arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1903.07348stat.MLcs.LG
收藏

关于深度集合学习和聚合的选择

On Deep Set Learning and the Choice of Aggregations

Volkswagen Group Machine Learning Research Lab(大众集团机器学习研究实验室)

Maximilian Soelch, Adnan Akhundov, Patrick van der Smagt, Justin Bayer

摘要:

最近,它已被证明,许多功能集可以表示的和分解。这些分解很容易适用于神经近似,将神经网络的适用性扩展到集值输入-深度集学习。这项工作研究了Deep Set架构的核心组件:聚合函数。我们建议并研究常用聚合函数的替代方案,包括可学习的经常性聚合函数。从经验上讲,我们证明了深度集网络对聚合函数的选择高度敏感:除了提高性能外,我们发现可学习的聚合降低了超参数敏感性,并更好地推广到分布外的输入大小。

英文摘要:

Recently, it has been shown that many functions on sets can be represented by sum decompositions. These decompositons easily lend themselves to neural approximations, extending the applicability of neural nets to set-valued inputs---Deep Set learning. This work investigates a core component of Deep Set architecture: aggregation functions. We suggest and examine alternatives to commonly used aggregation functions, including learnable recurrent aggregation functions. Empirically, we show that the Deep Set networks are highly sensitive to the choice of aggregation functions: beyond improved performance, we find that learnable aggregations lower hyper-parameter sensitivity and generalize better to out-of-distribution input size.

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。

↑