urbanczik_synapse

urbanczik_synapse#

class brainpy.state.urbanczik_synapse(*args, **kwargs)#

Dendritic prediction-error plasticity synapse spec (NEST urbanczik_synapse).

Implements the Urbanczik-Senn rule: with \(\bar{s}_L\), \(\bar{s}_s\) the presynaptic traces at the dendritic membrane (tau_L) and synaptic (tau_s) time constants, and \(\delta\Pi\) the post neuron’s dendritic prediction error read per edge each step,

\[\begin{split}\mathrm{PI}(t) &= (\bar{s}_L(t) - \bar{s}_s(t))\,\delta\Pi(t) \\ \mathrm{PI_{int}}(t) &= \textstyle\sum_{s \le t} \mathrm{PI}(s) \\ \mathrm{PI_{exp}}(t) &= \textstyle\sum_{s \le t} e^{-(t-s)/\tau_\Delta}\,\mathrm{PI}(s) \\ w(t) &= \mathrm{clip}\big(w_0 + (\mathrm{PI_{int}} - \mathrm{PI_{exp}})\, P,\; W_{\min},\, W_{\max}\big)\end{split}\]

with the NEST prefactor \(P = 15\,C_m\,\tau_s\,\eta / (g_L (\tau_L - \tau_s))\) (dendritic C_m [pF], g_L [nS], tau_L = C_m/g_L [ms], tau_s [ms]) and \(w_0\) the spec’s initial weight. The two sums are carried per edge (edge_state_init); the presynaptic traces are the substrate’s two-column per-pre trace seam (pre_trace_tau = (tau_L, tau_s)).

Parameters:

  • weight (ArrayLike or Quantity, optional) – Initial per-edge weight; bare numbers default to pA (the dendritic current port). Its sign selects tau_s and must agree with Wmin/Wmax under NEST’s sign tests. Default 1.0 pA.

  • delay (Quantity, optional) – Axonal/dendritic delay (> 0). Default 1.0 ms. For live-NEST parity the drive sets delay = dt (one grid step) to align the online reader’s one-step lag to NEST’s minimal dendritic-delay window.

  • eta (float, optional) – Learning rate \(\eta\). Default 0.07.

  • tau_Delta (Quantity, optional) – Plasticity low-pass time constant \(\tau_\Delta\) (> 0). Default 100.0 ms.

  • Wmin (float, optional) – Lower / upper weight clamp. Defaults 0.0 / 100.0.

  • Wmax (float, optional) – Lower / upper weight clamp. Defaults 0.0 / 100.0.

  • receptor_type (int, optional) – Informational default target port (3 = dendritic excitatory). The actual routing is driven by connect(..., receptor_type=...); the post neuron resolves it to a delta-input channel label. Default 3.

  • dend_C_m (Quantity, optional) – Dendritic compartment capacitance / leak conductance for the prefactor and tau_L = C_m/g_L. Defaults 300.0 pF / 30.0 nS (pp_cond_exp_mc_urbanczik dendrite).

  • dend_g_L (Quantity, optional) – Dendritic compartment capacitance / leak conductance for the prefactor and tau_L = C_m/g_L. Defaults 300.0 pF / 30.0 nS (pp_cond_exp_mc_urbanczik dendrite).

  • dend_tau_syn_ex (Quantity, optional) – Dendritic excitatory / inhibitory synaptic time constants; tau_s is the one selected by the initial weight sign. Defaults 3.0 ms / 3.0 ms.

  • dend_tau_syn_in (Quantity, optional) – Dendritic excitatory / inhibitory synaptic time constants; tau_s is the one selected by the initial weight sign. Defaults 3.0 ms / 3.0 ms.

Notes

See the module docstring for the online-vs-event-driven equivalence, the dendritic-parameter-location divergence, and the parity posture.

Examples

>>> import brainunit as u
>>> from brainpy.state import urbanczik_synapse
>>> s = urbanczik_synapse(weight=1.0 * u.pA)
>>> s.is_homogeneous_weight, s.post_state_reads
(False, ('delta_Pi',))
>>> s.post_trace_tau, s.edge_state_init()
(None, {'PI_integral': 0.0, 'PI_exp_integral': 0.0})
>>> [float(u.Quantity(t).to_decimal(u.ms)) for t in s.pre_trace_tau]
[10.0, 3.0]