音声データによる歯磨き効果の評価

在今年的工作中，我们的重点是减轻研究人员设计和培训绩效评估系统的负担。我们提出了一种通过使用深度神经网络进行性能评估的方法。我们的方法通过在进行性能评估时不需要创建定制的特征和中间活动识别模型来简化评估任务。我们还通过提出将长短期记忆自编码器与支持向量机回归模型相结合的方法，解决了在进行性能评估时可用的训练数据数量有限的问题。我们的结果表明，我们的方法能够生成一个工作的回归模型，但性能还不如我们的方法依赖于手工制作的特征。

In this year’s work, we focused on reducing the burden placed on the researchers designing and training performance evaluation systems. We proposed a method for conducting performance evaluation through the use of deep neural networks. Our method simplified the evaluation task by removing the need to create tailored features and intermediate activity recognition models when conducting performance evaluation. We also addressed the limited amount of training data available when conducting performance evaluation by proposing a method of combining long short term memory autoencoders with SVM regression models. Our results indicate that our method is able to generate a working regression model, but the performance is not yet as good as from our method that relied on handcrafted features.