指导数据收集和模型改善的算法(策略) yes no yes yes no no yes no yes no yes no is the performance on the training set acceptable? make some changes on your AI architecture:e.g. using broader and wider ANN, fine-tuning some important hyper-parameters, such as learning rate and batch size is performance on the test set acceptable? is performance on the training set better? achieve reliable model can more high-quality data be found at an acceptable cost? start over, collect cleaner data or a richer set of feature to improve the quality of training data collect more data 1.augment training set:e.g.multi-task learning procedures2.use special architectures and model regularization:e.g. dropout re-train model is performance on the test set acceptable? achieve reliable model is collecting more data practicable? collect more data improve the learning algorithm