Hyperbolic Activation Function Calculator

Calculator Inputs

Function type

Input value x

Scale α

Slope β

Inverse target y

Range start

Range end

Step size

Decimal places

Example Data Table

Case	Function	x	α	β	Expected behavior
Centered learning	Standard tanh	0	1	1	Output is zero and derivative is highest.
Positive saturation	Scaled hyperbolic tangent	4	1.7159	0.6667	Output moves close to the upper limit.
Soft transition	Bipolar sigmoid through tanh	1.5	1	0.5	Output changes slowly with a lower slope.

Formula Used

Basic hyperbolic tangent: tanh(x) = (e^x - e^-x) / (e^x + e^-x)

Scaled activation: f(x) = α × tanh(βx)

First derivative: f'(x) = αβ × [1 - tanh²(βx)]

Second derivative: f''(x) = -2αβ² × tanh(βx) × [1 - tanh²(βx)]

Inverse for scaled form: x = atanh(y / α) / β

For the standard form, α and β are treated as one. For the LeCun form, α is 1.7159 and β is two divided by three.

How to Use This Calculator

Select the hyperbolic activation function type.
Enter the input value x for the main result.
Set α for output scale and β for curve steepness.
Enter a target y value to calculate the inverse x.
Choose range start, range end, and step size.
Press Calculate to show results above the form.
Use CSV for spreadsheet data and PDF for a report.

What Is a Hyperbolic Activation Function?

A hyperbolic activation function maps any real input into a bounded, centered output. The common form is tanh. It returns values between minus one and one. This makes it useful for signals that need positive and negative meaning. Neural networks use it when centered gradients are helpful.

Why Tanh Matters

The tanh curve is S shaped. It is almost linear near zero. It becomes flat for large positive or negative inputs. That flat area is called saturation. In saturation, the derivative becomes small. Small derivatives can slow training. The calculator shows this behavior with output, slope, curvature, and gradient health.

Scaling and Slope Control

Advanced models may use a scaled form. The scale value changes the output range. The slope value changes how quickly the curve bends. A high slope creates a sharper transition. A low slope gives a softer response. The LeCun form is a popular scaled tanh version. It can improve signal flow in some older neural network designs.

Derivative Insight

The first derivative measures sensitivity. It tells how much the activation output changes when input changes. Near zero, tanh has its largest derivative. Far from zero, the derivative approaches zero. The second derivative shows curve direction. It helps explain bending and inflection around the origin.

Range Analysis

A single input result is useful. A range table is better for study. This calculator builds many sample points from your chosen interval. It reports minimum, maximum, mean, and saturation percentage. The graph compares activation output and derivative. This helps you see where the function is responsive.

Practical Use

Use this tool when checking neural math, classroom examples, or model notes. Enter an input value first. Then choose the function type. Adjust scale, slope, range start, range end, and step size. Press calculate to view results above the form. Export the table for reports. Download the PDF summary for quick sharing.

Good Interpretation

A strong activation is not always better. Values near the limits may look stable, but gradients can vanish. Values near zero keep learning more active. Use the derivative and graph together. They show whether your chosen input range supports useful updates.

FAQs

What is a hyperbolic activation function?

It is a smooth function based on hyperbolic math. The most common example is tanh. It maps real inputs into a bounded range, usually from minus one to one.

Why is tanh used in neural networks?

Tanh gives centered outputs. Negative inputs can produce negative outputs. This can help some models balance signals better than functions that only return positive values.

What does the derivative show?

The derivative shows sensitivity. A larger derivative means the output changes faster. A very small derivative means the neuron may be in saturation.

What does α mean?

Alpha controls the output scale. A larger alpha expands the upper and lower limits. In scaled tanh, the output range becomes approximately minus alpha to alpha.

What does β mean?

Beta controls steepness. A larger beta makes the curve transition faster around zero. A smaller beta makes the activation softer and less steep.

What is saturation?

Saturation happens when output gets close to its limit. In that region, the derivative becomes small. Learning can slow because gradients carry less change.

Can this calculator find inverse values?

Yes. Enter a target output y. The tool checks the valid output range and calculates the matching x value when the inverse exists.

Why use the range graph?

The graph shows output and derivative across many x values. It helps identify linear regions, flat regions, and inputs with healthier gradient behavior.