It shows the effect of learning random noise when training a neural network model.
- Related Paper(my paper) : “Fault Diagnosis of Inverter Current Sensor Using Artificial Neural Network Considering Out-of-distribution”, 2021 IEEE Energy Conversion Congress and Exposition-Asia(ECCE-Asia), Singapore, May 24-27, 2021 (Accepted, 1st author, To be published)
This idea came up with below refereces which deal with using a random noise when training a machine learning model. Random noise dataset could be generally applied as the additional input data, weights, gradients, or outputs.
Ref1 - A. Neelakantan et al., "Adding Gradient Noise Improves Learning for Very Deep Networks," arXiv:1511.06807 [cs, stat], Nov. 2015, Accessed: Mar. 31, 2021. [Online]. Available: http://arxiv.org/abs/1511.06807.
Ref2 - C. M. Bishop, "Training with Noise is Equivalent to Tikhonov Regularization," Neural Computation, vol. 7, no. 1, pp. 108–116, Jan. 1995, doi: 10.1162/neco.1995.7.1.108.
Ref3 - J. Sietsma and R. J. F. Dow, "Creating artificial neural networks that generalize," Neural Networks, vol. 4, no. 1, pp. 67–79, Jan. 1991, doi: 10.1016/0893-6080(91)90033-2.
Ref4 - D. Hendrycks, M. Mazeika, and T. Dietterich, "Deep Anomaly Detection with Outlier Exposure," arXiv:1812.04606 [cs, stat], Jan. 2019, Accessed: Mar. 31, 2021. [Online]. Available: http://arxiv.org/abs/1812.04606.