tfa.activations.lisht

LiSHT: Non-Parameteric Linearly Scaled Hyperbolic Tangent Activation Function.

Computes linearly scaled hyperbolic tangent (LiSHT):

\[ \mathrm{lisht}(x) = x * \tanh(x). \]

See LiSHT: Non-Parameteric Linearly Scaled Hyperbolic Tangent Activation Function for Neural Networks.

Usage:

x = tf.constant([1.0, 0.0, 1.0])
tfa.activations.lisht(x)
<tf.Tensor: shape=(3,), dtype=float32, numpy=array([0.7615942, 0.       , 0.7615942], dtype=float32)>

x A Tensor. Must be one of the following types: bfloat16, float16, float32, float64.

A Tensor. Has the same type as x.