Communications of the ACM
A Formal Approach to Identifying the Impact of Noise on Neural Networks
Credit: Andrij Borys Associates, Auxin
The past few years have seen an incredible rise in the use of smart systems based on artificial neural networks (ANNs), owing to their remarkable classification capability and decision making comparable to that of humans. Yet, as shown in Figure 1, the addition of even a small amount of noise to the input may trigger these networks to give incorrect results.13 This is an alarming limitation of the ANNs, particularly for those deployed in safety-critical applications such as autonomous vehicles, aviation, and healthcare. For instance, consider a self-driving car using an ANN to perceive traffic signs as shown in Figure 2; the correct classification by the ANN in noisy real-world environments is crucial for the safety of humans and objects in the vicinity of the car.
Figure 1. Magnitudes of image input and the noise applied to it. The addition of noise causes the input previously classified correctly to be misclassified by the trained network.
Figure 2. The addition of small, imperceptible noise to traffic signs makes the trained network provide incorrect output classification.5
The standard design cycle of an ANN involves training the network on clean data, tuning the network's hyperparameters based on clean data, and ensuring acceptable accuracy of the trained network using clean data. However, noise is a ubiquitous component of the real world.
No entries found