Struct ReadoutNeuron 

pub struct ReadoutNeuron {
    pub membrane: i32,
    pub kappa: i16,
}

Available on crate feature alloc only.

Non-spiking leaky integrator readout neuron.

The readout accumulates weighted spike inputs with exponential decay. Its membrane potential (i32 for extra precision) is the network’s continuous output.

Precision

While hidden layer neurons use i16 membranes, the readout uses i32 to prevent overflow during long sequences. The decay factor kappa is still Q1.14 (i16), but the membrane has ~18 bits of headroom above the Q1.14 range.

Fields

membrane: i32

Membrane potential (higher precision accumulator). Stored in Q1.14-compatible scale but with i32 range.

kappa: i16

Decay factor in Q1.14 (controls how fast past inputs are forgotten).

Implementations

impl ReadoutNeuron

pub fn new(kappa: i16) -> Self

Create a new readout neuron with the given decay factor.

Arguments

• kappa – decay factor in Q1.14 (0 = no memory, Q14_ONE = perfect memory)

pub fn step(&mut self, weighted_input: i32)

Advance the readout by one timestep.

Decays the membrane potential and integrates new weighted input:

membrane = (membrane * kappa) >> 14 + weighted_input

Arguments

• weighted_input – sum of W_kj * z_j[t] for active spikes, as i32

pub fn output_i32(&self) -> i32

Get the raw membrane potential as i32.

pub fn output_f64(&self, scale: f64) -> f64

Dequantize the membrane potential to f64.

Arguments

• scale – output scaling factor (typically 1.0 / Q14_ONE as f64)

pub fn reset(&mut self)

Reset the membrane potential to zero.