arXiv subjects

Dan Moldovan

Publications and source records attributed to Dan Moldovan.

12 recordsLinked to original sources

arXiv2024

Volt-PF Control Mode for Distribution Feeder Voltage Management Under High Penetration of Distributed Energy Resources

Volt-VAr control is a popular method for mitigating overvoltage violations caused by high penetration of distributed energy resources (DERs) in distribution feeders. An inherent limitation of volt-VAr control is that the reactive power (Q) absorbed/injected by the DER is determined based only on the terminal voltage, without considering the active power (P) generated by the DER. This leads to an inequitable burden of Q support, in the sense that those DERs generating lower P, and hence contributing less to overvoltage issues, may be required to provide more than their share of $Q$ support. The resulting PF of these DERs is required to vary over a wide range, which many current DERs do not support. A new control scheme, namely volt-PF control, is proposed here where the Q support is inherently a function of both the voltage and $P$ from DERs, which alleviates the above concerns while limiting the PF variation within a narrow range of 0.9 to 1. The proposed scheme is validated through extensive static and dynamic simulations on a real, large (8000+ nodes) feeder with very high penetration (>200%) of DERs.The implementation of the new scheme in new and existing commercial hardware inverters is described.

eess.SY↗

arXiv2023

TOI-3235 b: a transiting giant planet around an M4 dwarf star

We present the discovery of TOI-3235 b, a short-period Jupiter orbiting an M-dwarf with a stellar mass close to the critical mass at which stars transition from partially to fully convective. TOI-3235 b was first identified as a candidate from TESS photometry, and confirmed with radial velocities from ESPRESSO, and ground-based photometry from HATSouth, MEarth-South, TRAPPIST-South, LCOGT, and ExTrA. We find that the planet has a mass of $\mathrm{0.665\pm0.025\,M_J}$ and a radius of $\mathrm{1.017\pm0.044\,R_J}$. It orbits close to its host star, with an orbital period of $\mathrm{2.5926\,d}$, but has an equilibrium temperature of $\mathrm{\approx 604 \, K}$, well below the expected threshold for radius inflation of hot Jupiters. The host star has a mass of $\mathrm{0.3939\pm0.0030\,M_\odot}$, a radius of $\mathrm{0.3697\pm0.0018\,R_\odot}$, an effective temperature of $\mathrm{3389 \, K}$, and a J-band magnitude of $\mathrm{11.706\pm0.025}$. Current planet formation models do not predict the existence of gas giants such as TOI-3235 b around such low-mass stars. With a high transmission spectroscopy metric, TOI-3235 b is one of the best-suited giants orbiting M-dwarfs for atmospheric characterization.

astro-ph.EP↗

arXiv2023

Identifying Exoplanets with Deep Learning. V. Improved Light Curve Classification for TESS Full Frame Image Observations

The TESS mission produces a large amount of time series data, only a small fraction of which contain detectable exoplanetary transit signals. Deep learning techniques such as neural networks have proved effective at differentiating promising astrophysical eclipsing candidates from other phenomena such as stellar variability and systematic instrumental effects in an efficient, unbiased and sustainable manner. This paper presents a high quality dataset containing light curves from the Primary Mission and 1st Extended Mission full frame images and periodic signals detected via Box Least Squares (Kov\'acs et al. 2002; Hartman 2012). The dataset was curated using a thorough manual review process then used to train a neural network called Astronet-Triage-v2. On our test set, for transiting/eclipsing events we achieve a 99.6% recall (true positives over all data with positive labels) at a precision of 75.7% (true positives over all predicted positives). Since 90% of our training data is from the Primary Mission, we also test our ability to generalize on held-out 1st Extended Mission data. Here, we find an area under the precision-recall curve of 0.965, a 4% improvement over Astronet-Triage (Yu et al. 2019). On the TESS Object of Interest (TOI) Catalog through April 2022, a shortlist of planets and planet candidates, Astronet-Triage-v2 is able to recover 3577 out of 4140 TOIs, while Astronet-Triage only recovers 3349 targets at an equal level of precision. In other words, upgrading to Astronet-Triage-v2 helps save at least 200 planet candidates from being lost. The new model is currently used for planet candidate triage in the Quick-Look Pipeline (Huang et al. 2020a,b; Kunimoto et al. 2021).

astro-ph.EP↗

arXiv2022

An old warm Jupiter orbiting the metal-poor G-dwarf TOI-5542

We report the discovery of a 1.32$^{+0.10}_{-0.10}$ $\mathrm{M_{\rm Jup}}$ planet orbiting on a 75.12 day period around the G3V $10.8^{+2.1}_{-3.6}$ Gyr old star TOI-5542 (TIC 466206508; TYC 9086-1210-1). The planet was first detected by the Transiting Exoplanet Survey Satellite (TESS) as a single transit event in TESS Sector 13. A second transit was observed 376 days later in TESS Sector 27. The planetary nature of the object has been confirmed by ground-based spectroscopic and radial velocity observations from the CORALIE and HARPS spectrographs. A third transit event was detected by the ground-based facilities NGTS, EulerCam, and SAAO. We find the planet has a radius of 1.009$^{+0.036}_{-0.035}$ $\mathrm{R_{\rm Jup}}$ and an insolation of 9.6$^{+0.9}_{-0.8}$ $S_{\oplus}$, along with a circular orbit that most likely formed via disk migration or in situ formation, rather than high-eccentricity migration mechanisms. Our analysis of the HARPS spectra yields a host star metallicity of [Fe/H] = $-$0.21$\pm$0.08, which does not follow the traditional trend of high host star metallicity for giant planets and does not bolster studies suggesting a difference among low- and high-mass giant planet host star metallicities. Additionally, when analyzing a sample of 216 well-characterized giant planets, we find that both high masses (4 $\mathrm{M_{\rm Jup}}$ $ $ 10 days) and hot (P $<$ 10 days) giant planets are preferentially located around metal-rich stars (mean [Fe/H] $>$ 0.1). TOI-5542b is one of the oldest known warm Jupiters and it is cool enough to be unaffected by inflation due to stellar incident flux, making it a valuable contribution in the context of planetary composition and formation studies.

astro-ph.EP↗

arXiv2022

TOI-2046b, TOI-1181b and TOI-1516b, three new hot Jupiters from \textit{TESS}: planets orbiting a young star, a subgiant and a normal star

We present the confirmation and characterization of three hot Jupiters, TOI-1181b, TOI-1516b, and TOI-2046b, discovered by the TESS space mission. The reported hot Jupiters have orbital periods between 1.4 and 2.05 days. The masses of the three planets are $1.18\pm0.14$ M$_{\mathrm{J}}$, $3.16\pm0.12$\, M$_{\mathrm{J}}$, and 2.30 $\pm 0.28$ M$_{\mathrm{J}}$, for TOI-1181b, TOI-1516b, and TOI-2046b, respectively. The stellar host of TOI-1181b is a F9IV star, whereas TOI-1516b and TOI-2046b orbit F main sequence host stars. The ages of the first two systems are in the range of 2-5 Gyrs. However, TOI-2046 is among the few youngest known planetary systems hosting a hot Jupiter, with an age estimate of 100-400 Myrs. The main instruments used for the radial velocity follow-up of these three planets are located at Ond\v{r}ejov, Tautenburg and McDonald Observatory, and all three are mounted on 2-3 meter aperture telescopes, demonstrating that mid-aperture telescope networks can play a substantial role in the follow-up of gas giants discovered by \textit{TESS} and in the future by \textit{PLATO}.

astro-ph.EP↗

arXiv2022

TOI-2119: A transiting brown dwarf orbiting an active M-dwarf from NASA's TESS mission

We report the discovery of TOI-2119b, a transiting brown dwarf (BD) that orbits and is completely eclipsed by an active M-dwarf star. Using light curve data from the Transiting Exoplanet Survey Satellite mission and follow-up high-resolution Doppler spectroscopic observations, we find the BD has a radius of $R_b = 1.08 \pm 0.03{\rm R_J}$, a mass of $M_b = 64.4 \pm 2.3{\rm M_J}$, an orbital period of $P = 7.200865 \pm 0.00002$ days, and an eccentricity of $e=0.337\pm 0.002$. The host star has a mass of $M_\star = 0.53 \pm 0.02{\rm M_\odot}$, a radius of $R_\star= 0.50 \pm 0.01{\rm R_\odot}$, an effective temperature of $T_{\rm eff} = 3621 \pm 48$K, and a metallicity of $\rm [Fe/H]=+0.06\pm 0.08$. TOI-2119b joins an emerging population of transiting BDs around M-dwarf host stars, with TOI-2119 being the ninth such system. These M-dwarf--brown dwarf systems typically occupy mass ratios near $q = M_b/M_\star \approx 0.1-0.2$, which separates them from the typical mass ratios for systems with transiting substellar objects and giant exoplanets that orbit more massive stars. The nature of the secondary eclipse of the BD by the star enables us to estimate the effective temperature of the substellar object to be $2030\pm 84$K, which is consistent with predictions by substellar evolutionary models.

astro-ph.SR↗

arXiv2021

Causally motivated Shortcut Removal Using Auxiliary Labels

Shortcut learning, in which models make use of easy-to-represent but unstable associations, is a major failure mode for robust machine learning. We study a flexible, causally-motivated approach to training robust predictors by discouraging the use of specific shortcuts, focusing on a common setting where a robust predictor could achieve optimal \emph{iid} generalization in principle, but is overshadowed by a shortcut predictor in practice. Our approach uses auxiliary labels, typically available at training time, to enforce conditional independences implied by the causal graph. We show both theoretically and empirically that causally-motivated regularization schemes (a) lead to more robust estimators that generalize well under distribution shift, and (b) have better finite sample efficiency compared to usual regularization schemes, even when no shortcut is present. Our analysis highlights important theoretical properties of training techniques commonly used in the causal inference, fairness, and disentanglement literatures. Our code is available at https://github.com/mymakar/causally_motivated_shortcut_removal

cs.LG↗

arXiv2020

Underspecification Presents Challenges for Credibility in Modern Machine Learning

ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain. Underspecification is common in modern ML pipelines, such as those based on deep learning. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We show that this problem appears in a wide variety of practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.

cs.LG↗

arXiv2020

On Robustness and Transferability of Convolutional Neural Networks

Modern deep convolutional networks (CNNs) are often criticized for not generalizing under distributional shifts. However, several recent breakthroughs in transfer learning suggest that these networks can cope with severe distribution shifts and successfully adapt to new tasks from a few training examples. In this work we study the interplay between out-of-distribution and transfer performance of modern image classification CNNs for the first time and investigate the impact of the pre-training data size, the model scale, and the data preprocessing pipeline. We find that increasing both the training set and model sizes significantly improve the distributional shift robustness. Furthermore, we show that, perhaps surprisingly, simple changes in the preprocessing such as modifying the image resolution can significantly mitigate robustness issues in some cases. Finally, we outline the shortcomings of existing robustness evaluation datasets and introduce a synthetic dataset SI-Score we use for a systematic analysis across factors of variation common in visual data such as object size and position.

cs.CV↗

arXiv2019

AutoGraph: Imperative-style Coding with Graph-based Performance

There is a perceived trade-off between machine learning code that is easy to write, and machine learning code that is scalable or fast to execute. In machine learning, imperative style libraries like Autograd and PyTorch are easy to write, but suffer from high interpretive overhead and are not easily deployable in production or mobile settings. Graph-based libraries like TensorFlow and Theano benefit from whole-program optimization and can be deployed broadly, but make expressing complex models more cumbersome. We describe how the use of staged programming in Python, via source code transformation, offers a midpoint between these two library design patterns, capturing the benefits of both. A key insight is to delay all type-dependent decisions until runtime, via dynamic dispatch. We instantiate these principles in AutoGraph, a software system that improves the programming experience of the TensorFlow library, and demonstrate usability improvements with no loss in performance compared to native TensorFlow graphs. We also show that our system is backend agnostic, and demonstrate targeting an alternate IR with characteristics not found in TensorFlow graphs.

cs.PL↗

arXiv2018

Tangent: Automatic differentiation using source-code transformation for dynamically typed array programming

The need to efficiently calculate first- and higher-order derivatives of increasingly complex models expressed in Python has stressed or exceeded the capabilities of available tools. In this work, we explore techniques from the field of automatic differentiation (AD) that can give researchers expressive power, performance and strong usability. These include source-code transformation (SCT), flexible gradient surgery, efficient in-place array operations, higher-order derivatives as well as mixing of forward and reverse mode AD. We implement and demonstrate these ideas in the Tangent software library for Python, the first AD framework for a dynamic language that uses SCT.

cs.LG↗

arXiv2017

Tangent: Automatic Differentiation Using Source Code Transformation in Python

Automatic differentiation (AD) is an essential primitive for machine learning programming systems. Tangent is a new library that performs AD using source code transformation (SCT) in Python. It takes numeric functions written in a syntactic subset of Python and NumPy as input, and generates new Python functions which calculate a derivative. This approach to automatic differentiation is different from existing packages popular in machine learning, such as TensorFlow and Autograd. Advantages are that Tangent generates gradient code in Python which is readable by the user, easy to understand and debug, and has no runtime overhead. Tangent also introduces abstractions for easily injecting logic into the generated gradient code, further improving usability.

cs.MS↗