arXiv subjects

Michael Zhang

Publications and source records attributed to Michael Zhang.

105 records · Page 6Linked to original sources

arXiv2019

Ranking genetic factors related to age-related maculardegeneration by variable selection confidence sets

The widespread use of generalized linear models in case-control genetic studies has helped identify many disease-associated risk factors typically defined as DNA variants, or single nucleotide polymorphisms (SNPs). Up to now, most literature has focused on selecting a unique best subset of SNPs based on some statistical perspectives. In the presence of pronounced noise, however, multiple biological paths are often found to be equally supported by a given dataset when dealing with complex genetic diseases. We address the ambiguity related to SNP selection by constructing a list of models called variable selection confidence set (VSCS), which contains the collection of all well-supported SNP combinations at a user-specified confidence level. The VSCS extends the familiar notion of confidence intervals in the variable selection setting and provides the practitioner with new tools aiding the variable selection activity beyond trusting a single model. Based on the VSCS, we consider natural graphical and numerical statistics measuring the inclusion importance of a SNP based on its frequency in the most parsimonious VSCS models. This work is motivated by available case-control genetic data on age-related macular degeneration, a widespread complex disease and leading cause of vision loss.

stat.AP↗

arXiv2019

Forward modelling and retrievals with PLATON, a fast open source tool

We introduce PLanetary Atmospheric Transmission for Observer Noobs (PLATON), a Python package that calculates transmission spectra for exoplanets and retrieves atmospheric characteristics based on observed spectra. PLATON is easy to install and use, with common use cases taking no more than a few lines of code. It is also fast, with the forward model taking much less than one second and a typical retrieval finishing in minutes on an ordinary desktop. PLATON supports the most common atmospheric parameters, such as temperature, metallicity, C/O ratio, cloud-top pressure, and scattering slope. It also has less commonly included features, such as a Mie scattering cloud model and unocculted starspot corrections. The code is available online at https://github.com/ideasrule/platon under the open source GPL-3.0 license.

astro-ph.EP↗

arXiv2019

The long-term evolution and appearance of Type Iax postgenitor stars

Type Iax supernovae may arise from failed explosions of white dwarfs that leave behind a bound remnant (i.e., a "postgenitor" star) that could be identified in wide field surveys. To understand their observational signatures, we simulate these white dwarf (WD) postgenitors from shortly after explosion until they move back down the WD cooling track, and we consider several possible WD masses and explosion energies. To predict the peculiar surface abundances of the WD postgenitors, our models take into account gravitational settling and radiative levitation. We find that radiative levitation is significant at temperatures above a mass-dependent critical temperature, typically in the range Teff ~ 50-100 * 10^3 K, significantly increasing surface abundances of iron-group elements. Due to enhanced iron group opacity compared to normal WDs, the postgenitor peak luminosity and cooling timescale depend sensitively on mass, with more massive WDs becoming brighter but cooling much faster. We discuss our results in light of recently discovered hypervelocity white dwarfs with peculiar surface compositions, finding that our low-mass postgenitor models match many of their observational characteristics. Finally, we explore the effects of thermohaline diffusion, tentatively finding that it strongly suppresses abundance enhancements created by radiative levitation, but more realistic modeling is required to reach a firm conclusion.

astro-ph.SR↗

arXiv2018

Deep Weighted Averaging Classifiers

Recent advances in deep learning have achieved impressive gains in classification accuracy on a variety of types of data, including images and text. Despite these gains, however, concerns have been raised about the calibration, robustness, and interpretability of these models. In this paper we propose a simple way to modify any conventional deep architecture to automatically provide more transparent explanations for classification decisions, as well as an intuitive notion of the credibility of each prediction. Specifically, we draw on ideas from nonparametric kernel regression, and propose to predict labels based on a weighted sum of training instances, where the weights are determined by distance in a learned instance-embedding space. Working within the framework of conformal methods, we propose a new measure of nonconformity suggested by our model, and experimentally validate the accompanying theoretical expectations, demonstrating improved transparency, controlled error rates, and robustness to out-of-domain data, without compromising on accuracy or calibration.

cs.LG↗

arXiv2018

Reverse Curriculum Generation for Reinforcement Learning

Many relevant tasks require an agent to reach a certain state, or to manipulate objects into a desired configuration. For example, we might want a robot to align and assemble a gear onto an axle or insert and turn a key in a lock. These goal-oriented tasks present a considerable challenge for reinforcement learning, since their natural reward function is sparse and prohibitive amounts of exploration are required to reach the goal and receive some learning signal. Past approaches tackle these problems by exploiting expert demonstrations or by manually designing a task-specific reward shaping function to guide the learning agent. Instead, we propose a method to learn these tasks without requiring any prior knowledge other than obtaining a single state in which the task is achieved. The robot is trained in reverse, gradually learning to reach the goal from a set of start states increasingly far from the goal. Our method automatically generates a curriculum of start states that adapts to the agent's performance, leading to efficient training on goal-oriented tasks. We demonstrate our approach on difficult simulated navigation and fine-grained manipulation problems, not solvable by state-of-the-art reinforcement learning methods.

cs.AI↗

arXiv2018

Improved OpenCL-based Implementation of Social Field Pedestrian Model

Two aspects of improvements are proposed for the OpenCL-based implementation of the social field pedestrian model. In the aspect of algorithm, a method based on the idea of divide-and-conquer is devised in order to overcome the problem of global memory depletion when fields are of a larger size. This is of importance for the study of finer pedestrian walking behavior, which usually requires larger fields. In the aspect of computation, the OpenCL heterogeneous framework is thoroughly studied. Factors that may affect the numerical efficiency are evaluated, with regarding to the social field model previously proposed. This includes usage of local memory, deliberate patch of data structures for avoidance of bank conflicts, and so on. Numerical experiments disclose that the numerical efficiency is brought to an even higher level. Compared to the CPU model and the previous GPU model, the current GPU model can be at most 71.56 and 13.3 times faster respectively so that it is more qualified to be a core engine for analysis of super-large scale crowd.

cs.DC↗

arXiv2018

Berry curvature and symmetry broken induced Hall effect in MoS2

Inversion symmetry breaking and spin-orbit coupling lead to valley and spin Hall effect in MoS2. Because of the large valley separation in momentum space, the valley index is expected to be robust. In this paper, quantum Hall effect in MoS2 originated from Berry curvature is analyzed after review of symmetry structure and spin-orbit coupling Hamiltonian of MoS2. Finally an expression and rough calculation is given for valley and spin Hall effect.

cond-mat.mes-hall↗

arXiv2018

Detection of a Westward Hotspot Offset in the Atmosphere of a Hot Gas Giant CoRoT-2b

Short-period planets exhibit day-night temperature contrasts of hundreds to thousands of degrees K. They also exhibit eastward hotspot offsets whereby the hottest region on the planet is east of the substellar point; this has been widely interpreted as advection of heat due to eastward winds. We present thermal phase observations of the hot Jupiter CoRoT-2b obtained with the IRAC instrument on the Spitzer Space Telescope. These measurements show the most robust detection to date of a westward hotspot offset of 23 $\pm$ 4 degrees, in contrast with the nine other planets with equivalent measurements. The peculiar infrared flux map of CoRoT-2b may result from westward winds due to non-synchronous rotation magnetic effects, or partial cloud coverage, that obscures the emergent flux from the planet's eastern hemisphere. Non-synchronous rotation and magnetic effects may also explain the planet's anomalously large radius. On the other hand, partial cloud coverage could explain the featureless dayside emission spectrum of the planet. If CoRoT-2b is not tidally locked, then it means that our understanding of star-planet tidal interaction is incomplete. If the westward offset is due to magnetic effects, our result represents an opportunity to study an exoplanet's magnetic field. If it has Eastern clouds, then it means that our understanding of large-scale circulation on tidally locked planets is incomplete.

astro-ph.EP↗

arXiv2017

Phase curves of WASP-33b and HD 149026b and a New Correlation Between Phase Curve Offset and Irradiation Temperature

We present new 3.6 and 4.5 $μm$ Spitzer phase curves for the highly irradiated hot Jupiter WASP-33b and the unusually dense Saturn-mass planet HD 149026b. As part of this analysis, we develop a new variant of pixel level decorrelation that is effective at removing intrapixel sensitivity variations for long observations (>10 hours) where the position of the star can vary by a significant fraction of a pixel. Using this algorithm, we measure eclipse depths, phase amplitudes, and phase offsets for both planets at 3.6 $μm$ and 4.5 $μm$. We use a simple toy model to show that WASP-33b's phase offset, albedo, and heat recirculation efficiency are largely similar to those of other hot Jupiters despite its very high irradiation. On the other hand, our fits for HD 149026b prefer a very high albedo and an unusually high recirculation efficiency. We also compare our results to predictions from general circulation models, and find that while neither are a good match to the data, the discrepancies for HD 149026b are especially large. We speculate that this may be related to its high bulk metallicity, which could lead to enhanced atmospheric opacities and the formation of reflective cloud layers in localized regions of the atmosphere. We then place these two planets in a broader context by exploring relationships between the temperatures, albedos, heat transport efficiencies, and phase offsets of all planets with published thermal phase curves. We find a striking relationship between phase offset and irradiation temperature--the former drops with increasing temperature until around 3400 K, and rises thereafter. Although some aspects of this trend are mirrored in the circulation models, there are notable differences that provide important clues for future modeling efforts.

astro-ph.EP↗

arXiv2017

Reinforcement Mechanism Design, with Applications to Dynamic Pricing in Sponsored Search Auctions

In this study, we apply reinforcement learning techniques and propose what we call reinforcement mechanism design to tackle the dynamic pricing problem in sponsored search auctions. In contrast to previous game-theoretical approaches that heavily rely on rationality and common knowledge among the bidders, we take a data-driven approach, and try to learn, over repeated interactions, the set of optimal reserve prices. We implement our approach within the current sponsored search framework of a major search engine: we first train a buyer behavior model, via a real bidding data set, that accurately predicts bids given information that bidders are aware of, including the game parameters disclosed by the search engine, as well as the bidders' KPI data from previous rounds. We then put forward a reinforcement/MDP (Markov Decision Process) based algorithm that optimizes reserve prices over time, in a GSP-like auction. Our simulations demonstrate that our framework outperforms static optimization strategies including the ones that are currently in use, as well as several other dynamic ones.

cs.GT↗

arXiv2017

Probabilistically Safe Policy Transfer

Although learning-based methods have great potential for robotics, one concern is that a robot that updates its parameters might cause large amounts of damage before it learns the optimal policy. We formalize the idea of safe learning in a probabilistic sense by defining an optimization problem: we desire to maximize the expected return while keeping the expected damage below a given safety limit. We study this optimization for the case of a robot manipulator with safety-based torque limits. We would like to ensure that the damage constraint is maintained at every step of the optimization and not just at convergence. To achieve this aim, we introduce a novel method which predicts how modifying the torque limit, as well as how updating the policy parameters, might affect the robot's safety. We show through a number of experiments that our approach allows the robot to improve its performance while ensuring that the expected damage constraint is not violated during the learning process.

cs.RO↗

arXiv2014

Stars Get Dizzy After Lunch

Exoplanet searches have discovered a large number of 'hot Jupiters'--high-mass planets orbiting very close to their parent stars in nearly circular orbits. A number of these planets are sufficiently massive and close-in to be significantly affected by tidal dissipation in the parent star, to a degree parametrized by the tidal quality factor $Q_*$. This process speeds up their stars' rotation rate while reducing the planets' semimajor axis. In this paper, we investigate the tidal destruction of hot Jupiters. Because the orbital angular momenta of these planets are a significant fraction of their stars' rotational angular momenta, they spin up their stars significantly while spiralling to their deaths. Using Monte Carlo simulation, we predict that for $Q_* = 10^6$, $3.9\times 10^{-6}$ of stars with the Kepler Target Catalog's mass distribution should have a rotation period shorter than 1/3 day (8 h) due to accreting a planet. Exoplanet surveys such as SuperWASP, HATnet, HATsouth, and KELT have already produced light curves of millions of stars. These two facts suggest that it may be possible to search for tidally-destroyed planets by looking for stars with extremely short rotational periods, then looking for remnant planet cores around those candidates, anomalies in the metal distribution, or other signatures of the recent accretion of the planet.

astro-ph.EP↗

arXiv2014

POET: A Model for (P)lanetary (O)rbital (E)volution due to (T)ides on Evolving Stars

We make publicly available an efficient, versatile, easy to use and extend tool for calculating the evolution of circular aligned planetary orbits due to the tidal dissipation in the host star. This is the first model to fully account for the evolution of the angular momentum of the stellar convective envelope by the tidal coupling, the transfer of angular momentum between the stellar convective and radiative zones, the effects of the stellar evolution on the tidal dissipation efficiency and stellar core and envelope spins, the loss of stellar convective zone angular momentum to a magnetically launched wind and frequency dependent tidal dissipation. This is only a first release and further development is under way to allow calculating the evolution of inclined and eccentric orbits, with the latter including the tidal dissipation in the planet and its feedback on planetary structure. Considerable effort has been devoted to providing extensive documentation detailing both the usage and the complete implementation details, in order to make it as easy as possible for independent groups to use and/or extend the code for their purposes. POET represents a significant improvement over some previous models for planetary tidal evolution and so has many astrophysical applications. In this article, we describe and illustrate several key examples.

astro-ph.EP↗

arXiv2011

Poisson traces in positive characteristic

We study Poisson traces of the structure algebra A of an affine Poisson variety X defined over a field of characteristic p. According to arXiv:0908.3868v4, the dual space HP_0(A) to the space of Poisson traces arises as the space of coinvariants associated to a certain D-module M(X) on X. If X has finitely many symplectic leaves and the ground field has characteristic zero, then M(X) is holonomic, and thus HP_0(A) is finite dimensional. However, in characteristic p, the dimension of HP_0(A) is typically infinite. Our main results are complete computations of HP_0(A) for sufficiently large p when X is 1) a quasi-homogeneous isolated surface singularity in the three-dimensional space, 2) a quotient singularity V/G, for a symplectic vector space V by a finite subgroup G in Sp(V), and 3) a symmetric power of a symplectic vector space or a Kleinian singularity. In each case, there is a finite nonnegative grading, and we compute explicitly the Hilbert series. The proofs are based on the theory of D-modules in positive characteristic.

math.SG↗

arXiv2005

Direct Observation of Broadband Coating Thermal Noise in a Suspended Interferometer

We have directly observed broadband thermal noise in silica/tantala coatings in a high-sensitivity Fabry-Perot interferometer. Our result agrees well with the prediction based on indirect, ring-down measurements of coating mechanical loss, validating that method as a tool for the development of advanced interferometric gravitational-wave detectors.

gr-qc↗