Source code for dendritic_modeling.networks.architectures.excitation_inhibition.dendritic.level_interventions

"""Labelled interventions on single ``DendriNet`` levels.

Interventions are passed to :meth:`DendriNet.forward_levels`; ``forward``
never applies them. They replace a level's output in the pass where they are
given and change no parameter or buffer.
"""

from __future__ import annotations

from dataclasses import dataclass
from typing import Any

import torch


[docs] @dataclass(frozen=True, eq=False) class CurrentInjection: """Replace a shunting level's inhibitory conductance by injected current. The inhibition beyond a reference input acts as a current at a fixed voltage ``anchor`` instead of as a conductance: ``V = (E + C - (I - I_ref) * anchor) / (leak + E + G + I_ref)``, followed by the level's activation, where ``I`` and ``I_ref`` are the level's inhibitory conductances for the actual and the reference inhibitory input, ``C`` and ``G`` the child current and coupling conductance and ``E`` the excitatory conductance (absent at levels without that pathway). The level's ordinary forward still runs first, so its diagnostics describe the unperturbed conductance state. ``anchor`` has one value per compartment (for example a training-set mean voltage); it is the caller's to calibrate. """ reference_inhibitory_input: torch.Tensor anchor: torch.Tensor def output(self, layer: Any, x, inhibitory_input, branch_input): """The level's output under the intervention.""" if not layer.use_shunting: raise ValueError("CurrentInjection needs a shunting level") if layer.branch_inhibition is None or inhibitory_input is None: raise ValueError("CurrentInjection needs a level with inhibitory input") # Terms are summed in the order of the level's own forward pass. numerator = 0 base = layer.effective_leak if layer.input_excitatory and layer.branch_excitation is not None: excitation = layer.branch_excitation(x) numerator = numerator + excitation base = base + excitation if layer.input_branches and branch_input is not None: numerator = numerator + layer.branches_to_output(branch_input) base = base + layer.branches_to_output.sum_conductances() inhibition = layer.branch_inhibition(inhibitory_input) reference = layer.branch_inhibition(self.reference_inhibitory_input) numerator = numerator - (inhibition - reference) * self.anchor denominator = base + reference if layer.epsilon: denominator = denominator + layer.epsilon return layer.reactivation(numerator / denominator)
__all__ = ["CurrentInjection"]