mod to gated-hebb

Author: ago109

ngclearn/components/synapses/hebbian/gatedHebbianSynapse.py

@@ -7,7 +7,7 @@ class GatedHebbianSynapse(DenseSynapse):
 
     # Define Functions
     def __init__(self, name, shape, eta=0., weight_init=None, bias_init=None,
-                 w_bound=1., w_decay=0., p_conn=1., resist_scale=1.,
+                 w_bound=1., w_decay=0., alpha=0., p_conn=1., resist_scale=1.,
                  batch_size=1, **kwargs):
         super().__init__(name, shape, weight_init, bias_init, resist_scale,
                          p_conn, batch_size=batch_size, **kwargs)
@@ -17,6 +17,7 @@ def __init__(self, name, shape, eta=0., weight_init=None, bias_init=None,
         self.w_bound = w_bound
         self.w_decay = w_decay ## synaptic decay
         self.eta = eta
+        self.alpha = alpha
 
         # compartments (state of the cell, parameters, will be updated through stateless calls)
         self.preVals = jnp.zeros((self.batch_size, shape[0]))
@@ -29,19 +30,19 @@ def __init__(self, name, shape, eta=0., weight_init=None, bias_init=None,
         self.dBiases = Compartment(jnp.zeros(shape[1]))
 
     @staticmethod
-    def _compute_update(w_bound, pre, post, weights):
+    def _compute_update(alpha, pre, post, weights):
         ## calculate synaptic update values
         dW = jnp.matmul(pre.T, post)
         db = jnp.sum(post, axis=0, keepdims=True)
-        # if w_bound > 0.:
-        #     dW = dW * (w_bound - jnp.abs(weights))
+        if alpha > 0.: ## apply synaptic dependency weighting
+            dW = dW * (alpha - jnp.abs(weights))
         return dW, db
 
     @staticmethod
-    def _evolve(bias_init, eta, w_decay, w_bound, pre, post, preSpike,
+    def _evolve(bias_init, eta, alpha, w_decay, w_bound, pre, post, preSpike,
                 postSpike, weights, biases):
         ## calculate synaptic update values
-        dWeights, dBiases = GatedHebbianSynapse._compute_update(w_bound, pre, post, weights)
+        dWeights, dBiases = GatedHebbianSynapse._compute_update(alpha, pre, post, weights)
         weights = weights + dWeights * eta
         if w_decay > 0.:
             Wdec = jnp.matmul((1. - preSpike).T, postSpike) * w_decay