Recently deep learning has successfully achieved state-of-the-art performance on many difficulttasks. Deep neural networks allow for model flexibility and process features without the needof domain knowledge. Advantag...Recently deep learning has successfully achieved state-of-the-art performance on many difficulttasks. Deep neural networks allow for model flexibility and process features without the needof domain knowledge. Advantage learning (A-learning) is a popular method in dynamic treatment regime (DTR). It models the advantage function, which is of direct relevance to optimaltreatment decision. No assumptions on baseline function are made. However, there is a paucityof literature on deep A-learning. In this paper, we present a deep A-learning approach to estimate optimal DTR. We use an inverse probability weighting method to estimate the differencebetween potential outcomes. Parameter sharing of convolutional neural networks (CNN) greatlyreduces the amount of parameters in neural networks, which allows for high scalability. Convexified convolutional neural networks (CCNN) relax the constraints of CNN for optimisation purpose.Different architectures of CNN and CCNN are implemented for contrast function estimation.Both simulation results and application to the STAR*D (Sequenced Treatment Alternatives toRelieve Depression) trial indicate that the proposed methods outperform penalised least squareestimator.展开更多
We present quantum mechanical vibrational computations beyond the harmonic approximation from effective second order perturbative and variation perturbation treatments defined as static approaches, as well as vibratio...We present quantum mechanical vibrational computations beyond the harmonic approximation from effective second order perturbative and variation perturbation treatments defined as static approaches, as well as vibrational analysis from density functional theory molecular dynamics trajectories at 300 and 600 K. The four schemes are compared in terms of prediction of fundamental transitions, and simulation of the corresponding medium infrared spectrum at the same level of theory using the B3LYP/6-31＋G（d,p） description of the electronic structure. We summarize conclusions about advantages and drawbacks of these two approaches and report the main results obtained for semi-rigid and flexible molecules.展开更多
基金This work was supported by National Institutes of Health[5P01CA142538].
文摘Recently deep learning has successfully achieved state-of-the-art performance on many difficulttasks. Deep neural networks allow for model flexibility and process features without the needof domain knowledge. Advantage learning (A-learning) is a popular method in dynamic treatment regime (DTR). It models the advantage function, which is of direct relevance to optimaltreatment decision. No assumptions on baseline function are made. However, there is a paucityof literature on deep A-learning. In this paper, we present a deep A-learning approach to estimate optimal DTR. We use an inverse probability weighting method to estimate the differencebetween potential outcomes. Parameter sharing of convolutional neural networks (CNN) greatlyreduces the amount of parameters in neural networks, which allows for high scalability. Convexified convolutional neural networks (CCNN) relax the constraints of CNN for optimisation purpose.Different architectures of CNN and CCNN are implemented for contrast function estimation.Both simulation results and application to the STAR*D (Sequenced Treatment Alternatives toRelieve Depression) trial indicate that the proposed methods outperform penalised least squareestimator.
文摘We present quantum mechanical vibrational computations beyond the harmonic approximation from effective second order perturbative and variation perturbation treatments defined as static approaches, as well as vibrational analysis from density functional theory molecular dynamics trajectories at 300 and 600 K. The four schemes are compared in terms of prediction of fundamental transitions, and simulation of the corresponding medium infrared spectrum at the same level of theory using the B3LYP/6-31＋G（d,p） description of the electronic structure. We summarize conclusions about advantages and drawbacks of these two approaches and report the main results obtained for semi-rigid and flexible molecules.
