stdp_nn_restr_synapse

stdp_nn_restr_synapse#

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

Restricted symmetric nearest-neighbour STDP synapse spec (NEST stdp_nn_restr_synapse).

Like stdp_nn_symm_synapse, both K+ and K- run in the substrate’s 'nearest' mode (reset-to-1 on spike); the difference is a one-pair-per-spike restriction carried by two per-edge eligibility flags:

  • pre_avail — a pre has occurred since the previous post (a post may facilitate),

  • post_avail — a post has occurred since the previous pre (a pre may depress).

On a post spike, potentiation fires only if pre_avail (then the pre is consumed); on a pre spike, depression fires only if post_avail (then the post is consumed). A spike sets its own side available and clears the opposite. The weight maps are the same as stdp_synapse:

\[ \begin{align}\begin{aligned}\hat w \leftarrow \hat w + \lambda (1-\hat w)^{\mu_+} K^+ \quad\text{(post spike, if } \mathtt{pre\_avail})\\\hat w \leftarrow \hat w - \alpha \lambda \hat w^{\mu_-} K^- \quad\text{(pre spike, if } \mathtt{post\_avail})\end{aligned}\end{align} \]

with \(\hat w = w/W_{\max}\), clamped to \([0, W_{\max}]\). The gathered trace excludes the current step’s own spike (K+ = pre_trace - pre_spike, K- = post_trace - post_spike), which recovers the second-latest partner on an exactly-coinciding step (stdp_nn_restr_synapse.h:62-66). The restriction makes restr diverge from symm whenever several spikes of one side fall between two of the other: symm pairs every one, restr only the first.

Parameters:

  • weight (ArrayLike or Quantity, optional) – Per-edge weight (pA; bare numbers are pA). Same sign as Wmax. Default 1.0 pA.

  • delay (Quantity, optional) – Homogeneous axonal delay (> 0). Default 1.0 ms.

  • receptor_type (int, optional) – Postsynaptic receptor port (>= 0). Default 0.

  • tau_plus (Quantity, optional) – Presynaptic K+ trace constant (> 0). Default 20.0 ms.

  • tau_minus (Quantity, optional) – Postsynaptic K- trace constant (> 0). Default 20.0 ms.

  • lambda (float, optional) – Learning rate \(\lambda\). Default 0.01.

  • alpha (float, optional) – Depression asymmetry \(\alpha\). Default 1.0.

  • mu_plus (float, optional) – Potentiation/depression exponents (0 additive, 1 multiplicative soft-bound). Default 1.0.

  • mu_minus (float, optional) – Potentiation/depression exponents (0 additive, 1 multiplicative soft-bound). Default 1.0.

  • Wmax (float, optional) – Weight bound (same sign as weight). Default 100.0.

Notes

NEST divergence — ``tau_minus`` location. In NEST tau_minus is a parameter of the postsynaptic neuron (ArchivingNode), not the synapse; here it is a synapse-spec attribute driving the substrate’s per-post K- trace.

Phantom-pre-at-0 (shared with symm). NEST’s first send (t_lastspike_=0) facilitates a post preceding the first pre against a virtual pre at t=0; the substrate’s pre_avail flag starts at 0, so that post is simply not eligible — the physically correct nearest behaviour. Parity is asserted where this is absent/below tolerance.

The eager substrate applies potentiation at post-spike steps and depression at pre-spike steps; NEST defers both to the next send. The cumulative op set is identical at every send, so the trajectories coincide where the weight_recorder samples (pre-spike steps).

Parity note. The exact nearest-neighbour pairing convention, the NEST source citation, and the single-pair regression test are documented in STDP parity: where state lives and how spikes pair (stdp_nn_restr_synapse — restricted symmetric nearest-neighbour).

References

See also

stdp_nn_symm_synapse, stdp_synapse, stdp_nn_pre_centered_synapse

Examples

>>> import brainunit as u
>>> from brainpy.state import stdp_nn_restr_synapse
>>> s = stdp_nn_restr_synapse(weight=1.0, tau_plus=20.0 * u.ms)
>>> s.pre_trace_mode, s.post_trace_mode
('nearest', 'nearest')
>>> s.edge_state_init()
{'pre_avail': 0.0, 'post_avail': 0.0}