In environmental health research there is often interest in the effect of an exposure on a health outcome assessed on the same day and several subsequent days or lags. Distributed lag nonlinear models (DLNM) are a well-established statistical framework for estimating an exposure-lag-response function. We propose methods to allow for prior information to be incorporated into DLNMs. First, we impose a monotonicity constraint in the exposure-response at lagged time periods which matches with knowledge on how biological mechanisms respond to increased levels of exposures. Second, we introduce variable selection into the DLNM to identify lagged periods of susceptibility with respect to the outcome of interest. The variable selection approach allows for direct application of informative priors on which lags have nonzero association with the outcome. We propose a tree-of-trees model that uses two layers of trees: one for splitting the exposure time frame and one for fitting exposure-response functions over different time periods. We introduce a zero-inflated alternative to the tree splitting prior in Bayesian additive regression trees to allow for lag selection and the addition of informative priors. We develop a computational approach for efficient posterior sampling and perform a comprehensive simulation study to compare our method to existing DLNM approaches. We apply our method to estimate time-lagged extreme temperature relationships with mortality during summer or winter in Chicago, IL.
相關內容
High-dimensional data can often display heterogeneity due to heteroscedastic variance or inhomogeneous covariate effects. Penalized quantile and expectile regression methods offer useful tools to detect heteroscedasticity in high-dimensional data. The former is computationally challenging due to the non-smooth nature of the check loss, and the latter is sensitive to heavy-tailed error distributions. In this paper, we propose and study (penalized) robust expectile regression (retire), with a focus on iteratively reweighted $\ell_1$-penalization which reduces the estimation bias from $\ell_1$-penalization and leads to oracle properties. Theoretically, we establish the statistical properties of the retire estimator under two regimes: (i) low-dimensional regime in which $d \ll n$; (ii) high-dimensional regime in which $s\ll n\ll d$ with $s$ denoting the number of significant predictors. In the high-dimensional setting, we carefully characterize the solution path of the iteratively reweighted $\ell_1$-penalized retire estimation, adapted from the local linear approximation algorithm for folded-concave regularization. Under a mild minimum signal strength condition, we show that after as many as $\log(\log d)$ iterations the final iterate enjoys the oracle convergence rate. At each iteration, the weighted $\ell_1$-penalized convex program can be efficiently solved by a semismooth Newton coordinate descent algorithm. Numerical studies demonstrate the competitive performance of the proposed procedure compared with either non-robust or quantile regression based alternatives.
Understanding dynamics in complex systems is challenging because there are many degrees of freedom, and those that are most important for describing events of interest are often not obvious. The leading eigenfunctions of the transition operator are useful for visualization, and they can provide an efficient basis for computing statistics such as the likelihood and average time of events (predictions). Here we develop inexact iterative linear algebra methods for computing these eigenfunctions (spectral estimation) and making predictions from a data set of short trajectories sampled at finite intervals. We demonstrate the methods on a low-dimensional model that facilitates visualization and a high-dimensional model of a biomolecular system. Implications for the prediction problem in reinforcement learning are discussed.
When data is collected in an adaptive manner, even simple methods like ordinary least squares can exhibit non-normal asymptotic behavior. As an undesirable consequence, hypothesis tests and confidence intervals based on asymptotic normality can lead to erroneous results. We propose a family of online debiasing estimators to correct these distributional anomalies in least squares estimation. Our proposed methods take advantage of the covariance structure present in the dataset and provide sharper estimates in directions for which more information has accrued. We establish an asymptotic normality property for our proposed online debiasing estimators under mild conditions on the data collection process and provide asymptotically exact confidence intervals. We additionally prove a minimax lower bound for the adaptive linear regression problem, thereby providing a baseline by which to compare estimators. There are various conditions under which our proposed estimators achieve the minimax lower bound. We demonstrate the usefulness of our theory via applications to multi-armed bandit, autoregressive time series estimation, and active learning with exploration.
This paper studies the third-order characteristic of nonsingular discrete memoryless channels and the Gaussian channel with a maximal power constraint. The third-order term in our expansions employs a new quantity here called the \emph{channel skewness}, which affects the approximation accuracy more significantly as the error probability decreases. For the Gaussian channel, evaluating Shannon's (1959) random coding and sphere-packing bounds in the central limit theorem (CLT) regime enables exact computation of the channel skewness. For discrete memoryless channels, this work generalizes Moulin's (2017) bounds on the asymptotic expansion of the maximum achievable message set size for nonsingular channels from the CLT regime to include the moderate deviations (MD) regime, thereby refining Altu\u{g} and Wagner's (2014) MD result. For an example binary symmetric channel and most practically important $(n, \epsilon)$ pairs, including $n \in [100, 500]$ and $\epsilon \in [10^{-10}, 10^{-1}]$, an approximation up to the channel skewness is the most accurate among several expansions in the literature. A derivation of the third-order term in the type-II error exponent of binary hypothesis testing in the MD regime is also included; the resulting third-order term is similar to the channel skewness.
Standard bandit algorithms that assume continual reallocation of measurement effort are challenging to implement due to delayed feedback and infrastructural/organizational difficulties. Motivated by practical instances involving a handful of reallocation epochs in which outcomes are measured in batches, we develop a new adaptive experimentation framework that can flexibly handle any batch size. Our main observation is that normal approximations universal in statistical inference can also guide the design of scalable adaptive designs. By deriving an asymptotic sequential experiment, we formulate a dynamic program that can leverage prior information on average rewards. State transitions of the dynamic program are differentiable with respect to the sampling allocations, allowing the use of gradient-based methods for planning and policy optimization. We propose a simple iterative planning method, Residual Horizon Optimization, which selects sampling allocations by optimizing a planning objective via stochastic gradient-based methods. Our method significantly improves statistical power over standard adaptive policies, even when compared to Bayesian bandit algorithms (e.g., Thompson sampling) that require full distributional knowledge of individual rewards. Overall, we expand the scope of adaptive experimentation to settings which are difficult for standard adaptive policies, including problems with a small number of reallocation epochs, low signal-to-noise ratio, and unknown reward distributions.
On a daily basis, data centers process huge volumes of data backed by the proliferation of inexpensive hard disks. Data stored in these disks serve a range of critical functional needs from financial, and healthcare to aerospace. As such, premature disk failure and consequent loss of data can be catastrophic. To mitigate the risk of failures, cloud storage providers perform condition-based monitoring and replace hard disks before they fail. By estimating the remaining useful life of hard disk drives, one can predict the time-to-failure of a particular device and replace it at the right time, ensuring maximum utilization whilst reducing operational costs. In this work, large-scale predictive analyses are performed using severely skewed health statistics data by incorporating customized feature engineering and a suite of sequence learners. Past work suggests using LSTMs as an excellent approach to predicting remaining useful life. To this end, we present an encoder-decoder LSTM model where the context gained from understanding health statistics sequences aid in predicting an output sequence of the number of days remaining before a disk potentially fails. The models developed in this work are trained and tested across an exhaustive set of all of the 10 years of S.M.A.R.T. health data in circulation from Backblaze and on a wide variety of disk instances. It closes the knowledge gap on what full-scale training achieves on thousands of devices and advances the state-of-the-art by providing tangible metrics for evaluation and generalization for practitioners looking to extend their workflow to all years of health data in circulation across disk manufacturers. The encoder-decoder LSTM posted an RMSE of 0.83 during training and 0.86 during testing over the exhaustive 10 year data while being able to generalize competitively over other drives from the Seagate family.
Bayesian persuasion and its derived information design problem has been one of the main research agendas in the economics and computation literature over the past decade. However, when attempting to apply its model and theory, one is often limited by the fact that the sender can only implement very restricted information structures. Moreover, in this case, the sender can possibly achieve higher expected utility by performing a sequence of feasible experiments, where the choice of each experiment depends on the outcomes of all previous experiments. Indeed, it has been well observed that real life persuasions often take place in rounds during which the sender exhibits experiments/arguments sequentially. We study the sender's expected utility maximization using finite and infinite sequences of experiments. For infinite sequences of experiments, we characterize the supremum of the sender's expected utility using a function that generalizes the concave closure definition in the standard Bayesian persuasion problem. With this characterization, we first study a special case where the sender can use feasible experiments to achieve the optimal expected utility of the standard Bayesian persuasion without feasibility constraints, which is a trivial utility upper bound, and establish structural findings about the sender's optimal sequential design in this case. Then we derive conditions under which the sender's optimal sequential design exists; when an optimal sequential design exists, there exists an optimal design that is Markovian, i.e., the choice of the next experiment only depends on the receiver's current belief.
In the present academic landscape, the process of collecting data is slow, and the lax infrastructures for data collaborations lead to significant delays in coming up with and disseminating conclusive findings. Therefore, there is an increasing need for a secure, scalable, and trustworthy data-sharing ecosystem that promotes and rewards collaborative data-sharing efforts among researchers, and a robust incentive mechanism is required to achieve this objective. Reputation-based incentives, such as the h-index, have historically played a pivotal role in the academic community. However, the h-index suffers from several limitations. This paper introduces the SCIENCE-index, a blockchain-based metric measuring a researcher's scientific contributions. Utilizing the Microsoft Academic Graph and machine learning techniques, the SCIENCE-index predicts the progress made by a researcher over their career and provides a soft incentive for sharing their datasets with peer researchers. To incentivize researchers to share their data, the SCIENCE-index is augmented to include a data-sharing parameter. DataCite, a database of openly available datasets, proxies this parameter, which is further enhanced by including a researcher's data-sharing activity. Our model is evaluated by comparing the distribution of its output for geographically diverse researchers to that of the h-index. We observe that it results in a much more even spread of evaluations. The SCIENCE-index is a crucial component in constructing a decentralized protocol that promotes trust-based data sharing, addressing the current inequity in dataset sharing. The work outlined in this paper provides the foundation for assessing scientific contributions in future data-sharing spaces powered by decentralized applications.
This paper considers statistical inference of time-varying network vector autoregression models for large-scale time series. A latent group structure is imposed on the heterogeneous and node-specific time-varying momentum and network spillover effects so that the number of unknown time-varying coefficients to be estimated can be reduced considerably. A classic agglomerative clustering algorithm with normalized distance matrix estimates is combined with a generalized information criterion to consistently estimate the latent group number and membership. A post-grouping local linear smoothing method is proposed to estimate the group-specific time-varying momentum and network effects, substantially improving the convergence rates of the preliminary estimates which ignore the latent structure. In addition, a post-grouping specification test is conducted to verify the validity of the parametric model assumption for group-specific time-varying coefficient functions, and the asymptotic theory is derived for the test statistic constructed via a kernel weighted quadratic form under the null and alternative hypotheses. Numerical studies including Monte-Carlo simulation and an empirical application to the global trade flow data are presented to examine the finite-sample performance of the developed model and methodology.
Classic algorithms and machine learning systems like neural networks are both abundant in everyday life. While classic computer science algorithms are suitable for precise execution of exactly defined tasks such as finding the shortest path in a large graph, neural networks allow learning from data to predict the most likely answer in more complex tasks such as image classification, which cannot be reduced to an exact algorithm. To get the best of both worlds, this thesis explores combining both concepts leading to more robust, better performing, more interpretable, more computationally efficient, and more data efficient architectures. The thesis formalizes the idea of algorithmic supervision, which allows a neural network to learn from or in conjunction with an algorithm. When integrating an algorithm into a neural architecture, it is important that the algorithm is differentiable such that the architecture can be trained end-to-end and gradients can be propagated back through the algorithm in a meaningful way. To make algorithms differentiable, this thesis proposes a general method for continuously relaxing algorithms by perturbing variables and approximating the expectation value in closed form, i.e., without sampling. In addition, this thesis proposes differentiable algorithms, such as differentiable sorting networks, differentiable renderers, and differentiable logic gate networks. Finally, this thesis presents alternative training strategies for learning with algorithms.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191