arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1909.04013cs.LGcond-mat.dis-nncond-mat.stat-mechstat.ML
收藏

超参数空间中非局部路径优化训练深度神经网络

Training Deep Neural Networks by optimizing over nonlocal paths in hyperparameter space

SoftServe Inc.(SoftServe公司)

Vlad Pushkarov, Jonathan Efroni, Mykola Maksymenko, Maciej Koch-Janusz

摘要:

超参数优化是深层结构训练中的一个实际问题,也是一个有趣的理论问题。尽管最近取得了许多进展,但最常用的方法几乎普遍涉及训练模型的多个解耦副本,实际上是对超参数空间进行采样。我们表明,在一个可以忽略不计的额外计算成本,结果可以改善采样非局部路径,而不是超参数空间中的点。为此,我们将超参数解释为控制训练中相关噪声的水平,它可以映射到有效温度。模型的通常独立的实例被耦合,并允许使用统计物理学的完善的并行回火技术在整个训练过程中交换它们的超参数。然后,每个模拟对应于联合超参数/模型参数空间中的唯一路径或历史。我们提供了我们的方法的实证测试,特别是辍学和学习率优化。我们观察到更快的训练和更好的抗过拟合能力,并显示绝对验证误差的系统性降低,优于基准测试结果。

英文摘要:

Hyperparameter optimization is both a practical issue and an interesting theoretical problem in training of deep architectures. Despite many recent advances the most commonly used methods almost universally involve training multiple and decoupled copies of the model, in effect sampling the hyperparameter space. We show that at a negligible additional computational cost, results can be improved by sampling nonlocal paths instead of points in hyperparameter space. To this end we interpret hyperparameters as controlling the level of correlated noise in training, which can be mapped to an effective temperature. The usually independent instances of the model are coupled and allowed to exchange their hyperparameters throughout the training using the well established parallel tempering technique of statistical physics. Each simulation corresponds then to a unique path, or history, in the joint hyperparameter/model-parameter space. We provide empirical tests of our method, in particular for dropout and learning rate optimization. We observed faster training and improved resistance to overfitting and showed a systematic decrease in the absolute validation error, improving over benchmark results.

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。

↑