Skip to main content
Journal of Big Data
Fig. 1 | Journal of Big Data

Fig. 1

From: Novel sensitivity method for evaluating the first derivative of the feed-forward neural network outputs

Fig. 1

Illustration of the subtractive cancellation errors in finite difference methods and the CSDA. Both FDA and CFDA suffer from subtractive cancellation errors unlike CSDA. The truncation errors in CSDA can be minimized by choosing a very low \(h\) value. (CSDA: Complex-Step Derivative Approximation; FDA: Finite Difference, and CFDA: Central Finite Difference Approximation)

Back to article page