Blockwise coordinate descent procedures for the multi-task lasso, with applications to neural semantic basis discovery
We develop a cyclical blockwise coordinate descent algorithm for the multi-task Lasso that efficiently solves problems with thousands of features and tasks. The main result shows that a closed-form Winsorization operator can be obtained for the sup-norm penalized least squares regression. This allows the algorithm to find solutions to very large-scale problems far more efficiently than existing methods. This result complements the pioneering work of Friedman, et al. (2007) for the single-task Lasso. As a case study, we use the multi-task Lasso as a variable selector to discover a semantic basis for predicting human neural activation. The learned solution outperforms the standard basis for this task on the majority of test participants, while requiring far fewer assumptions about cognitive neuroscience. We demonstrate how this learned basis can yield insights into how the brain represents the meanings of words.
Prices, Wages and the U.S. NAIRU in the 1990s
Using quarterly macro data and annual state panel data, we examine various explanations of the low rate of price inflation, strong real wage growth, and low rate of unemployment in the U.S. economy during the late 1990s. Many of these explanations imply shifts in the coefficients of price and wage Phillips curves. We find, however, that once one accounts for the univariate trends in the unemployment rate and in the rate of productivity growth, these coefficients are stable. This suggests that many explanations, such as persistent beneficial supply shocks, changes in firms' pricing power, changes in price expectations arising from shifts in Fed policy, and changes in wage setting behavior miss the mark. Rather, we suggest that explanations of movements of wages, prices and unemployment over the 1990s, and indeed over the past forty years, must focus on understanding the univariate trends in the unemployment rate and in productivity growth and, perhaps, the relation between the two.
Comparison of eight published static finite element models of the intact lumbar spine: predictive power of models improves when combined together.
Finite element (FE) model studies have made important contributions to our understanding of functional biomechanics of the lumbar spine. However, if a model is used to answer clinical and biomechanical questions over a certain population, their inherently large inter-subject variability has to be considered. Current FE model studies, however, generally account only for a single distinct spinal geometry with one set of material properties. This raises questions concerning their predictive power, their range of results and on their agreement with in vitro and in vivo values. Eight well-established FE models of the lumbar spine (L1-5) of different research centers around the globe were subjected to pure and combined loading modes and compared to in vitro and in vivo measurements for intervertebral rotations, disc pressures and facet joint forces. Under pure moment loading, the predicted L1-5 rotations of almost all models fell within the reported in vitro ranges, and their median values differed on average by only 2° for flexion-extension, 1° for lateral bending and 5° for axial rotation. Predicted median facet joint forces and disc pressures were also in good agreement with published median in vitro values. However, the ranges of predictions were larger and exceeded those reported in vitro, especially for the facet joint forces. For all combined loading modes, except for flexion, predicted median segmental intervertebral rotations and disc pressures were in good agreement with measured in vivo values. In light of high inter-subject variability, the generalization of results of a single model to a population remains a concern. This study demonstrated that the pooled median of individual model results, similar to a probabilistic approach, can be used as an improved predictive tool in order to estimate the response of the lumbar spine.
The 2012 free and open source GIS software map - A guide to facilitate research, development, and adoption
Over the last decade an increasing number of free and open source software projects have been founded that concentrate on developing several types of software for geographic data collection, storage, analysis and visualization. We first identify the drivers of such software projects and identify different types of geographic information software, e.g. desktop GIS, remote sensing software, server GIS etc. We then list the major projects for each software category. Afterwards we discuss the points that should be considered if free and open source software is to be selected for use in business and research, such as software functionality, license types and their restrictions, developer and user community characteristics, etc. Finally possible future developments are addressed.
Carbon nanotube mode lockers with enhanced nonlinearity via evanescent field interaction in D-shaped fibers.
We demonstrate a novel passive mode-locking scheme for pulsed lasers enhanced by the interaction of carbon nanotubes (CNTs) with the evanescent field of propagating light in a D-shaped optical fiber. The scheme features all-fiber operation as well as a long lateral interaction length, which guarantees a strong nonlinear effect from the nanotubes. Mode locking is achieved with less than 30% of the CNTs compared with the amount of nanotubes used for conventional schemes. Our method also ensures the preservation of the original morphology of the individual CNTs. The demonstrated pulsed laser with our CNT mode locker has a repetition rate of 5.88 MHz and a temporal pulse width of 470 fs.
Low rates of colorectal, cervical, and breast cancer screening in Asian Americans compared with non‐Hispanic whites
Asian Americans have lower cancer screening rates compared with non‐Hispanic whites (NHWs). Little is known about mechanisms that underlie disparities in cancer screening. The objectives of the current study were 1) to determine the relation between nativity, years in the United States, language, and cancer screening in NHWs and Asian Americans, independent of access to care and 2) to determine whether Asians reported different reasons than NHWs for not obtaining cancer screening.
ipyrad: Interactive assembly and analysis of RADseq datasets
SUMMARY ipyrad is a free and open source tool for assembling and analyzing restriction-site associated DNA sequence (RADseq) datasets using de novo and/or reference-based approaches. It is designed to be massively scalable to hundreds of taxa and thousands of samples, and can be efficiently parallelized on high performance computing clusters. It is available as both a command line interface (CLI) and as a Python package with an application programming interface (API), the latter of which can be used interactively to write complex, reproducible scripts, and implement a suite of downstream analysis tools. AVAILABILITY AND IMPLEMENTATION ipyrad is a free and open source program written in Python. Source code is available from the GitHub repository (https://github.com/dereneaton/ipyrad/), and Linux and MacOS installs are distributed through the conda package manager. SUPPLEMENTARY INFORMATION Complete documentation, including numerous tutorials, and Jupyter notebooks demonstrating example assemblies and applications of downstream analysis tools are available online: https://ipyrad.readthedocs.io/.
Multi-Task CNN Model for Attribute Prediction
This paper proposes a joint multi-task learning algorithm to better predict attributes in images using deep convolutional neural networks (CNN). We consider learning binary semantic attributes through a multi-task CNN model, where each CNN will predict one binary attribute. The multi-task learning allows CNN models to simultaneously share visual knowledge among different attribute categories. Each CNN will generate attribute-specific feature representations, and then we apply multi-task learning on the features to predict their attributes. In our multi-task framework, we propose a method to decompose the overall model's parameters into a latent task matrix and combination matrix. Furthermore, under-sampled classifiers can leverage shared statistics from other classifiers to improve their performance. Natural grouping of attributes is applied such that attributes in the same group are encouraged to share more knowledge. Meanwhile, attributes in different groups will generally compete with each other, and consequently share less knowledge. We show the effectiveness of our method on two popular attribute datasets.
Heterogeneous Multi-task Learning for Human Pose Estimation with Deep Convolutional Neural Network
We propose a heterogeneous multi-task learning framework for human pose estimation from monocular images using a deep convolutional neural network. In particular, we simultaneously learn a human pose regressor and sliding-window body-part and joint-point detectors in a deep network architecture. We show that including the detection tasks helps to regularize the network, directing it to converge to a good solution. We report competitive and state-of-art results on several datasets. We also empirically show that the learned neurons in the middle layer of our network are tuned to localized body parts.
Multi-task warped Gaussian process for personalized age estimation
Automatic age estimation from facial images has aroused research interests in recent years due to its promising potential for some computer vision applications. Among the methods proposed to date, personalized age estimation methods generally outperform global age estimation methods by learning a separate age estimator for each person in the training data set. However, since typical age databases only contain very limited training data for each person, training a separate age estimator using only training data for that person runs a high risk of overfitting the data and hence the prediction performance is limited. In this paper, we propose a novel approach to age estimation by formulating the problem as a multi-task learning problem. Based on a variant of the Gaussian process (GP) called warped Gaussian process (WGP), we propose a multi-task extension called multi-task warped Gaussian process (MTWGP). Age estimation is formulated as a multi-task regression problem in which each learning task refers to estimation of the age function for each person. While MTWGP models common features shared by different tasks (persons), it also allows task-specific (person-specific) features to be learned automatically. Moreover, unlike previous age estimation methods which need to specify the form of the regression functions or determine many parameters in the functions using inefficient methods such as cross validation, the form of the regression functions in MTWGP is implicitly defined by the kernel function and all its model parameters can be learned from data automatically. We have conducted experiments on two publicly available age databases, FG-NET and MORPH. The experimental results are very promising in showing that MTWGP compares favorably with state-of-the-art age estimation methods.
Free and open source development practices in the game community
The free and open source software (FOSS) approach lets community of like-minded participants develop software systems and related artifacts that are shared freely instead of offered as closed-source commercial products. Free (as in freedom) software and open source are closely related but slightly different approaches and licensing schemes for developing publicly shared software. FOSS development communities don't seem to adopt modern software engineering processes. FOSS communities develop software that's extremely valuable, generally reliable, globally distributed, made available for acquisition at little or no cost, and readily used in its community. Free and open source software development practices gives rise to new view of how complex software systems can be constructed, deployed, and evolved. They rely on lean electronic communication media, virtual project management, and version management mechanisms to coordinate globally dispersed development efforts. These FOSS processes offer new directions for developing complex software systems. We look at the FOSS computer game community to provide examples of common development processes and practices.
Politics and the Structural Dependence of the State in Democratic Capitalist Nations
I explore empirically a central claim of the structural dependence thesis, namely, that capitalists' ability to disinvest fundamentally conditions policy choices in democratic capitalist systems. Utilizing time-series data for 16 affluent democracies from 1965 to 1984, I find that, indeed, low rates of business investment are associated with reductions in corporate tax burdens and that these reductions are more pronounced in periods of economic crisis. Moreover, low rates of capital formation engender cuts in personal income taxes during periods of economic stress. However, I also find that the magnitude of responsiveness of taxation to low rates of investment is relatively small and that analyses of the political context of investment and taxation indicate that governments have choices. The responsiveness of corporate tax burdens to capital formation may, under some governments, be part of a policy mix designed to maintain adequate investment and to address the demands of core constituencies.
Accelerated Gradient Method for Multi-task Sparse Learning Problem
Many real world learning problems can be recast as multi-task learning problems which utilize correlations among different tasks to obtain better generalization performance than learning each task individually. The feature selection problem in multi-task setting has many applications in fields of computer vision, text classification and bio-informatics. Generally, it can be realized by solving a L-1-infinity regularized optimization problem. And the solution automatically yields the joint sparsity among different tasks. However, due to the nonsmooth nature of the L-1-infinity norm, there lacks an efficient training algorithm for solving such problem with general convex loss functions. In this paper, we propose an accelerated gradient method based on an ``optimal'' first order black-box method named after Nesterov and provide the convergence rate for smooth convex loss functions. For nonsmooth convex loss functions, such as hinge loss, our method still has fast convergence rate empirically. Moreover, by exploiting the structure of the L-1-infinity ball, we solve the black-box oracle in Nesterov's method by a simple sorting scheme. Our method is suitable for large-scale multi-task learning problem since it only utilizes the first order information and is very easy to implement. Experimental results show that our method significantly outperforms the most state-of-the-art methods in both convergence speed and learning accuracy.
Surface plasmon resonance sensor based on D-shaped microstructured optical fiber with hollow core.
To solve the phase matching and analyte filling problems in the microstructured optical fiber (MOF)-based surface plasmon resonance (SPR) sensors, we present the D-shaped hollow core MOF-based SPR sensor. The air hole in the fiber core can lower the refractive index of a Gaussian-like core mode to match with that of a plasmon mode. The analyte is deposited directly onto the D-shaped flat surface instead of filling the fiber holes. We numerically investigate the effect of the air hole in the core on the SPR sensing performance, and identify the sensor sensitivity on wavelength, amplitude and phase. This work allows us to determine the feasibility of using the D-shaped hollow-core MOFs to develop a high-sensitivity, real-time and distributed SPR sensor.
Approximating optimal spare capacity allocation by successive survivable routing
The design of survivable mesh based communication networks has received considerable attention in recent years. One task is to route backup paths and allocate spare capacity in the network to guarantee seamless communications services survivable to a set of failure scenarios. This is a complex multi-constraint optimization problem, called the spare capacity allocation (SCA) problem. This paper unravels the SCA problem structure using a matrix-based model, and develops a fast and efficient approximation algorithm, termed successive survivable routing (SSR). First, per-flow spare capacity sharing is captured by a spare provision matrix (SPM) method. The SPM matrix has a dimension the number of failure scenarios by the number of links. It is used by each demand to route the backup path and share spare capacity with other backup paths. Next, based on a special link metric calculated from SPM, SSR iteratively routes/updates backup paths in order to minimize the cost of total spare capacity. A backup path can be further updated as long as it is not carrying any traffic. Furthermore, the SPM method and SSR algorithm are generalized from protecting all single link failures to any arbitrary link failures such as those generated by Shared Risk Link Groups or all single node failures. Numerical results comparing several SCA algorithms show that SSR has the best trade-off between solution optimality and computation speed.
Free and open source geographic information tools for landscape ecology
Geographic Information tools (GI tools) have become an essential component of research in landscape ecology. In this article we review the use of GIS (Geographic Information Systems) and GI tools in landscape ecology, with an emphasis on free and open source software (FOSS) projects. Specifically, we introduce the background and terms related to the free and open source software movement, then compare eight FOSS desktop GIS with proprietary GIS to analyse their utility for landscape ecology research. We also provide a summary of related landscape analysis FOSS applications, and extensions. Our results indicate that (i) all eight GIS provide the basic GIS functionality needed in landscape ecology, (ii) they all facilitate customisation, and (iii) they all provide good support via forums and email lists. Drawbacks that have been identified are related to the fact that most projects are relatively young. This currently affects the size of their user and developer communities, and their ability to include advanced spatial analysis functions and up-to-date documentation. However, we expect these drawbacks to be addressed over time, as systems mature. In general, we see great potential for the use of free and open source desktop GIS in landscape ecology research and advocate concentrated efforts by the landscape ecology community towards a common, customisable and free research platform.
Efficiently Identifying Task Groupings for Multi-Task Learning
Multi-task learning can leverage information learned by one task to benefit the training of other tasks. Despite this capacity, naively training all tasks together in one model often degrades performance, and exhaustively searching through combinations of task groupings can be prohibitively expensive. As a result, efficiently identifying the tasks that would benefit from training together remains a challenging design question without a clear solution. In this paper, we suggest an approach to select which tasks should train together in multi-task learning models. Our method determines task groupings in a single run by training all tasks together and quantifying the effect to which one task's gradient would affect another task's loss. On the large-scale Taskonomy computer vision dataset, we find this method can decrease test loss by 10.0% compared to simply training all tasks together while operating 11.6 times faster than a state-of-the-art task grouping method.
Implications of rRNA Operon Copy Number and Ribosome Content in the Marine Oligotrophic Ultramicrobacterium Sphingomonassp. Strain RB2256
Sphingomonas sp. strain RB2256 is a representative of the dominant class of ultramicrobacteria that are present in marine oligotrophic waters. In this study we examined the rRNA copy number and ribosome content of RB2256 to identify factors that may be associated with the relatively low rate of growth exhibited by the organism. It was found that RB2256 contains a single copy of the rRNA operon, in contrast to Vibrio spp., which contain more than eight copies. The maximum number of ribosomes per cell was observed during mid-log phase; however, this maximum content was low compared to those of faster-growing, heterotrophic bacteria (approximately 8% of the maximum ribosome content of Escherichia coli with a growth rate of 1.5 h−1). The low number of ribosomes per cell appears to correlate with the low rate of growth (0.16 to 0.18 h−1) and the presence of a single copy of the rRNA operon. However, on the basis of cell volume, RB2256 appears to have a higher concentration of ribosomes than E. coli (approximately double that of E. coli with a growth rate of 1.5 h−1). Ribosome numbers reached maximum levels during mid-log-phase growth but decreased rapidly to 10% of maximum during late log phase through 7 days of starvation. The cells in late log phase and at the onset of starvation displayed an immediate response to a sudden addition of excess glucose (3 mM). This result demonstrates that a ribosome content 10% of maximum is sufficient to allow cells to immediately respond to nutrient upshift and achieve maximum rates of growth. These data indicate that the bulk of the ribosome pool is not required for protein synthesis and that ribosomes are not the limiting factor contributing to a low rate of growth. Our findings show that the regulation of ribosome content, the number of ribosomes per cell, and growth rate responses in RB2256 are fundamentally different from those characteristics in fast-growing heterotrophs like E. coliand that they may be characteristics typical of oligotrophic ultramicrobacteria.
BioSig: The Free and Open Source Software Library for Biomedical Signal Processing
BioSig is an open source software library for biomedical signal processing. The aim of the BioSig project is to foster research in biomedical signal processing by providing free and open source software tools for many different application areas. Some of the areas where BioSig can be employed are neuroinformatics, brain-computer interfaces, neurophysiology, psychology, cardiovascular systems, and sleep research. Moreover, the analysis of biosignals such as the electroencephalogram (EEG), electrocorticogram (ECoG), electrocardiogram (ECG), electrooculogram (EOG), electromyogram (EMG), or respiration signals is a very relevant element of the BioSig project. Specifically, BioSig provides solutions for data acquisition, artifact processing, quality control, feature extraction, classification, modeling, and data visualization, to name a few. In this paper, we highlight several methods to help students and researchers to work more efficiently with biomedical signals.
Hölder continuity of the gradient of p(x)-harmonic mappings
We prove that local minimizers u : R n → R N of the functional ∫ ∣D u ( x )∣ p(r) d x are of class C 1, α for some α > 0, provided p(x) > 1 is Holder continuous.
neural network differential equation deep learning convolutional neural network convolutional neural software development deep neural network feature selection open source deep neural optical network learning approach geographic information system distributed generation geographic information stochastic differential equation power allocation open source software stochastic differential deep convolutional neural source software deep convolutional named entity recognition loss function human action space telescope personal area network mental health feature learning wireless personal area elastic optical multi-task learning personal area elastic optical network network reconfiguration wireless personal learning problem hubble space telescope learning task software developer high rate water pressure hubble space doubly stochastic open source tool source software development free and open capacity allocation low rate source software project group lasso path protection sparse learning backup path rate wireles open source gi spare capacity optimal reconfiguration source gi multi-task deep backward doubly stochastic doubly stochastic differential backward doubly big bang-big crunch low rate wireles survivable routing multi-task deep learning guidance sensor bang-big crunch big bang-big multi-task learning approach source software tool deep multi-task deep multi-task learning pose-invariant face high water pressure fine guidance shared backup fine guidance sensor multi-task convolutional neural multi-task learning model source gis software shared backup path multi-task feature bang-big crunch algorithm backup path protection approximating optimal crunch algorithm multi-task learning method software movement multi-task network spare capacity allocation multi-task convolutional multi-task cnn multi-task sparse multi-task feature learning multi-task regression multi-task allocation feature learning algorithm multiple related task multi-task multi-view multi-task learning algorithm soils contaminated