large-scale Language Models on Mobile

Pre-trained large-scale language models have increasingly demonstrated high accuracy on many natural language processing (NLP) tasks. However, the limited weight storage and computational speed on hardware platforms have impeded the popularity of pre-trained models, especially in the era of edge computing. In this paper, we seek to find the best model structure of BERT for a given computation size to match specific devices. We propose the first compiler-aware neural architecture optimization framework. Our framework can guarantee the identified model to meet both resource and real-time specifications of mobile devices, thus achieving real-time execution of large transformer-based models like BERT variants. We evaluate our model on several NLP tasks, achieving competitive results on well-known benchmarks with lower latency on mobile devices. Specifically, our model is 5.2x faster on CPU and 4.1x faster on GPU with 0.5-2% accuracy loss compared with BERT-base. Our overall framework achieves up to 7.8x speedup compared with TensorFlow-Lite with only minor accuracy loss.

cs.CL↗

arXiv2020

On Efficient Constructions of Checkpoints

Efficient construction of checkpoints/snapshots is a critical tool for training and diagnosing deep learning models. In this paper, we propose a lossy compression scheme for checkpoint constructions (called LC-Checkpoint). LC-Checkpoint simultaneously maximizes the compression rate and optimizes the recovery speed, under the assumption that SGD is used to train the model. LC-Checkpointuses quantization and priority promotion to store the most crucial information for SGD to recover, and then uses a Huffman coding to leverage the non-uniform distribution of the gradient scales. Our extensive experiments show that LC-Checkpoint achieves a compression rate up to $28\times$ and recovery speedup up to $5.77\times$ over a state-of-the-art algorithm (SCAR).

cs.LG↗

arXiv2020

A Privacy-Preserving-Oriented DNN Pruning and Mobile Acceleration Framework

Weight pruning of deep neural networks (DNNs) has been proposed to satisfy the limited storage and computing capability of mobile edge devices. However, previous pruning methods mainly focus on reducing the model size and/or improving performance without considering the privacy of user data. To mitigate this concern, we propose a privacy-preserving-oriented pruning and mobile acceleration framework that does not require the private training dataset. At the algorithm level of the proposed framework, a systematic weight pruning technique based on the alternating direction method of multipliers (ADMM) is designed to iteratively solve the pattern-based pruning problem for each layer with randomly generated synthetic data. In addition, corresponding optimizations at the compiler level are leveraged for inference accelerations on devices. With the proposed framework, users could avoid the time-consuming pruning process for non-experts and directly benefit from compressed models. Experimental results show that the proposed framework outperforms three state-of-art end-to-end DNN frameworks, i.e., TensorFlow-Lite, TVM, and MNN, with speedup up to 4.2X, 2.5X, and 2.0X, respectively, with almost no accuracy loss, while preserving data privacy.

cs.LG↗

arXiv2020

Petascale XCT: 3D Image Reconstruction with Hierarchical Communications on Multi-GPU Nodes

X-ray computed tomography is a commonly used technique for noninvasive imaging at synchrotron facilities. Iterative tomographic reconstruction algorithms are often preferred for recovering high quality 3D volumetric images from 2D X-ray images, however, their use has been limited to small/medium datasets due to their computational requirements. In this paper, we propose a high-performance iterative reconstruction system for terabyte(s)-scale 3D volumes. Our design involves three novel optimizations: (1) optimization of (back)projection operators by extending the 2D memory-centric approach to 3D; (2) performing hierarchical communications by exploiting "fat-node" architecture with many GPUs; (3) utilization of mixed-precision types while preserving convergence rate and quality. We extensively evaluate the proposed optimizations and scaling on the Summit supercomputer. Our largest reconstruction is a mouse brain volume with 9Kx11Kx11K voxels, where the total reconstruction time is under three minutes using 24,576 GPUs, reaching 65 PFLOPS: 34% of Summit's peak performance.

cs.DC↗

arXiv2020

Dynamical Evidence of a Spiral Arm--Driving Planet in the MWC 758 Protoplanetary Disk

More than a dozen young stars host spiral arms in their surrounding protoplanetary disks. The excitation mechanisms of such arms are under debate. The two leading hypotheses -- companion-disk interaction and gravitational instability (GI) -- predict distinct motion for spirals. By imaging the MWC 758 spiral arm system at two epochs spanning ${\sim}5$ yr using the SPHERE instrument on the Very Large Telescope (VLT), we test the two hypotheses for the first time. We find that the pattern speeds of the spirals are not consistent with the GI origin. Our measurements further evince the existence of a faint "missing planet" driving the disk arms. The average spiral pattern speed is $0.\!^\circ22\pm0.\!^\circ03$ yr$^{-1}$, pointing to a driver at $172_{-14}^{+18}$ au around a $1.9$ $M_\odot$ central star if it is on a circular orbit. In addition, we witness time varying shadowing effects on a global scale that are likely originated from an inner disk.

astro-ph.EP↗

arXiv2020

An Image Enhancing Pattern-based Sparsity for Real-time Inference on Mobile Devices

Weight pruning has been widely acknowledged as a straightforward and effective method to eliminate redundancy in Deep Neural Networks (DNN), thereby achieving acceleration on various platforms. However, most of the pruning techniques are essentially trade-offs between model accuracy and regularity which lead to impaired inference accuracy and limited on-device acceleration performance. To solve the problem, we introduce a new sparsity dimension, namely pattern-based sparsity that comprises pattern and connectivity sparsity, and becoming both highly accurate and hardware friendly. With carefully designed patterns, the proposed pruning unprecedentedly and consistently achieves accuracy enhancement and better feature extraction ability on different DNN structures and datasets, and our pattern-aware pruning framework also achieves pattern library extraction, pattern selection, pattern and connectivity pruning and weight training simultaneously. Our approach on the new pattern-based sparsity naturally fits into compiler optimization for highly efficient DNN execution on mobile platforms. To the best of our knowledge, it is the first time that mobile devices achieve real-time inference for the large-scale DNN models thanks to the unique spatial property of pattern-based sparsity and the help of the code generation capability of compilers.

cs.CV↗

arXiv2020

Multiband GPI Imaging of the HR 4796A Debris Disk

We have obtained Gemini Planet Imager (GPI) J-, H-, K1-, and K2-Spec observations of the iconic debris ring around the young, main-sequence star HR 4796A. We applied several point-spread function (PSF) subtraction techniques to the observations (Mask-and-Interpolate, RDI-NMF, RDI-KLIP, and ADI-KLIP) to measure the geometric parameters and the scattering phase function for the disk. To understand the systematic errors associated with PSF subtraction, we also forward-modeled the observations using a Markov Chain Monte Carlo framework and a simple model for the disk. We found that measurements of the disk geometric parameters were robust, with all of our analyses yielding consistent results; however, measurements of the scattering phase function were challenging to reconstruct from PSF-subtracted images, despite extensive testing. As a result, we estimated the scattering phase function using disk modeling. We searched for a dependence of the scattering phase function with respect to the GPI filters but found none. We compared the H-band scattering phase function with that measured by Hubble Space Telescope STIS at visual wavelengths and discovered a blue color at small scattering angles and a red color at large scattering angles, consistent with predictions and laboratory measurements of large grains. Finally, we successfully modeled the SPHERE H2 HR 4796A scattered phase function using a distribution of hollow spheres composed of silicates, carbon, and metallic iron.

astro-ph.SR↗

arXiv2020

Debris Disk Results from the Gemini Planet Imager Exoplanet Survey's Polarimetric Imaging Campaign

We report the results of a ${\sim}4$-year direct imaging survey of 104 stars to resolve and characterize circumstellar debris disks in scattered light as part of the Gemini Planet Imager Exoplanet Survey. We targeted nearby (${\lesssim}150$ pc), young (${\lesssim}500$ Myr) stars with high infrared excesses ($L_{\mathrm{IR}} / L_\star > 10^{-5}$), including 38 with previously resolved disks. Observations were made using the Gemini Planet Imager high-contrast integral field spectrograph in $H$-band (1.6 $μ$m) coronagraphic polarimetry mode to measure both polarized and total intensities. We resolved 26 debris disks and three protoplanetary/transitional disks. Seven debris disks were resolved in scattered light for the first time, including newly presented HD 117214 and HD 156623, and we quantified basic morphologies of five of them using radiative transfer models. All of our detected debris disks but HD 156623 have dust-poor inner holes, and their scattered-light radii are generally larger than corresponding radii measured from resolved thermal emission and those inferred from spectral energy distributions. To assess sensitivity, we report contrasts and consider causes of non-detections. Detections were strongly correlated with high IR excess and high inclination, although polarimetry outperformed total intensity angular differential imaging for detecting low inclination disks (${\lesssim} 70 °$). Based on post-survey statistics, we improved upon our pre-survey target prioritization metric predicting polarimetric disk detectability. We also examined scattered-light disks in the contexts of gas, far-IR, and millimeter detections. Comparing $H$-band and ALMA fluxes for two disks revealed tentative evidence for differing grain properties. Finally, we found no preference for debris disks to be detected in scattered light if wide-separation substellar companions were present.

astro-ph.EP↗

arXiv2020

Keck/NIRC2 $L$'-Band Imaging of Jovian-Mass Accreting Protoplanets around PDS 70

We present $L$'-band imaging of the PDS 70 planetary system with Keck/NIRC2 using the new infrared pyramid wavefront sensor. We detected both PDS 70 b and c in our images, as well as the front rim of the circumstellar disk. After subtracting off a model of the disk, we measured the astrometry and photometry of both planets. Placing priors based on the dynamics of the system, we estimated PDS 70 b to have a semi-major axis of $20^{+3}_{-4}$~au and PDS 70 c to have a semi-major axis of $34^{+12}_{-6}$~au (95\% credible interval). We fit the spectral energy distribution (SED) of both planets. For PDS 70 b, we were able to place better constraints on the red half of its SED than previous studies and inferred the radius of the photosphere to be 2-3~$R_{Jup}$. The SED of PDS 70 c is less well constrained, with a range of total luminosities spanning an order of magnitude. With our inferred radii and luminosities, we used evolutionary models of accreting protoplanets to derive a mass of PDS 70 b between 2 and 4 $M_{\textrm{Jup}}$ and a mean mass accretion rate between $3 \times 10^{-7}$ and $8 \times 10^{-7}~M_{\textrm{Jup}}/\textrm{yr}$. For PDS 70 c, we computed a mass between 1 and 3 $M_{\textrm{Jup}}$ and mean mass accretion rate between $1 \times 10^{-7}$ and $5 \times~10^{-7} M_{\textrm{Jup}}/\textrm{yr}$. The mass accretion rates imply dust accretion timescales short enough to hide strong molecular absorption features in both planets' SEDs.

astro-ph.EP↗

arXiv2020

CoCoPIE: Making Mobile AI Sweet As PIE --Compression-Compilation Co-Design Goes a Long Way

Assuming hardware is the major constraint for enabling real-time mobile intelligence, the industry has mainly dedicated their efforts to developing specialized hardware accelerators for machine learning and inference. This article challenges the assumption. By drawing on a recent real-time AI optimization framework CoCoPIE, it maintains that with effective compression-compiler co-design, it is possible to enable real-time artificial intelligence on mainstream end devices without special hardware. CoCoPIE is a software framework that holds numerous records on mobile AI: the first framework that supports all main kinds of DNNs, from CNNs to RNNs, transformer, language models, and so on; the fastest DNN pruning and acceleration framework, up to 180X faster compared with current DNN pruning on other frameworks such as TensorFlow-Lite; making many representative AI applications able to run in real-time on off-the-shelf mobile devices that have been previously regarded possible only with special hardware support; making off-the-shelf mobile devices outperform a number of representative ASIC and FPGA solutions in terms of energy efficiency and/or performance.

cs.LG↗

arXiv2020

Towards Real-Time DNN Inference on Mobile Platforms with Model Pruning and Compiler Optimization

High-end mobile platforms rapidly serve as primary computing devices for a wide range of Deep Neural Network (DNN) applications. However, the constrained computation and storage resources on these devices still pose significant challenges for real-time DNN inference executions. To address this problem, we propose a set of hardware-friendly structured model pruning and compiler optimization techniques to accelerate DNN executions on mobile devices. This demo shows that these optimizations can enable real-time mobile execution of multiple DNN applications, including style transfer, DNN coloring and super resolution.

cs.LG↗

arXiv2020

The Gemini Planet Imager view of the HD 32297 debris disk

We present new $H$-band scattered light images of the HD 32297 edge-on debris disk obtained with the Gemini Planet Imager (GPI). The disk is detected in total and polarized intensity down to a projected angular separation of 0.15", or 20au. On the other hand, the large scale swept-back halo remains undetected, likely a consequence of its markedly blue color relative to the parent body belt. We analyze the curvature of the disk spine and estimate a radius of $\approx$100au for the parent body belt, smaller than past scattered light studies but consistent with thermal emission maps of the system. We employ three different flux-preserving post-processing methods to suppress the residual starlight and evaluate the surface brightness and polarization profile along the disk spine. Unlike past studies of the system, our high fidelity images reveal the disk to be highly symmetric and devoid of morphological and surface brightness perturbations. We find the dust scattering properties of the system to be consistent with those observed in other debris disks, with the exception of HR 4796. Finally, we find no direct evidence for the presence of a planetary-mass object in the system.

astro-ph.SR↗

arXiv2020

HD 165054: an astrometric calibration field for high-contrast imagers in Baade's Window

We present a study of the HD 165054 astrometric calibration field that has been periodically observed with the Gemini Planet Imager. HD 165054 is a bright star within Baade's Window, a region of the galactic plane with relatively low extinction from interstellar dust. HD 165054 was selected as a calibrator target due to the high number density of stars within this region ($\sim 3$ stars per square arcsecond with $H<22$), necessary because of the small field-of-view of the Gemini Planet Imager. Using nine epochs spanning over five years, we have fit a standard five-parameter astrometric model to the astrometry of seven background stars within close proximity to HD 165054 (angular separation $< 2$ arcsec). We achieved a proper motion precision of $\sim 0.3$ mas/yr, and constrained the parallax of each star to be $\lesssim 1$ mas. Our measured proper motions and parallax limits are consistent with the background stars being a part of the galactic bulge. Using these measurements we find no evidence of any systematic trend of either the plate scale or the north angle offset of GPI between 2014 and 2019. We compared our model describing the motions of the seven background stars to observations of the same field in 2014 and 2018 obtained with Keck/NIRC2, an instrument with an excellent astrometric calibration. We find that predicted position of the background sources is consistent with that measured by NIRC2, within the uncertainties of the calibration of the two instruments. In the future, we will use this field as a standard astrometric calibrator for the upgrade of GPI and potentially for other high-contrast imagers.

astro-ph.IM↗

arXiv2020

Using Data Imputation for Signal Separation in High Contrast Imaging

To characterize circumstellar systems in high contrast imaging, the fundamental step is to construct a best point spread function (PSF) template for the non-circumstellar signals (i.e., star light and speckles) and separate it from the observation. With existing PSF construction methods, the circumstellar signals (e.g., planets, circumstellar disks) are unavoidably altered by over-fitting and/or self-subtraction, making forward modeling a necessity to recover these signals. We present a forward modeling--free solution to these problems with data imputation using sequential non-negative matrix factorization (DI-sNMF). DI-sNMF first converts this signal separation problem to a "missing data" problem in statistics by flagging the regions which host circumstellar signals as missing data, then attributes PSF signals to these regions. We mathematically prove it to have negligible alteration to circumstellar signals when the imputation region is relatively small, which thus enables precise measurement for these circumstellar objects. We apply it to simulated point source and circumstellar disk observations to demonstrate its proper recovery of them. We apply it to Gemini Planet Imager (GPI) K1-band observations of the debris disk surrounding HR 4796A, finding a tentative trend that the dust is more forward scattering as the wavelength increases. We expect DI-sNMF to be applicable to other general scenarios where the separation of signals is needed.

astro-ph.IM↗

arXiv2020

PCONV: The Missing but Desirable Sparsity in DNN Weight Pruning for Real-time Execution on Mobile Devices

Model compression techniques on Deep Neural Network (DNN) have been widely acknowledged as an effective way to achieve acceleration on a variety of platforms, and DNN weight pruning is a straightforward and effective method. There are currently two mainstreams of pruning methods representing two extremes of pruning regularity: non-structured, fine-grained pruning can achieve high sparsity and accuracy, but is not hardware friendly; structured, coarse-grained pruning exploits hardware-efficient structures in pruning, but suffers from accuracy drop when the pruning rate is high. In this paper, we introduce PCONV, comprising a new sparsity dimension, -- fine-grained pruning patterns inside the coarse-grained structures. PCONV comprises two types of sparsities, Sparse Convolution Patterns (SCP) which is generated from intra-convolution kernel pruning and connectivity sparsity generated from inter-convolution kernel pruning. Essentially, SCP enhances accuracy due to its special vision properties, and connectivity sparsity increases pruning rate while maintaining balanced workload on filter computation. To deploy PCONV, we develop a novel compiler-assisted DNN inference framework and execute PCONV models in real-time without accuracy compromise, which cannot be achieved in prior work. Our experimental results show that, PCONV outperforms three state-of-art end-to-end DNN frameworks, TensorFlow-Lite, TVM, and Alibaba Mobile Neural Network with speedup up to 39.2x, 11.4x, and 6.3x, respectively, with no accuracy loss. Mobile devices can achieve real-time inference on large-scale DNNs.

cs.LG↗

arXiv2020

BLK-REW: A Unified Block-based DNN Pruning Framework using Reweighted Regularization Method

Accelerating DNN execution on various resource-limited computing platforms has been a long-standing problem. Prior works utilize l1-based group lasso or dynamic regularization such as ADMM to perform structured pruning on DNN models to leverage the parallel computing architectures. However, both of the pruning dimensions and pruning methods lack universality, which leads to degraded performance and limited applicability. To solve the problem, we propose a new block-based pruning framework that comprises a general and flexible structured pruning dimension as well as a powerful and efficient reweighted regularization method. Our framework is universal, which can be applied to both CNNs and RNNs, implying complete support for the two major kinds of computation-intensive layers (i.e., CONV and FC layers). To complete all aspects of the pruning-for-acceleration task, we also integrate compiler-based code optimization into our framework that can perform DNN inference in a real-time manner. To the best of our knowledge, it is the first time that the weight pruning framework achieves universal coverage for both CNNs and RNNs with real-time mobile acceleration and no accuracy compromise.

cs.LG↗

arXiv2020

RTMobile: Beyond Real-Time Mobile Acceleration of RNNs for Speech Recognition

Recurrent neural networks (RNNs) based automatic speech recognition has nowadays become prevalent on mobile devices such as smart phones. However, previous RNN compression techniques either suffer from hardware performance overhead due to irregularity or significant accuracy loss due to the preserved regularity for hardware friendliness. In this work, we propose RTMobile that leverages both a novel block-based pruning approach and compiler optimizations to accelerate RNN inference on mobile devices. Our proposed RTMobile is the first work that can achieve real-time RNN inference on mobile platforms. Experimental results demonstrate that RTMobile can significantly outperform existing RNN hardware acceleration methods in terms of inference accuracy and time. Compared with prior work on FPGA, RTMobile using Adreno 640 embedded GPU on GRU can improve the energy-efficiency by about 40$\times$ while maintaining the same inference time.

cs.SD↗