arXiv subjects

Andrew Howard

Publications and source records attributed to Andrew Howard.

At least 19 recordsLinked to original sources

arXiv2026

Gaia Exoplanet Orbits, Demographics, and Evolution Survey (GEODES): Characteristics of Three Long-Period Companions Accelerating their Host Stars

The upcoming release of $Gaia$ DR4 will yield thousands of giant planet candidates, eventually enabling studies of giant planet eccentricities, masses, and occurrence rates across a broad range of stellar host masses, metallicities, and ages. However, some of these planet candidates are expected to be false positives, and even genuine detections will require additional observations to precisely determine their orbits and masses. We present here the first results of the $Gaia$ Exoplanet Orbits, Demographics, and Evolution Survey (GEODES), an observational campaign to identify the most promising planet candidate hosts for pre-DR4 vetting and post-DR4 validation and characterization. In this paper we showcase three systems from our broader sample exhibiting both tangential and radial accelerations, each representing a distinct outcome of our survey strategy. We combine $Hipparcos$, $Hipparcos$-$Gaia$, $Gaia$ DR2, and $Gaia$ DR3 absolute astrometry with adaptive optics (AO) imaging and precision RVs to constrain companion masses and orbits. HIP 18512, a nearby (15.3 pc) K4V dwarf, hosts a low-mass stellar companion at 10.87" $\pm$ 0.07" (166 AU) which produces significant RV and astrometric accelerations on its host star. The RV trend and astrometric acceleration of the nearby (24.2 pc) K4V star HIP 45839, together with an AO imaging non-detection, constrain the companion to $a$ = 17.9^{+4.8}_{-2.7} AU ($P$ = 70--127 years) and $M$ = 45.2^{+10.5}_{-12.7} $M_{Jup}$. In the case of HIP 81991 (43.8 pc, G5V), the astrometric and RV data indicate that the companion has a separation of 6.4^{+0.6}_{-0.3} AU ($P$ = 14.4--17.7 years) and a mass of 9.5^{+5.4}_{-2.2} $M_{Jup}$, and is more likely a planet (65%) than a brown dwarf (35%).

astro-ph.EP↗

arXiv2025

A Smooth Transition from Giant Planets to Brown Dwarfs from the Radial Occurrence Distribution

Measuring the occurrence rates of celestial objects is a valuable way to study their origins and evolution. Giant planets and brown dwarfs produce large Doppler signatures that are easily detectable by modern instrumentation, and legacy radial velocity (RV) surveys have now achieved full orbital coverage for periods $\lesssim$30 years. However, the Doppler method's sensitivity to companion minimum mass $M_c \sin i$ -- as opposed to true mass $M_c$ -- prevents unambiguous characterization using RVs alone, as purported giant planets may be brown dwarfs or stars on inclined orbits. Here we combined legacy RVs with absolute astrometry to re-fit the orbits of 195 companions from the California Legacy Survey. Nearly 50% (8/18) of the ``brown dwarfs" ($M_c \sin i$=13--80 $M_{Jup}$) we refit had true masses above 80 $M_{Jup}$. We incorporated our orbital posteriors and target sensitivity maps into a Poisson likelihood model to calculate occurrence as a function of true companion mass $M_c$ (0.8--80 $M_{Jup}$) and separation $a$ (0.3--30 AU). The semi-major axis distributions of objects in this range vary smoothly with mass, with Jupiter analogs favoring an abrupt increase in occurrence near 1 AU and brown dwarfs exhibiting a gradual enhancement at wider separations. Marginalized companion occurrence between 1--10 AU decreases smoothly with mass, with brown dwarfs having the lowest occurrence rate: $0.9^{+0.5}_{-0.4}$%. Jupiter analogs are 10 times as common as brown dwarfs per mass interval in this range, demonstrating that the brown dwarf desert extends to 10 AU. The smooth variation in these distributions disfavors a sharp transition mass between ``bottom-up" core accretion and ``top-down" gravitational instability formation mechanisms, and rather suggests that these processes may produce companions in overlapping mass ranges.

astro-ph.EP↗

arXiv2025

The OATMEAL Survey. II. The 3D spin-orbit obliquity of an eccentric transiting brown dwarf in the Ruprecht 147 open cluster

We present new analysis of the CWW 89 system as part of the Orbital Architectures of Transiting Massive Exoplanets And Low-mass stars (OATMEAL) survey. The CWW 89 system is a member of the 2.8 Gyr old Ruprecht 147 (NGC 6774) cluster and features two stars, CWW 89A (EPIC 219388192) and CWW 89B, with the primary hosting a transiting brown dwarf. We use in-transit, highly precise radial velocity measurements with the Keck Planet Finder (KPF) to characterize the Rossiter-McLaughlin (RM) effect and measure the projected spin-orbit obliquity $|\lambda|=1.4\pm2.5^\circ$ and the full 3D spin-orbit obliquity of the brown dwarf to be $\psi=15.1^{+15.0^\circ}_{-10.9}$. This value of $\lambda$ implies that the brown dwarf's orbit is prograde and well-aligned with the equator of the host star, continuing the trend of transiting brown dwarfs showing a preference for alignment ($\lambda \approx 0^\circ$) regardless of the stellar effective temperature. We find that this contrast with the transiting giant planet population, whose spin-orbit alignments depend on host $T_{\rm eff}$, shows an increasingly clear distinction in the formation and orbital migration mechanisms between transiting giant planets and transiting brown dwarfs like CWW 89Ab. For this system in particular, we find it plausible that the brown dwarf may have undergone coplanar high-eccentricity migration influence by CWW 89B.

astro-ph.EP↗

arXiv2024

Robust Training of Neural Networks at Arbitrary Precision and Sparsity

The discontinuous operations inherent in quantization and sparsification introduce a long-standing obstacle to backpropagation, particularly in ultra-low precision and sparse regimes. While the community has long viewed quantization as unfriendly to gradient descent due to its lack of smoothness, we pinpoint-for the first time-that the key issue is the absence of a proper gradient path that allows training to learn robustness to quantization noise. The standard Straight-Through Estimator (STE) exacerbates this with its well-understood mismatch: a quantization-aware forward pass but oblivious backward pass, leading to unmanaged error and instability. We solve this by explicitly modeling quantization as additive noise, making the full forward-backward path well-defined without heuristic gradient estimation. As one natural solution, we introduce a denoising dequantization transform derived from a principled ridge regression objective, creating an explicit, corrective gradient path that makes learning robust to the noise STE bypasses. We extend this to sparsification by treating it as a special form of quantization that zeros out small values. Our unified framework trains models at arbitrary precisions and sparsity levels with off-the-shelf recipes, enabling stable A1W1 and sub-1-bit networks where others falter. It yields state-of-the-art results, mapping efficiency frontiers for modern LLMs and providing a theoretically grounded path to hyper-efficient neural networks.

cs.LG↗

arXiv2024

Custom Gradient Estimators are Straight-Through Estimators in Disguise

Quantization-aware training comes with a fundamental challenge: the derivative of quantization functions such as rounding are zero almost everywhere and nonexistent elsewhere. Various differentiable approximations of quantization functions have been proposed to address this issue. In this paper, we prove that when the learning rate is sufficiently small, a large class of weight gradient estimators is equivalent with the straight through estimator (STE). Specifically, after swapping in the STE and adjusting both the weight initialization and the learning rate in SGD, the model will train in almost exactly the same way as it did with the original gradient estimator. Moreover, we show that for adaptive learning rate algorithms like Adam, the same result can be seen without any modifications to the weight initialization and learning rate. We experimentally show that these results hold for both a small convolutional model trained on the MNIST dataset and for a ResNet50 model trained on ImageNet.

cs.LG↗

arXiv2024

MobileNetV4 -- Universal Models for the Mobile Ecosystem

We present the latest generation of MobileNets, known as MobileNetV4 (MNv4), featuring universally efficient architecture designs for mobile devices. At its core, we introduce the Universal Inverted Bottleneck (UIB) search block, a unified and flexible structure that merges Inverted Bottleneck (IB), ConvNext, Feed Forward Network (FFN), and a novel Extra Depthwise (ExtraDW) variant. Alongside UIB, we present Mobile MQA, an attention block tailored for mobile accelerators, delivering a significant 39% speedup. An optimized neural architecture search (NAS) recipe is also introduced which improves MNv4 search effectiveness. The integration of UIB, Mobile MQA and the refined NAS recipe results in a new suite of MNv4 models that are mostly Pareto optimal across mobile CPUs, DSPs, GPUs, as well as specialized accelerators like Apple Neural Engine and Google Pixel EdgeTPU - a characteristic not found in any other models tested. Finally, to further boost accuracy, we introduce a novel distillation technique. Enhanced by this technique, our MNv4-Hybrid-Large model delivers 87% ImageNet-1K accuracy, with a Pixel 8 EdgeTPU runtime of just 3.8ms.

cs.CV↗

arXiv2023

ESPRESSO observations of Gaia BH1: high-precision orbital constraints and no evidence for an inner binary

We present high-precision radial velocity (RV) observations of Gaia BH1, the nearest known black hole (BH). The system contains a solar-type G star orbiting a massive dark companion, which could be either a single BH or an inner BH + BH binary. A BH + BH binary is expected in some models where Gaia BH1 formed as a hierarchical triple, which are attractive because they avoid many of the difficulties associated with forming the system through isolated binary evolution. Our observations test the inner binary scenario. We have measured 115 precise RVs of the G star, including 40 from ESPRESSO with a precision of $3$-$5$ m s$^{-1}$, and 75 from other instruments with a typical precision of $30$-$100$ m s$^{-1}$. Our observations span $2.33$ orbits of the G star and are concentrated near a periastron passage, when perturbations due to an inner binary would be largest. The RVs are well-fit by a Keplerian two-body orbit and show no convincing evidence of an inner binary. Using REBOUND simulations of hierarchical triples with a range of inner periods, mass ratios, eccentricities, and orientations, we show that plausible inner binaries with periods $P_{\text{inner}} \gtrsim 1.5$ days would have produced larger deviations from a Keplerian orbit than observed. Binaries with $P_{\text{inner}} \lesssim 1.5$ days are consistent with the data, but these would merge within a Hubble time and would thus imply fine-tuning. We present updated parameters of Gaia BH1's orbit. The RVs yield a spectroscopic mass function $f\left(M_{\text{BH}}\right)=3.9358 \pm 0.0002\,M_{\odot}$ - about $7000\sigma$ above the $\sim2.5\,M_{\odot}$ maximum neutron star mass. Including the inclination constraint from Gaia astrometry, this implies a BH mass of $M_{\text{BH}} = 9.27 \pm 0.10 ~ M_{\odot}$.

astro-ph.SR↗

arXiv2023

Hot Jupiters Have Giant Companions: Evidence for Coplanar High-Eccentricity Migration

This study considers the characteristics of planetary systems with giant planets based on a population-level analysis of the California Legacy Survey planet catalog. We identified three characteristics common to hot Jupiters. First, while not all hot Jupiters have a detected outer giant planet companion ($M \sin i$ = 0.3--30 $M_{\textrm{Jup}}$), such companions are ubiquitous when survey completeness corrections are applied for orbital periods out to 40,000 days. Giant harboring systems without a hot Jupiter also host at least one outer giant planet companion per system. Second, the mass distributions of hot Jupiters and other giant planets are indistinguishable. However, within a planetary system that includes a hot Jupiter, the outer giant planet companions are at least $3\times$ more massive than the inner hot Jupiters. Third, the eccentricity distribution of the outer companions in hot Jupiter systems (with an average model eccentricity of $\langle e\rangle=0.34\pm0.05$) is different from the corresponding outer planets in planetary systems without hot Jupiters ($\langle e\rangle=0.19\pm0.02$). We conclude that the existence of two gas giants, where the outermost planet has an eccentricity $\ge0.2$ and is $3\times$ more massive, are key factors in the production of a hot Jupiter. Our simple model based on these factors predicts that $\sim$10\% of warm and cold Jupiter systems will by chance meet these assembly criteria, which is consistent with our measurement of $16\pm6\%$ relative occurrence of hot Jupiter systems to all giant-harboring systems. We find that these three features favor coplanar high-eccentricity migration as the dominant mechanism for hot Jupiter formation.

astro-ph.EP↗

arXiv2023

The development of HISPEC for Keck and MODHIS for TMT: science cases and predicted sensitivities

HISPEC is a new, high-resolution near-infrared spectrograph being designed for the W.M. Keck II telescope. By offering single-shot, R=100,000 between 0.98 - 2.5 um, HISPEC will enable spectroscopy of transiting and non-transiting exoplanets in close orbits, direct high-contrast detection and spectroscopy of spatially separated substellar companions, and exoplanet dynamical mass and orbit measurements using precision radial velocity monitoring calibrated with a suite of state-of-the-art absolute and relative wavelength references. MODHIS is the counterpart to HISPEC for the Thirty Meter Telescope and is being developed in parallel with similar scientific goals. In this proceeding, we provide a brief overview of the current design of both instruments, and the requirements for the two spectrographs as guided by the scientific goals for each. We then outline the current science case for HISPEC and MODHIS, with focuses on the science enabled for exoplanet discovery and characterization. We also provide updated sensitivity curves for both instruments, in terms of both signal-to-noise ratio and predicted radial velocity precision.

astro-ph.IM↗

arXiv2023

Data-Driven Approaches to Searches for the Technosignatures of Advanced Civilizations

Humanity has wondered whether we are alone for millennia. The discovery of life elsewhere in the Universe, particularly intelligent life, would have profound effects, comparable to those of recognizing that the Earth is not the center of the Universe and that humans evolved from previous species. There has been rapid growth in the fields of extrasolar planets and data-driven astronomy. In a relatively short interval, we have seen a change from knowing of no extrasolar planets to now knowing more potentially habitable extrasolar planets than there are planets in the Solar System. In approximately the same interval, astronomy has transitioned to a field in which sky surveys can generate 1 PB or more of data. The Data-Driven Approaches to Searches for the Technosignatures of Advanced Civilizations_ study at the W. M. Keck Institute for Space Studies was intended to revisit searches for evidence of alien technologies in light of these developments. Data-driven searches, being able to process volumes of data much greater than a human could, and in a reproducible manner, can identify *anomalies* that could be clues to the presence of technosignatures. A key outcome of this workshop was that technosignature searches should be conducted in a manner consistent with Freeman Dyson's "First Law of SETI Investigations," namely "every search for alien civilizations should be planned to give interesting results even when no aliens are discovered." This approach to technosignatures is commensurate with NASA's approach to biosignatures in that no single observation or measurement can be taken as providing full certainty for the detection of life. Areas of particular promise identified during the workshop were (*) Data Mining of Large Sky Surveys, (*) All-Sky Survey at Far-Infrared Wavelengths, (*) Surveys with Radio Astronomical Interferometers, and (*) Artifacts in the Solar System.

astro-ph.IM↗

arXiv2023

ReMaX: Relaxing for Better Training on Efficient Panoptic Segmentation

This paper presents a new mechanism to facilitate the training of mask transformers for efficient panoptic segmentation, democratizing its deployment. We observe that due to its high complexity, the training objective of panoptic segmentation will inevitably lead to much higher false positive penalization. Such unbalanced loss makes the training process of the end-to-end mask-transformer based architectures difficult, especially for efficient models. In this paper, we present ReMaX that adds relaxation to mask predictions and class predictions during training for panoptic segmentation. We demonstrate that via these simple relaxation techniques during training, our model can be consistently improved by a clear margin \textbf{without} any extra computational cost on inference. By combining our method with efficient backbones like MobileNetV3-Small, our method achieves new state-of-the-art results for efficient panoptic segmentation on COCO, ADE20K and Cityscapes. Code and pre-trained checkpoints will be available at \url{https://github.com/google-research/deeplab2}.

cs.CV↗

arXiv2022

Kepler-102: Masses and Compositions for a Super-Earth and Sub-Neptune Orbiting an Active Star

Radial velocity (RV) measurements of transiting multiplanet systems allow us to understand the densities and compositions of planets unlike those in the Solar System. Kepler-102, which consists of 5 tightly packed transiting planets, is a particularly interesting system since it includes a super-Earth (Kepler-102d) and a sub-Neptune-sized planet (Kepler-102e) for which masses can be measured using radial velocities. Previous work found a high density for Kepler-102d, suggesting a composition similar to that of Mercury, while Kepler-102e was found to have a density typical of sub-Neptune size planets; however, Kepler-102 is an active star, which can interfere with RV mass measurements. To better measure the mass of these two planets, we obtained 111 new RVs using Keck/HIRES and TNG/HARPS-N and modeled Kepler-102's activity using quasi-periodic Gaussian Process Regression. For Kepler-102d, we report a mass upper limit of M$_{d} < $5.3 M$_{\oplus}$ [95\% confidence], a best-fit mass of M$_{d}$=2.5 $\pm$ 1.4 M$_{\oplus}$, and a density of $\rho_{d}$=5.6 $\pm$ 3.2 g/cm$^{3}$ which is consistent with a rocky composition similar in density to the Earth. For Kepler-102e we report a mass of M$_{e}$=4.7 $\pm$ 1.7 M$_{\oplus}$ and a density of $\rho_{e}$=1.8 $\pm$ 0.7 g/cm$^{3}$. These measurements suggest that Kepler-102e has a rocky core with a thick gaseous envelope comprising 2-4% of the planet mass and 16-50% of its radius. Our study is yet another demonstration that accounting for stellar activity in stars with clear rotation signals can yield more accurate planet masses, enabling a more realistic interpretation of planet interiors.

astro-ph.EP↗

arXiv2022

On Label Granularity and Object Localization

Weakly supervised object localization (WSOL) aims to learn representations that encode object location using only image-level category labels. However, many objects can be labeled at different levels of granularity. Is it an animal, a bird, or a great horned owl? Which image-level labels should we use? In this paper we study the role of label granularity in WSOL. To facilitate this investigation we introduce iNatLoc500, a new large-scale fine-grained benchmark dataset for WSOL. Surprisingly, we find that choosing the right training label granularity provides a much larger performance boost than choosing the best WSOL algorithm. We also show that changing the label granularity can significantly improve data efficiency.

cs.CV↗

arXiv2022

Report on Community Cadence Observing to Maximize the Scientific Output of the Keck Planet Finder

The arrival of the Keck Planet Finder (KPF) in 2022 represents a major advance in the precision radial velocity (PRV) capabilities of the W. M. Keck Observatory. In preparation for KPF science, our committee of PRV experts and WMKO staff studied the current implementation of cadence observing at Keck and other PRV facilities. We find that many of KPF's major science cases are not feasible through Keck's standard allocations of full or half nights to individual PIs. Pooling time among several PIs as is currently done by the California Planet Search (CPS) collaboration with HIRES results in lower quality science results than is possible when KPF is available at higher observational cadence. This strategy also creates barriers to entry, particularly for researchers wishing to lead small proposals. This report makes recommendations for optimizing PRV cadence at Keck subject to the following constraints: preservation of clear boundaries between cadence observations and classically scheduled time; and ensuring fairness and scientific independence of different Keck TACs and different KPF PIs. We recommend establishing a new category of Keck time allocation, "KPF Community Cadence" (KPF-CC). In many ways, KPF-CC will formalize observing strategies provided by CPS, but with higher observational cadence appropriate for KPF science and with universal access to the program for all Keck users. We recommend that KPF-CC time be scheduled classically into blocks as small as a quarter night subject to considerations of bright/dark time, variations in proposal pressure with the seasons, and the needs of non-KPF observing programs. Within KPF-CC time, the Keck Observing Assistants would execute observations generated by a dynamic scheduler. We recommend that Keck staff and a board of PRV experts design and maintain the scheduling software.

astro-ph.IM↗

arXiv2021

MOSAIC: Mobile Segmentation via decoding Aggregated Information and encoded Context

We present a next-generation neural network architecture, MOSAIC, for efficient and accurate semantic image segmentation on mobile devices. MOSAIC is designed using commonly supported neural operations by diverse mobile hardware platforms for flexible deployment across various mobile platforms. With a simple asymmetric encoder-decoder structure which consists of an efficient multi-scale context encoder and a light-weight hybrid decoder to recover spatial details from aggregated information, MOSAIC achieves new state-of-the-art performance while balancing accuracy and computational cost. Deployed on top of a tailored feature extraction backbone based on a searched classification network, MOSAIC achieves a 5% absolute accuracy gain surpassing the current industry standard MLPerf models and state-of-the-art architectures.

cs.CV↗

arXiv2021

Panoramic SETI: Overall focal plane electronics and timing and network protocols

The PANOSETI experiment is an all-sky, all-the-time visible search for nanosecond to millisecond time-scale transients. The experiment will deploy observatory domes at several sites, each dome containing ~45 telescopes and covering ~4,440 square degrees. Here we describe the focal-plane electronics for the visible wavelength telescopes, each of which contains a Mother Board and four Quadrant Boards. On each quadrant board, 256 silicon photomultiplier (SiPM) photon detectors are arranged to measure pulse heights to search for nanosecond time-scale pulses. To simultaneously examine pulse widths over a large range of time scales (nanoseconds to milliseconds), the instrument implements both a Continuous Imaging Mode (CI-Mode) and a Pulse Height Mode (PH-Mode). Precise timing is implemented in the gateware with the White Rabbit protocol.

astro-ph.IM↗

arXiv2021

Bridging the Gap Between Object Detection and User Intent via Query-Modulation

When interacting with objects through cameras, or pictures, users often have a specific intent. For example, they may want to perform a visual search. With most object detection models relying on image pixels as their sole input, undesired results are not uncommon. Most typically: lack of a high-confidence detection on the object of interest, or detection with a wrong class label. The issue is especially severe when operating capacity-constrained mobile object detectors on-device. In this paper we investigate techniques to modulate mobile detectors to explicitly account for the user intent, expressed as an embedding of a simple query. Compared to standard detectors, query-modulated detectors show superior performance at detecting objects for a given user query. Thanks to large-scale training data synthesized from standard object detection annotations, query-modulated detectors also outperform a specialized referring expression recognition system. Query-modulated detectors can also be trained to simultaneously solve for both localizing a user query and standard detection, even outperforming standard mobile detectors at the canonical COCO task.

cs.CV↗

arXiv2021

BasisNet: Two-stage Model Synthesis for Efficient Inference

In this work, we present BasisNet which combines recent advancements in efficient neural network architectures, conditional computation, and early termination in a simple new form. Our approach incorporates a lightweight model to preview the input and generate input-dependent combination coefficients, which later controls the synthesis of a more accurate specialist model to make final prediction. The two-stage model synthesis strategy can be applied to any network architectures and both stages are jointly trained. We also show that proper training recipes are critical for increasing generalizability for such high capacity neural networks. On ImageNet classification benchmark, our BasisNet with MobileNets as backbone demonstrated clear advantage on accuracy-efficiency trade-off over several strong baselines. Specifically, BasisNet-MobileNetV3 obtained 80.3% top-1 accuracy with only 290M Multiply-Add operations, halving the computational cost of previous state-of-the-art without sacrificing accuracy. With early termination, the average cost can be further reduced to 198M MAdds while maintaining accuracy of 80.0% on ImageNet.

cs.CV↗