Struct LeCun 

pub struct LeCun;

LeCun’s tanh activation function.

This activation function is defined as: f(x) = 1.7159 * tanh(0.666 * x)

It provides a smoother alternative to standard tanh with better gradient flow properties. The scaling factors (1.7159 and 0.666) are chosen such that:

• The function approximates the identity near the origin
• The output range is approximately [-1.7159, 1.7159]

Example

use burn::backend::NdArray;
use burn::tensor::Tensor;
use ncps::activation::LeCun;

type Backend = NdArray<f32>;
let device = Default::default();

let x = Tensor::<Backend, 1>::from_floats([0.0, 1.0, -1.0], &device);
let y = LeCun::forward(x);

Implementations

impl LeCun

pub fn forward<B: Backend, const D: usize>(x: Tensor<B, D>) -> Tensor<B, D>

Applies the LeCun tanh activation function.

Arguments

• x - Input tensor of any dimension

Returns

Tensor with LeCun activation applied element-wise