Page 86 - Data Science Algorithms in a Week

70 Olmer Garcia and Cesar Diaz

to low accuracy in the testing process. This model developed provided opportunities to
analyze new research questions such as:

 Will this model work with my country traffic signals – How about the climate
and the cultural environment?
 How to improve performance?
 Is feasible to implement the feedforward process in real time?

CONCLUSION

A brief review of machine learning and the architecture of autonomous vehicles was
discussed in this chapter. It is important to note that the use of machine learning required
two hardware/software systems: one for training in the cloud and the other one in the
autonomous vehicle. Another point to take into account was that modeling by machine
learning using examples requires sufficient data to let machine learning models
generalize at appropriate levels. There are some potential applications for deep learning
in the field of autonomous vehicles. For example, it is possible that a deep learning
neural network becomes the “driver” of the autonomous vehicle: where the inputs are
road conditions and the risk profile of the passenger and the outputs are turning degrees
and speed of the car. Driving scenarios are a good fit for multiclass and multi label
classification problems. The mapping is hidden in the different and multiple hierarchical
layers but deep learning does not need the exact form of the function (if it maps well
from input to output). The results are very promising. However, safety regulations (and
public acceptance) will require numerous tests and validations of the deep learning based
systems to be certified by the respective agencies.

REFERENCES

Amsalu, S., Homaifar, A., Afghah, F., Ramyar, S., & Kurt, A. (2015). Driver behavior
modeling near intersections using support vector machines based on statistical feature
extraction. In 2015 IEEE Intelligent Vehicles Symposium (IV), 1270–1275.
Bahadorimonfared, A., Soori, H., Mehrabi, Y., Delpisheh, A., Esmaili, A., Salehi, M., &
Bakhtiyari, M. (2013). Trends of fatal road traffic injuries in Iran (2004–2011). PloS
one, 8(5):e65198.
Bedoya, O. G. (2016). Análise de risco para a cooperação entre o condutor e sistema de
controle de veículos autônomos[Risk analisys for cooperation between the driver and