arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2012.14873cs.LGstat.ML
收藏

孪生神经网络回归

Twin Neural Network Regression

Perimeter Institute for Theoretical Physics(圆周理论物理研究所);University of Ottawa(渥太华大学);Vector Institute for Artificial Intelligence(向量人工智能研究所);University of Waterloo(滑铁卢大学);National Research Council of Canada(加拿大国家研究委员会)

Sebastian J. Wetzel, Kevin Ryczko, Roger G. Melko, Isaac Tamblyn

摘要:

介绍了孪生神经网络(TNN)回归方法。该方法预测两个不同数据点的目标值之间的差异,而不是目标本身。然后,通过在未看见的数据点的目标和所有训练数据点的目标之间的所有预测差异的集合上求平均来获得传统回归问题的解。尽管集合的产生通常是昂贵的,但TNN回归本质上创建了两倍于训练集大小的预测集合,同时只训练单个神经网络。由于集成已被证明是更准确的比单一的模型,这个属性自然转移到TNN回归。我们表明,与其他最先进的方法相比,TNN能够竞争或产生更准确的预测不同的数据集。此外,TNN回归受到自洽条件的约束。我们发现,违反这些条件提供了一个估计的预测不确定性。

英文摘要:

We introduce twin neural network (TNN) regression. This method predicts differences between the target values of two different data points rather than the targets themselves. The solution of a traditional regression problem is then obtained by averaging over an ensemble of all predicted differences between the targets of an unseen data point and all training data points. Whereas ensembles are normally costly to produce, TNN regression intrinsically creates an ensemble of predictions of twice the size of the training set while only training a single neural network. Since ensembles have been shown to be more accurate than single models this property naturally transfers to TNN regression. We show that TNNs are able to compete or yield more accurate predictions for different data sets, compared to other state-of-the-art methods. Furthermore, TNN regression is constrained by self-consistency conditions. We find that the violation of these conditions provides an estimate for the prediction uncertainty.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑