fixed scaling issue in exp/alpha syn

Author: Alexander Ororbia

docs/tutorials/neurocog/dynamic_synapses.md

@@ -4,8 +4,7 @@ In this lesson, we will study dynamic synapses, or synaptic cable components in
 ngc-learn that evolve on fast time-scales in response to their pre-synaptic inputs. 
 These types of chemical synapse components are useful for modeling time-varying 
 conductance which ultimately drives eletrical current input into neuronal units 
-(such as spiking cells).  
-Here, we will learn how to build two important types of dynamic synapses in 
+(such as spiking cells).  Here, we will learn how to build two important types of dynamic synapses in 
 ngc-learn -- the exponential synapse and the alpha synapse -- and visualize 
 the time-course of their resulting conductances. In addition, we will then 
 construct and study a small neuronal circuit involving a leaky integrator that 
@@ -366,3 +365,8 @@ voltage threshold.
    +--------------------------------------------------------------------+
 ```
 
+Notice that the above shows the behavior of the post-synaptic LIF in response to the integration of pulses coming from two Poisson spike trains both at rates of $10$ Hz (since both `exc_freq` and `inh_freq` have been set to ten). Messing with the frequencies of the excitatory and inhibitory pulse trains can lead to sparser or denser post-synaptic spike outputs. 
+
+## References
+
+<b>[1]</b> Sterratt, David, et al. Principles of computational modelling in neuroscience. Cambridge university press, 2023.

ngclearn/components/synapses/alphaSynapse.py

@@ -90,18 +90,18 @@ def advance_state(
             dt, tau_syn, g_syn_bar, syn_rest, Rscale, inputs, weights, i_syn, g_syn, h_syn, v
     ):
         s = inputs
-        ## advance conductance variable
+        ## advance conductance variable(s)
         _out = jnp.matmul(s, weights) ## sum all pre-syn spikes at t going into post-neuron)
         dhsyn_dt = -h_syn/tau_syn + _out * g_syn_bar
         h_syn = h_syn + dhsyn_dt * dt ## run Euler step to move intermediate conductance h
 
         dgsyn_dt = -g_syn/tau_syn + h_syn # or -g_syn/tau_syn + h_syn/tau_syn
         g_syn = g_syn + dgsyn_dt * dt ## run Euler step to move conductance g
-        g_syn = g_syn * Rscale
 
-        i_syn = -g_syn
+        ## compute derive electrical current variable
+        i_syn = -g_syn * Rscale
         if syn_rest is not None:
-            i_syn =  -g_syn * (v - syn_rest)
+            i_syn =  -(g_syn  * Rscale) * (v - syn_rest)
         outputs = i_syn #jnp.matmul(inputs, Wdyn * Rscale) + biases
         return outputs, i_syn, g_syn, h_syn
 

ngclearn/components/synapses/exponentialSynapse.py

@@ -90,12 +90,12 @@ def advance_state(
         s = inputs
         ## advance conductance variable
         _out = jnp.matmul(s, weights) ## sum all pre-syn spikes at t going into post-neuron)
-        dgsyn_dt = -g_syn/tau_syn + _out * g_syn_bar
+        dgsyn_dt = -g_syn/tau_syn + (_out * g_syn_bar) * (1./dt)
         g_syn = g_syn + dgsyn_dt * dt ## run Euler step to move conductance
-        g_syn = g_syn * Rscale
-        i_syn = -g_syn
+        ## compute derive electrical current variable
+        i_syn = -g_syn * Rscale
         if syn_rest is not None:
-            i_syn =  -g_syn * (v - syn_rest)
+            i_syn =  -(g_syn * Rscale) * (v - syn_rest)
         outputs = i_syn #jnp.matmul(inputs, Wdyn * Rscale) + biases
         return outputs, i_syn, g_syn