多任务Lasso回归法在恒星光谱物理参量估计中的应用

发布时间：2019-01-25 10:43

【摘要】：多任务学习方法在机器学习、计算机视觉、人工智能领域已得到广泛关注,利用任务间的相关性,将多个任务同时学习的效果优于每个任务单独学习的情况.采用多任务Lasso回归法(Multi-task Lasso Regression)用于恒星光谱物理参量的估计,不仅可以获取不同物理参量间的共同的特征信息,而且也可以很好地保留不同物理参量的特有的补充信息.使用恒星大气模拟模型合成光谱库ELODIE中的光谱数据和美国大型巡天项目Sloan发布的SDSS实测光谱数据进行实验,模型估算精度优于相关文献中的方法,特别是对重力加速度(lg g)和化学丰度([Fe/H])的估计.实验中通过改变光谱的分辨率,施加不同信噪比(SNR)的噪声,来说明模型的稳定性强.结果表明,模型精度受光谱分辨率和噪声的影响,但噪声对其影响更大,可见,多任务Lasso回归法不仅操作简便,稳定性强,而且也提高了模型的整体预测精度.
[Abstract]:Multi-task learning methods have been paid more attention in machine learning, computer vision and artificial intelligence. Using the correlation between tasks, the effect of multi-task learning at the same time is better than that of each task alone. Using multitask Lasso regression (Multi-task Lasso Regression) to estimate the spectral physical parameters of stars can not only obtain the common characteristic information among different physical parameters, but also preserve the special supplementary information of different physical parameters. Using the stellar atmosphere simulation model to synthesize the spectral data from the spectral library ELODIE and the measured spectral data from the SDSS published by Sloan, a large-scale survey project in the United States, the accuracy of the model estimation is better than that of the methods in the relevant literature. In particular, the estimation of the gravitational acceleration (lg g) and the chemical abundance ([Fe/H]). The stability of the model is demonstrated by changing the spectral resolution and adding different SNR (SNR) noises. The results show that the precision of the model is affected by spectral resolution and noise, but the noise is more significant. It is obvious that the multitask Lasso regression method is not only easy to operate and stable, but also improves the overall prediction accuracy of the model.
【作者单位】：暨南大学信息科学技术学院;
【基金】：教育部人文社会科学研究一般项目(11YJAZH118)资助
【分类号】：P144.1

【参考文献】