PAC Confidence Sets for Deep Neural Networks via Calibrated Prediction
PAC Confidence Sets for Deep Neural Networks via Calibrated Prediction
复制标题
DOI：
--
复制
发表时间：
2019-12
期刊：
影响因子：
--
通讯作者：
Sangdon Park;O. Bastani;N. Matni;Insup Lee
中科院分区：
文献类型：
--
作者：
Sangdon Park;O. Bastani;N. Matni;Insup Lee
我们提出了一种结合校正预测和学习理论泛化界的算法来构造具有PAC保证的深度神经网络的置信度集-即给定输入的置信度集包含高概率的真实标签。我们演示了如何使用我们的方法在ResNet for ImageNet上构造PAC置信集，并在一个动力学模型上构造半猎豹强化学习问题。
We propose an algorithm combining calibrated prediction and generalization bounds from learning theory to construct confidence sets for deep neural networks with PAC guarantees---i.e., the confidence set for a given input contains the true label with high probability. We demonstrate how our approach can be used to construct PAC confidence sets on ResNet for ImageNet, and on a dynamics model the half-cheetah reinforcement learning problem.