Use this function when exploring or implementing a neural-network layer that needs a smooth, bounded, zero-centered nonlinear response.

X

Input value.

Apply the hyperbolic tangent activation. The tool evaluates the supplied inputs together and returns the named outputs below; it does not infer omitted operating conditions or change the units shown.^[1]

Y

tanh(x).

• Hyperbolic tangent saturates near −1 and 1, where gradients become small. Network behavior also depends on initialization, normalization, architecture, data, and optimization rather than the activation alone.
• Use finite inputs in the displayed units and preserve more precision than the final presentation requires. Independently verify safety-critical, financial, compliance, or production decisions.

Compare tanh with sigmoid, ReLU, leaky ReLU, or another activation under the same model and training conditions. Inspect both forward values and gradients when diagnosing saturation.

• Leaky ReLU

  Apply the leaky ReLU activation, scaling negatives by a small slope.

• ReLU

  Apply the rectified linear activation, flooring negatives at zero.

• Sigmoid

  Apply the logistic sigmoid activation.