Engineering Papers⌕ Search

Engineering topics

Nugent, Peter

Publications and source records attributed to Nugent, Peter.

At least 19 records

Identifying Transient Candidates in the Dark Energy Survey Using Convolutional Neural Networks

The ability to discover new transient candidates via image differencing without direct human intervention is an important task in observational astronomy. For these kind of image classification problems, machine learning techniques such as Convolutional Neural Networks (CNNs) have shown remarkable success. In this work, we present the results of an automated transient candidate identification on images with CNNs for an extant data set from the Dark Energy Survey Supernova program, whose main focus was on using Type Ia supernovae for cosmology. By performing an architecture search of CNNs, we identify networks that efficiently select non-artifacts (e.g., supernovae, variable stars, AGN, etc.) from artifacts (image defects, mis-subtractions, etc.), achieving the efficiency of previous work performed with random Forests, without the need to expend any effort in feature identification. The CNNs also help us identify a subset of mislabeled images. Performing a relabeling of the images in this subset, the resulting classification with CNNs is significantly better than previous results, lowering the false positive rate by 27% at a fixed missed detection rate of 0.05.

79 ASTRONOMY AND ASTROPHYSICS↗

Variability-selected Intermediate-mass Black Hole Candidates in Dwarf Galaxies from ZTF and WISE

While it is difficult to observe the first black hole seeds in the early universe, we can study intermediate-mass black holes (IMBHs) in local dwarf galaxies for clues about their origins. In this paper we present a sample of variability-selected active galactic nuclei (AGN) in dwarf galaxies using optical photometry from the Zwicky Transient Facility (ZTF) and forward-modeled mid-IR photometry of time-resolved Wide-field Infrared Survey Explorer (WISE) co-added images. We found that 44 out of 25,714 dwarf galaxies had optically variable AGN candidates and 148 out of 79,879 dwarf galaxies had mid-IR variable AGN candidates, corresponding to active fractions of 0.17% ± 0.03% and 0.19% ± 0.02%, respectively. We found that spectroscopic approaches to AGN identification would have missed 81% of our ZTF IMBH candidates and 69% of our WISE IMBH candidates. Only nine candidates have been detected previously in radio, X-ray, and variability searches for dwarf galaxy AGN. The ZTF and WISE dwarf galaxy AGN with broad Balmer lines have virial masses of 10 5 M ⊙ < M BH < 10 7 M ⊙ , but for the rest of the sample, BH masses predicted from host galaxy mass range between 10 5.2 M ⊙ < M BH < 10 7.25 M ⊙ . We found that only 5 of 152 previously reported variability-selected AGN candidates from the Palomar Transient Factory in common with our parent sample were variable in ZTF. We also determined a nuclear supernova fraction of 0.05% ± 0.01% yr -1 for dwarf galaxies in ZTF. Our ZTF and WISE IMBH candidates show the promise of variability searches for the discovery of otherwise hidden low-mass AGN.

79 ASTRONOMY AND ASTROPHYSICS↗

Spatial Graph Attention and Curiosity-driven Policy for Antiviral Drug Discovery

We developed Distilled Graph Attention Policy Network (DGAPN), a reinforcement learning model to generate novel graph-structured chemical representations that optimize user-defined objectives by efficiently navigating a physically constrained domain. The framework is examined on the task of generating molecules that are designed to bind, noncovalently, to functional sites of SARS-CoV-2 proteins. We present a spatial Graph Attention (sGAT) mechanism that leverages self-attention over both node and edge attributes as well as encoding the spatial structure --- this capability is of considerable interest in synthetic biology and drug discovery. An attentional policy network is introduced to learn the decision rules for a dynamic, fragment-based chemical environment, and state-of-the-art policy gradient techniques are employed to train the network with stability. Exploration is driven by the stochasticity of the action space design and the innovation reward bonuses learned and proposed by random network distillation. In experiments, our framework achieved outstanding results compared to state-of-the-art algorithms, while reducing the complexity of paths to chemical synthesis.

Wu, Yulun↗

Spatial Graph Attention and Curiosity-driven Policy for Antiviral Drug Discovery

We developed Distilled Graph Attention Policy Networks (DGAPNs), a curiosity-driven reinforcement learning model to generate novel graph-structured chemical representations that optimize user-defined objectives by efficiently navigating a physically constrained domain. The framework is examined on the task of generating molecules that are designed to bind, noncovalently, to functional sites of SARS-CoV-2 proteins. We present a spatial Graph Attention Network (sGAT) that leverages self-attention over both node and edge attributes as well as encoding spatial structure -- this capability is of considerable interest in areas such as molecular and synthetic biology and drug discovery. An attentional policy network is then introduced to learn decision rules for a dynamic, fragment-based chemical environment, and state-of-the-art policy gradient techniques are employed to train the network with enhanced stability. Exploration is efficiently encouraged by incorporating innovation reward bonuses learned and proposed by random network distillation. In experiments, our framework achieved outstanding results compared to state-of-the-art algorithms, while increasing the diversity of proposed molecules and reducing the complexity of paths to chemical synthesis.

Wu, Yulun↗

2019 Computing Sciences Strategic Plan

Computing has transformed nearly every aspect of scientific inquiry — across disciplines and across scales — from the behavior of subatomic particles to the formation of structures in the early universe, from the assembly of the human genome to the evolution of earth systems. Over the past two decades, computing has become an integral part of how Berkeley Lab is “Bringing Science Solutions to the World.” Advances in computing and mathematics have been key, with new mathematical models of complex physical phenomena, new methods for analyzing complex data, new algorithms for accuracy and scaling and sophisticated software systems that encapsulate these techniques into open, reusable tools. The performance of NERSC computers and the ESnet network have grown by several orders of magnitude, along with our understanding of how to map scientific computations and workflows onto these systems. From research to facility operations, the passion, talent and dedication of the Computing Sciences Area staff has been the cornerstone of our success. The plan outlined in this document describes the next step in a journey to expand the influence and impact of our efforts, building an increasingly connected global enterprise for science that places more powerful instruments in the hands of scientists, along with more powerful methods and tools for modeling, analysis and prediction.

97 MATHEMATICS AND COMPUTING↗

Co-design Center for Exascale Machine Learning Technologies (ExaLearn)

We report rapid growth in data, computational methods, and computing power is driving a remarkable revolution in what variously is termed machine learning (ML), statistical learning, computational learning, and artificial intelligence. In addition to highly visible successes in machine-based natural language translation, playing the game Go, and self-driving cars, these new technologies also have profound implications for computational and experimental science and engineering, as well as for the exascale computing systems that the Department of Energy (DOE) is developing to support those disciplines. Not only do these learning technologies open up exciting opportunities for scientific discovery on exascale systems, they also appear poised to have important implications for the design and use of exascale computers themselves, including high-performance computing (HPC) for ML and ML for HPC. The overarching goal of the ExaLearn co-design project is to provide exascale ML software for use by Exascale Computing Project (ECP) applications, other ECP co-design centers, and DOE experimental facilities and leadership class computing facilities.

97 MATHEMATICS AND COMPUTING↗

Removing Atmospheric Fringes from Zwicky Transient Facility i-band Images using Principal Component Analysis

The Zwicky Transient Facility is a time-domain optical survey that has substantially increased our ability to observe and construct massive catalogs of astronomical objects by use of its 47 square degree camera that can observe in multiple filters. However the telescope’s i-band filter suffers from significant atmospheric fringes that reduce photometric precision, especially for faint sources and in multi-epoch co-additions. Here we present a method for constructing models of these atmospheric fringes using Principal Component Analysis that can be used to identify and remove these artifacts from contaminated images. In addition, we present the Uniform Background Indicator as a quantitative measurement of the reduced correlated background noise and photometric error present after removing fringes. We conclude by evaluating the effect of our method on measuring faint sources through the injection and recovery of artificial stars in both single-image epochs and co-additions. Our method for constructing atmospheric fringe models and applying those models to produce cleaned images is available for public download in the open source Python package fringez.

79 ASTRONOMY AND ASTROPHYSICS↗

Spatial Graph Attention and Curiosity-driven Policy for Antiviral Drug Discovery

We developed Distilled Graph Attention Policy Networks (DGAPNs), a curiosity-driven reinforcement learning model to generate novel graph-structured chemical representations that optimize user-defined objectives by efficiently navigating a physically constrained domain. Here, the framework is examined on the task of generating molecules that are designed to bind, noncovalently, to functional sites of SARS-CoV-2 proteins. We present a spatial Graph Attention Network (sGAT) that leverages self-attention over both node and edge attributes as well as encoding spatial structure -- this capability is of considerable interest in areas such as molecular and synthetic biology and drug discovery. An attentional policy network is then introduced to learn decision rules for a dynamic, fragment-based chemical environment, and state-of-the-art policy gradient techniques are employed to train the network with enhanced stability. Exploration is efficiently encouraged by incorporating innovation reward bonuses learned and proposed by random network distillation. In experiments, our framework achieved outstanding results compared to state-of-the-art algorithms, while increasing the diversity of proposed molecules and reducing the complexity of paths to chemical synthesis.

Wu, Yulun↗

AGNs on the Move: A Search for Off-nuclear AGNs from Recoiling Supermassive Black Holes and Ongoing Galaxy Mergers with the Zwicky Transient Facility

A supermassive black hole (SMBH) ejected from the potential well of its host galaxy via gravitational wave recoil carries important information about the mass ratio and spin alignment of the pre-merger SMBH binary. Such a recoiling SMBH may be detectable as an active galactic nucleus (AGN) broad-line region offset by up to 10 kpc from a disturbed host galaxy. We describe a novel methodology using forward modeling with The Tractor to search for such offset AGNs in a sample of 5493 optically variable AGNs detected with the Zwicky Transient Facility (ZTF). We present the discovery of nine AGNs that may be spatially offset from their host galaxies and are candidates for recoiling SMBHs. Five of these offset AGNs exhibit double-peaked broad Balmer lines, which may have arisen from unobscured accretion disk emission, and four show radio emission indicative of a relativistic jet. The fraction of double-peaked emitters in our spatially offset AGN sample is significantly larger than the 16% double-peaked emitter fraction observed for ZTF AGNs overall. In our sample of variable AGNs we also identified 52 merging galaxies, including a new spectroscopically confirmed dual AGN. Finally, we detected the dramatic rebrightening of SDSS 1133, a previously discovered variable object and recoiling SMBH candidate, in ZTF. The flare was accompanied by the reemergence of strong P Cygni line features, indicating that SDSS 1133 may be an outbursting luminous blue variable star.

79 ASTRONOMY AND ASTROPHYSICS↗

Nebular Models of Sub-Chandrasekhar Mass Type Ia Supernovae: Clues to the Origin of Ca-rich Transients

In this work we use non-local thermal equilibrium radiative transport modeling to examine observational signatures of sub-Chandrasekhar mass double detonation explosions in the nebular phase. Results range from spectra that look like typical and subluminous Type Ia supernovae (SNe) for higher mass progenitors to spectra that look like Ca-rich transients for lower mass progenitors. This ignition mechanism produces an inherent relationship between emission features and the progenitor mass as the ratio of the nebular [Ca II ]/[Fe III ] emission lines increases with decreasing white dwarf mass. Examining the [Ca II ]/[Fe III ] nebular line ratio in a sample of observed SNe we find further evidence for the two distinct classes of SNe Ia identified in Polin et al. by their relationship between Si ii velocity and B-band magnitude, both at time of peak brightness. This suggests that SNe Ia arise from more than one progenitor channel, and provides an empirical method for classifying events based on their physical origin. Furthermore, we provide insight to the mysterious origin of Ca-rich transients. Low-mass double detonation models with only a small mass fraction of Ca (1%) produce nebular spectra that cool primarily through forbidden [Ca II ] emission.

79 ASTRONOMY AND ASTROPHYSICS↗

The Case for Strong Scaling in Deep Learning: Training Large 3D CNNs with Hybrid Parallelism

Here, we present scalable hybrid-parallel algorithms for training large-scale 3D convolutional neural networks. Deep learning-based emerging scientific workflows often require model training with large, high-dimensional samples, which can make training much more costly and even infeasible due to excessive memory usage. We solve these challenges by extensively applying hybrid parallelism throughout the end-to-end training pipeline, including both computations and I/O. Our hybrid-parallel algorithm extends the standard data parallelism with spatial parallelism, which partitions a single sample in the spatial domain, realizing strong scaling beyond the mini-batch dimension with a larger aggregated memory capacity. We evaluate our proposed training algorithms with two challenging 3D CNNs, CosmoFlow and 3D U-Net. Our comprehensive performance studies show that good weak and strong scaling can be achieved for both networks using up to 2K GPUs. More importantly, we enable training of CosmoFlow with much larger samples than previously possible, realizing an order-of-magnitude improvement in prediction accuracy.

97 MATHEMATICS AND COMPUTING↗

SN 2018fif: The Explosion of a Large Red Supergiant Discovered in Its Infancy by the Zwicky Transient Facility

High-cadence transient surveys are able to capture supernovae closer to their first light than ever before. Applying analytical models to such early emission, we can constrain the progenitor stars' properties. Here, we present observations of SN 2018fif (ZTF 18abokyfk). The supernova was discovered close to first light and monitored by the Zwicky Transient Facility (ZTF) and the Neil Gehrels Swift Observatory. Early spectroscopic observations suggest that the progenitor of SN 2018fif was surrounded by relatively small amounts of circumstellar material compared to all previous cases. This particularity, coupled with the high-cadence multiple-band coverage, makes it a good candidate to investigate using shock-cooling models. We employ the SOPRANOS code, an implementation of the model by Sapir & Waxman and its extension to early times by Morag et al. Compared with previous implementations, SOPRANOS has the advantage of including a careful account of the limited temporal validity domain of the shock-cooling model as well as allowing usage of the entirety of the early UV data. We find that the progenitor of SN 2018fif was a large red supergiant with a radius of $R={744.0}_{-128.0}^{+183.0}\,{R}_{\odot }$ and an ejected mass of ${M}_{\mathrm{ej}}={9.3}_{-5.8}^{+0.4}\,{M}_{\odot }$. Our model also gives information on the explosion epoch, the progenitor's inner structure, the shock velocity, and the extinction. The distribution of radii is double-peaked, with smaller radii corresponding to lower values of the extinction, earlier recombination times, and a better match to the early UV data. If these correlations persist in future objects, denser spectroscopic monitoring constraining the time of recombination, as well as accurate UV observations (e.g., with ULTRASAT), will help break the extinction/radius degeneracy and independently determine both.

79 ASTRONOMY AND ASTROPHYSICS↗

Gravitational Microlensing Event Statistics for the Zwicky Transient Facility

Microlensing surveys have discovered thousands of events, with almost all events discovered within the Galactic bulge or toward the Magellanic Clouds. The Zwicky Transient Facility (ZTF), while not designed to be a microlensing campaign, is an optical time-domain survey that observes the entire northern sky, including the Galactic plane, every few nights. The ZTF observes ~10 9 stars in the g and r bands and can significantly contribute to the observed microlensing population. In this paper, we predict that the ZTF will observe ~1100 microlensing events in 3 yr of observing within 10° latitude of the Galactic plane, with ~500 events in the outer Galaxy (ℓ ≥ 10°). This yield increases to ~1400 (~800) events by combining every three ZTF exposures, ~1800 (~900) events if the ZTF observes for a total of 5 yr, and ~2400 (~1300) events for a 5 yr survey with postprocessing image stacking. Using the microlensing modeling software PopSyCLE, we compare the microlensing populations in the Galactic bulge and the outer Galaxy. We also present an analysis of the microlensing event ZTF18abhxjmj to demonstrate how to leverage these population statistics in event modeling. The ZTF will constrain Galactic structure, stellar populations, and primordial black holes through photometric microlensing.

79 ASTRONOMY AND ASTROPHYSICS↗

HPC-enabled computation of demand models at scale

The purpose of this project is to examine the energy impact of urban-scale traffic for the Los Angeles Basin by developing and implementing a scalable traffic assignment model. An energy optimization function will be posed and when integrated into the optimization code for travel assignment it can be mathematically proven to converge. The energy optimization function can then be compared to the typical travel time optimization that is traditionally used in traffic assignment models. The analysis will begin with static traffic assignment models with the routing for all origin and destinations computed in parallel on high performance computing facilities. Convergence of the numerical methods rely on the solution of convex programs (or extensions of these). This step will mostly consist of demonstrating the ability to parallelize the Frank Wolfe algorithm on various platforms.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

HPC4Mobilty w/ UCB

The purpose of this project is to examine the energy impact of urban-scale traffic for the Los Angeles Basin by developing and implementing a scalable traffic assignment model. An energy optimization function will be posed and when integrated into the optimization code for travel assignment it can be mathematically proven to converge. The energy optimization function can then be compared to the typical travel time optimization that is traditionally used in traffic assignment models. The analysis will begin with static traffic assignment models with the routing for all origin and destinations computed in parallel on high performance computing facilities. Convergence of the numerical methods rely on the solution of convex programs (or extensions of these). This step will mostly consist of demonstrating the ability to parallelize the Frank Wolfe algorithm on various platforms. This work will contribute to LBNL’s efforts to develop new processes, analytical tools, program designs, and business models to advance the state of the art in next-generation sustainable transportation solutions.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗