arXiv subjects

Cheng Li

Publications and source records attributed to Cheng Li.

At least 469 records · Page 26Linked to original sources

arXiv2018

Rapid Bayesian optimisation for synthesis of short polymer fiber materials

The discovery of processes for the synthesis of new materials involves many decisions about process design, operation, and material properties. Experimentation is crucial but as complexity increases, exploration of variables can become impractical using traditional combinatorial approaches. We describe an iterative method which uses machine learning to optimise process development, incorporating multiple qualitative and quantitative objectives. We demonstrate the method with a novel fluid processing platform for synthesis of short polymer fibers, and show how the synthesis process can be efficiently directed to achieve material and process objectives.

stat.ML↗

arXiv2018

High Dimensional Bayesian Optimization Using Dropout

Scaling Bayesian optimization to high dimensions is challenging task as the global optimization of high-dimensional acquisition function can be expensive and often infeasible. Existing methods depend either on limited active variables or the additive form of the objective function. We propose a new method for high-dimensional Bayesian optimization, that uses a dropout strategy to optimize only a subset of variables at each iteration. We derive theoretical bounds for the regret and show how it can inform the derivation of our algorithm. We demonstrate the efficacy of our algorithms for optimization on two benchmark functions and two real-world applications- training cascade classifiers and optimizing alloy composition.

stat.ML↗

arXiv2018

SDSS-IV MaNGA: Global stellar population and gradients for about 2000 early-type and spiral galaxies on the mass-size plane

We perform full spectrum fitting stellar population analysis and Jeans Anisotropic modelling (JAM) of the stellar kinematics for about 2000 early-type galaxies (ETGs) and spiral galaxies from the MaNGA DR14 sample. Galaxies with different morphologies are found to be located on a remarkably tight mass plane which is close to the prediction of the virial theorem, extending previous results for ETGs. By examining an inclined projection (`the mass-size' plane), we find that spiral and early-type galaxies occupy different regions on the plane, and their stellar population properties (i.e. age, metallicity and stellar mass-to-light ratio) vary systematically along roughly the direction of velocity dispersion, which is a proxy for the bulge fraction. Galaxies with higher velocity dispersions have typically older ages, larger stellar mass-to-light ratios and are more metal rich, which indicates that galaxies increase their bulge fractions as their stellar populations age and become enriched chemically. The age and stellar mass-to-light ratio gradients for low-mass galaxies in our sample tend to be positive ($\rm centre<outer$), while the gradients for most massive galaxies are negative. The metallicity gradients show a clear peak around velocity dispersion $\log_{10} σ_{\rm e}\approx 2.0$, which corresponds to the critical mass $\sim 3\times 10^{10}M_{\odot}$ of the break in the mass-size relation. Spiral galaxies with large mass and size have the steepest gradients, while the most massive ETGs, especially above the critical mass $M_{\rm crit}\ge 2\times 10^{11} M_{\odot}$, where slow rotator ETGs start dominating, have much flatter gradients. This may be due to differences in their evolution histories, e.g. mergers.

astro-ph.GA↗

arXiv2018

On Bayesian Oracle Properties

When model uncertainty is handled by Bayesian model averaging (BMA) or Bayesian model selection (BMS), the posterior distribution possesses a desirable "oracle property" for parametric inference, if for large enough data it is nearly as good as the oracle posterior, obtained by assuming unrealistically that the true model is known and only the true model is used. We study the oracle properties in a very general context of quasi-posterior, which can accommodate non-regular models with cubic root asymptotics and partial identification. Our approach for proving the oracle properties is based on a unified treatment that bounds the posterior probability of model mis-selection. This theoretical framework can be of interest to Bayesian statisticians who would like to theoretically justify their new model selection or model averaging methods in addition to empirical results. Furthermore, for non-regular models, we obtain nontrivial conclusions on the choice of prior penalty on model complexity, the temperature parameter of the quasi-posterior, and the advantage of BMA over BMS.

math.ST↗

arXiv2018

Moist adiabats with multiple condensing species: A new theory with application to giant planet atmospheres

We derived a new formula for calculating the moist adiabatic temperature profile of an atmosphere consisting of ideal gases with multiple condensing species. This expression unifies various formulas published in the literature and can be generalized to account for chemical reactions. Unlike previous methods, it converges to machine precision independent of mesh size. It accounts for any ratio of condensable vapors to dry gas, from zero to infinity, and for variable heat capacities as a function of temperature. Because the derivation is generic, the new formula is not only applicable to planetary atmosphere in the solar system, but also to hot Jupiters and brown dwarfs in which a variety of alkali metals, silicates and exotic materials condense. We demonstrate that even though the vapors are ideal gases, they interact in their effects on the moist adiabatic lapse rate. Finally, we apply the new thermodynamic model to study the effects of downdrafts on the distribution of minor constituents and thermal profile in the Galileo probe hotspot. We find that the Galileo Probe measurements can be interpreted as a strong downdraft that displaces an air parcel from 1 bar to the 4 bar level.

astro-ph.EP↗

arXiv2018

Modulation Classification Using Received Signal's Amplitude Distribution for Coherent Receivers

In this letter, we propose a modulation classification algorithm which is based on the received signal's amplitude for coherent optical receivers. The proposed algorithm classifies the modulation format from several possible candidates by differentiating the cumulative distribution function (CDF) curves of their normalized amplitudes. The candidate with the most similar CDF to the received signal is selected. The measure of similarity is the minimum average distance between these CDFs. Five commonly used quadrature amplitude modulation formats in digital coherent optical systems are employed. Optical back-to-back experiments and extended simulations are carried out to investigate the performance of the proposed algorithm. Results show that the proposed algorithm achieves accurate classification at optical signal-to-noise ratios of interest. Furthermore, it does not require carrier recovery.

eess.SP↗

arXiv2017

Find the Conversation Killers: a Predictive Study of Thread-ending Posts

How to improve the quality of conversations in online communities has attracted considerable attention recently. Having engaged, urbane, and reactive online conversations has a critical effect on the social life of Internet users. In this study, we are particularly interested in identifying a post in a multi-party conversation that is unlikely to be further replied to, which therefore kills that thread of the conversation. For this purpose, we propose a deep learning model called the ConverNet. ConverNet is attractive due to its capability of modeling the internal structure of a long conversation and its appropriate encoding of the contextual information of the conversation, through effective integration of attention mechanisms. Empirical experiments on real-world datasets demonstrate the effectiveness of the proposal model. For the widely concerned topic, our analysis also offers implications for improving the quality and user experience of online conversations.

cs.CL↗

arXiv2017

SDSS-IV MaNGA: Constraints on the Conditions for Star Formation in Galaxy Discs

Regions of disc galaxies with widespread star formation tend to be both gravitationally unstable and self-shielded against ionizing radiation, whereas extended outer discs with little or no star formation tend to be stable and unshielded on average. We explore what drives the transition between these two regimes, specifically whether discs first meet the conditions for self-shielding (parameterized by dust optical depth, $τ$) or gravitational instability (parameterized by a modified version of Toomre's instability parameters, $Q_{\rm thermal}$, which quantifies the stability of a gas disc that is thermally supported at $T=10^4$ K). We first introduce a new metric formed by the product of these quantities, $Q_{\rm thermal}τ$, which indicates whether the conditions for disk instability or self-shielding are easier to meet in a given region of a galaxy, and we discuss how $Q_{\rm thermal}τ$ can be constrained even in the absence of direct gas information. We then analyse a sample of 13 galaxies with resolved gas measurements and find that on average galaxies will reach the threshold for disk instabilities ($Q_{\rm thermal}<1$) before reaching the threshold for self-shielding ($τ>1$). Using integral field spectroscopic observations of a sample of 236 galaxies from the MaNGA survey, we find that the value of $Q_{\rm thermal}τ$ in star-forming discs is consistent with similar behavior. These results support a scenario where disc fragmentation and collapse occurs before self-shielding, suggesting that gravitational instabilities are the primary condition for widespread star formation in galaxy discs. Our results support similar conclusions based on recent galaxy simulations.

astro-ph.GA↗

arXiv2017

From which world is your graph?

Discovering statistical structure from links is a fundamental problem in the analysis of social networks. Choosing a misspecified model, or equivalently, an incorrect inference algorithm will result in an invalid analysis or even falsely uncover patterns that are in fact artifacts of the model. This work focuses on unifying two of the most widely used link-formation models: the stochastic blockmodel (SBM) and the small world (or latent space) model (SWM). Integrating techniques from kernel learning, spectral graph theory, and nonlinear dimensionality reduction, we develop the first statistically sound polynomial-time algorithm to discover latent patterns in sparse graphs for both models. When the network comes from an SBM, the algorithm outputs a block structure. When it is from an SWM, the algorithm outputs estimates of each node's latent position.

cs.LG↗

arXiv2017

Resolved star formation and molecular gas properties of green valley galaxies: a first look with ALMA and MaNGA

We study the role of cold gas in quenching star formation in the green valley by analysing ALMA $^{12}$CO (1-0) observations of three galaxies with resolved optical spectroscopy from the MaNGA survey. We present resolution-matched maps of the star formation rate and molecular gas mass. These data are used to calculate the star formation efficiency (SFE) and gas fraction ($f_{\rm~gas}$) for these galaxies separately in the central `bulge' regions and outer disks. We find that, for the two galaxies whose global specific star formation rate (sSFR) deviates most from the star formation main sequence, the gas fraction in the bulges is significantly lower than that in their disks, supporting an `inside-out' model of galaxy quenching. For the two galaxies where SFE can be reliably determined in the central regions, the bulges and disks share similar SFEs. This suggests that a decline in $f_{\rm~gas}$ is the main driver of lowered sSFR in bulges compared to disks in green valley galaxies. Within the disks, there exist common correlations between the sSFR and SFE and between sSFR and $f_{\rm~gas}$ on kpc scales -- the local SFE or $f_{\rm~gas}$ in the disks declines with local sSFR. Our results support a picture in which the sSFR in bulges is primarily controlled by $f_{\rm~gas}$, whereas both SFE and $f_{\rm~gas}$ play a role in lowering the sSFR in disks. A larger sample is required to confirm if the trend established in this work is representative of green valley as a whole.

astro-ph.GA↗

arXiv2017

Quantifying Facial Age by Posterior of Age Comparisons

We introduce a novel approach for annotating large quantity of in-the-wild facial images with high-quality posterior age distribution as labels. Each posterior provides a probability distribution of estimated ages for a face. Our approach is motivated by observations that it is easier to distinguish who is the older of two people than to determine the person's actual age. Given a reference database with samples of known ages and a dataset to label, we can transfer reliable annotations from the former to the latter via human-in-the-loop comparisons. We show an effective way to transform such comparisons to posterior via fully-connected and SoftMax layers, so as to permit end-to-end training in a deep network. Thanks to the efficient and effective annotation approach, we collect a new large-scale facial age dataset, dubbed `MegaAge', which consists of 41,941 images. Data can be downloaded from our project page mmlab.ie.cuhk.edu.hk/projects/MegaAge and github.com/zyx2012/Age_estimation_BMVC2017. With the dataset, we train a network that jointly performs ordinal hyperplane classification and posterior distribution learning. Our approach achieves state-of-the-art results on popular benchmarks such as MORPH2, Adience, and the newly proposed MegaAge.

cs.CV↗

arXiv2017

Improving Secrecy with Nearly Collinear Main and Wiretap Channels via a Cooperative Jamming Relay

In physical layer security (PHY-security), the frequently observed high correlation between the main and wiretap channels can cause a significant loss of secrecy. This paper investigates a slow fading scenario, where a transmitter (Alice) sends a confidential message to a legitimate receiver (Bob) while a passive eavesdropper (Eve) attempts to decode the message from its received signal. It is assumed that Alice is equipped with multiple antennas while Bob and Eve each have a single antenna (i.e., a MISOSE system). In a MISOSE system, high correlation results in nearly collinear main and wiretap channel vectors, which help Eve to see and intercept confidential information. Unfortunately, the signal processing techniques at Alice, such as beamforming and artificial noise (AN), are helpless, especially in the extreme case of completely collinear main and wiretap channel vectors. On this background, we first investigate the achievable secrecy outage probability via beamforming and AN at Alice with the optimal power allocation between the information-bearing signal and AN. Then, an ingenious model, in which a cooperative jamming relay (Relay) is introduced, is proposed to effectively mitigate the adverse effects of high correlation. Based on the proposed model, the power allocation between the information-bearing signal at Alice and the AN at Relay is also studied to maximize secrecy. Finally, to validate our proposed schemes, numerical simulations are conducted, and the results show that a significant performance gain with respect to secrecy is achieved.

cs.IT↗

arXiv2017

Constraining the HI-Halo Mass Relation From Galaxy Clustering

We study the dependence of galaxy clustering on atomic gas mass using a sample of $\sim$16,000 galaxies with redshift in the range of $0.0025 10^8M_{\odot}$, drawn from the 70% complete sample of the Arecibo Legacy Fast ALFA survey. We construct subsamples of galaxies with $M_{\rm HI}$ above different thresholds, and make volume-limited clustering measurements in terms of three statistics: the projected two-point correlation function, the projected cross-correlation function with respect to a reference sample selected from the Sloan Digital Sky Survey, and the redshift-space monopole moment. In contrast to previous studies, which found no/weak HI-mass dependence, we find both the clustering amplitude on scales above a few Mpc and the bias factors to increase significantly with increasing HI mass for subsamples with HI mass thresholds above $10^9M_{\odot}$. For HI mass thresholds below $10^9M_{\odot}$, while the measurements have large uncertainties caused by the limited survey volume and sample size, the inferred galaxy bias factors are systematically lower than the minimum halo bias factor from mass-selected halo samples. The simple halo model, in which galaxy content is only determined by halo mass, has difficulties in interpreting the clustering measurements of the HI-selected samples. We extend the simple model by including the halo formation time as an additional parameter. A model that puts HI-rich galaxies into halos that formed late can reproduce the clustering measurements reasonably well. We present the implications of our best-fitting model on the correlation of HI mass with halo mass and formation time, as well as the halo occupation distributions and HI mass functions for central and satellite galaxies. These results are compared with the predictions from semi-analytic galaxy formation models and hydrodynamic galaxy formation simulations.

astro-ph.GA↗

arXiv2017

Merge or Not? Learning to Group Faces via Imitation Learning

Given a large number of unlabeled face images, face grouping aims at clustering the images into individual identities present in the data. This task remains a challenging problem despite the remarkable capability of deep learning approaches in learning face representation. In particular, grouping results can still be egregious given profile faces and a large number of uninteresting faces and noisy detections. Often, a user needs to correct the erroneous grouping manually. In this study, we formulate a novel face grouping framework that learns clustering strategy from ground-truth simulated behavior. This is achieved through imitation learning (a.k.a apprenticeship learning or learning by watching) via inverse reinforcement learning (IRL). In contrast to existing clustering approaches that group instances by similarity, our framework makes sequential decision to dynamically decide when to merge two face instances/groups driven by short- and long-term rewards. Extensive experiments on three benchmark datasets show that our framework outperforms unsupervised and supervised baselines.

cs.CV↗

arXiv2017

The SDSS-IV MaNGA Sample: Design, Optimization, and Usage Considerations

We describe the sample design for the SDSS-IV MaNGA survey and present the final properties of the main samples along with important considerations for using these samples for science. Our target selection criteria were developed while simultaneously optimizing the size distribution of the MaNGA integral field units (IFUs), the IFU allocation strategy, and the target density to produce a survey defined in terms of maximizing S/N, spatial resolution, and sample size. Our selection strategy makes use of redshift limits that only depend on i-band absolute magnitude ($M_i$), or, for a small subset of our sample, $M_i$ and color (NUV-i). Such a strategy ensures that all galaxies span the same range in angular size irrespective of luminosity and are therefore covered evenly by the adopted range of IFU sizes. We define three samples: the Primary and Secondary samples are selected to have a flat number density with respect to $M_i$ and are targeted to have spectroscopic coverage to 1.5 and 2.5 effective radii (Re), respectively. The Color-Enhanced supplement increases the number of galaxies in the low-density regions of color-magnitude space by extending the redshift limits of the Primary sample in the appropriate color bins. The samples cover the stellar mass range $5\times10^8 \leq M_* \leq 3\times10^{11} M_{\odot}$ and are sampled at median physical resolutions of 1.37 kpc and 2.5 kpc for the Primary and Secondary samples respectively. We provide weights that will statistically correct for our luminosity and color-dependent selection function and IFU allocation strategy, thus correcting the observed sample to a volume limited sample.

astro-ph.GA↗

arXiv2017

The peculiar filamentary HI structure of NGC 6145

In this paper, we report the peculiar HI morphology of the cluster spiral galaxy NGC 6145, which has a 150 kpc HI filament on one side that is nearly parallel to its major axis. This filament is made up of several HI clouds and the diffuse HI gas between them, with no optical counterparts. We compare its HI distribution with other one-sided HI distributions in the literature, and find that the overall HI distribution is very different from the typical tidal and ram-pressure stripped HI shape, and its morphology is inconsistent with being a pure accretion event. Only about 30% of the total HI gas is anchored on the stellar disk, while most of HI gas forms the filament in the west. At a projected distance of 122 kpc, we find a massive elliptical companion (NGC 6146) with extended radio emission, whose axis points to an HI gap in NGC 6145. The velocity of the HI filament shows an overall light-of- sight motion of 80 to 180 km/s with respect to NGC 6145. Using the long-slit spectra of NGC 6145 along its major stellar axis, we find that some outer regions show enhanced star formation, while in contrast, almost no star formation activities are found in its center (less than 2 kpc). Pure accretion, tidal or ram-pressure stripping is not likely to produce the observed HI filament. An alternative explanation is the jet-stripping from NGC 6146, although direct evidence for a jet-cold gas interaction has not been found.

astro-ph.GA↗

arXiv2017

Optimized spin-injection efficiency and spin MOSFET operation based on low-barrier ferromagnet/insulator/n-Si tunnel contact

We theoretically investigate the spin injection in different FM/I/n-Si tunnel contacts by using the lattice NEGF method. We find that the tunnel contacts with low barrier materials such as TiO$_2$ and Ta$_{2}$O$_{5}$, have much lower resistances than the conventional barrier materials, resulting in a wider and attainable optimum parameters window for improving the spin injection efficiency and MR ratio of a vertical spin MOSFET. Additionally, we find the spin asymmetry coefficient of TiO$_2$ tunnel contact has a negative value, while that of Ta$_{2}$O$_{5}$ contact can be tuned between positive and negative values, by changing the parameters.

cond-mat.mes-hall↗

arXiv2017

Regularizing Model Complexity and Label Structure for Multi-Label Text Classification

Multi-label text classification is a popular machine learning task where each document is assigned with multiple relevant labels. This task is challenging due to high dimensional features and correlated labels. Multi-label text classifiers need to be carefully regularized to prevent the severe over-fitting in the high dimensional space, and also need to take into account label dependencies in order to make accurate predictions under uncertainty. We demonstrate significant and practical improvement by carefully regularizing the model complexity during training phase, and also regularizing the label search space during prediction phase. Specifically, we regularize the classifier training using Elastic-net (L1+L2) penalty for reducing model complexity/size, and employ early stopping to prevent overfitting. At prediction time, we apply support inference to restrict the search space to label sets encountered in the training set, and F-optimizer GFM to make optimal predictions for the F1 metric. We show that although support inference only provides density estimations on existing label combinations, when combined with GFM predictor, the algorithm can output unseen label combinations. Taken collectively, our experiments show state of the art results on many benchmark datasets. Beyond performance and practical contributions, we make some interesting observations. Contrary to the prior belief, which deems support inference as purely an approximate inference procedure, we show that support inference acts as a strong regularizer on the label prediction structure. It allows the classifier to take into account label dependencies during prediction even if the classifiers had not modeled any label dependencies during training.

stat.ML↗