SIREN Nested-Sine Coordinate Derivatives
Differentiate a nested sine network and see how frequency affects the signal and its derivatives.
Free theory articleDifferentiate a nested sine network and see how frequency affects the signal and its derivatives.
Concept & interpretation
Sinusoidal Representation Networks (SIRENs) are implicit neural networks that use periodic activation functions instead of traditional ones like ReLU. This architecture allows them to model complex, high-frequency signals and their derivatives accurately, which is essential for tasks such as solving partial differential equations or representing implicit surfaces.
Experiment
Interact with the interactive model to observe how changing the first frequency and other parameters influences the network and its derivatives. By observing the visual output, you can grasp the concepts intuitively.
- Set the First frequency to 2 and note the baseline shape of the function and its derivative.
- Increase the First frequency to 4 and observe how the derivative's magnitude scales with the frequency.
- Change the Difference step from 0.1 to 0.01 and see how the central-difference approximation becomes more accurate, illustrating the risk of using coarse finite differences for high-frequency signals.