Many automotive companies leverage deep learning technology to improve object classification and action prediction in their autonomous cars. However, it is difficult to explain how Deep learning (DL) algorithm work, which is crucial for safety-critical decision-making.
The focus of the thesis will include three parts:
1. Literature review of methods to explain DL algorithm;
2. Take Baidu Apollo 3.5 software (https://github.com/ApolloAuto/apollo) as an example, use different explanation methods (such as decision tree , model distillation , feature visualization  etc.) to extract semantic information of perception module and prediction module in Baidu Apollo 3.5 software;
3. Compare the outcome of different explanation methods and evaluate their potential contributions for safety verification.
 N. Frosst and G. Hinton, "Distilling a Neural Network Into a Soft Decision Tree," arXiv preprint arXiv:1711.09784, 2017.
 S. Tan, R. Caruana, G. Hooker, P. Koch, and A. Gordo, "Learning Global Additive Explanations for Neural Nets Using Model Distillation," arXiv preprint arXiv:1801.08640, 2018.
 M. D. Zeiler and R. Fergus, "Visualizing and understanding convolutional networks," 2014, pp. 818-833: Springer.