improve url formatting

README.md

@@ -7,7 +7,7 @@
 
 This is the Go implementation of the Axon algorithm for spiking, biologically based models of cognition, based on the [emergent](https://github.com/emer/emergent) framework.  Development of Axon is supported by the Obelisk project at https://astera.org/ and by collaborations with scientists at the University of California Davis, and other institutions around the world.
 
-Axon is the spiking version of [Leabra](https://github.com/emer/leabra), with several advances.  As a backcronym, *axon* could stand for *Adaptive eXcitation Of Noise*, reflecting the ability to learn using the power of error-backpropagation in the context of noisy spiking activation.  The spiking function of the axon is what was previously missing from Leabra.  Axon is used to develop large-scale systems-neuroscience models of the brain, i.e., [Computational Cognitive Neuroscience](CompCogNeuro.org), centered around the [Rubicon](Rubicon.md) model of goal-driven, motivated cognition.
+Axon is the spiking version of [Leabra](https://github.com/emer/leabra), with several advances.  As a backcronym, *axon* could stand for *Adaptive eXcitation Of Noise*, reflecting the ability to learn using the power of error-backpropagation in the context of noisy spiking activation.  The spiking function of the axon is what was previously missing from Leabra.  Axon is used to develop large-scale systems-neuroscience models of the brain, i.e., [Computational Cognitive Neuroscience](compcogneuro.org), centered around the [Rubicon](Rubicon.md) model of goal-driven, motivated cognition.
 
 Axon and emergent use the [Cogent Core](https://cogentcore.org/core) GUI framework.  See [install](https://www.cogentcore.org/core/setup/install) instructions there.  Once those prerequisites are in place, then the simplest way to run a simulation is:
 
@@ -149,7 +149,7 @@ func (ss *Sim) NetViewCounters(tm etime.Times) {
 
 # Overview of the Axon Algorithm
 
-Axon is the spiking version of [Leabra](https://github.com/emer/leabra), which uses rate-code neurons instead of spiking.  Like Leabra, Axon is intended to capture a middle ground between neuroscience, computation, and cognition, providing a computationally effective framework based directly on the biology, to understand how cognitive function emerges from the brain.  See [Computational Cognitive Neuroscience](https://CompCogNeuro.org) for a full textbook on the principles and many implemented models.
+Axon is the spiking version of [Leabra](https://github.com/emer/leabra), which uses rate-code neurons instead of spiking.  Like Leabra, Axon is intended to capture a middle ground between neuroscience, computation, and cognition, providing a computationally effective framework based directly on the biology, to understand how cognitive function emerges from the brain.  See [Computational Cognitive Neuroscience](https://compcogneuro.org) for a full textbook on the principles and many implemented models.
 
 ## Pseudocode as a LaTeX doc for Paper Appendix
 
@@ -196,7 +196,7 @@ Furthermore, synaptic inputs are integrated first by separate pathways, and then
 
 ## Inhibitory Competition Function Simulating Effects of Interneurons
 
-The pyramidal cells of the neocortex that are the main target of axon models only send excitatory glutamatergic signals via positive-only discrete spiking communication, and are bidirectionally connected.  With all this excitation, it is essential to have pooled inhibition to balance things out and prevent runaway excitatory feedback loops.  Inhibitory competition provides many computational benefits for reducing the dimensionality of the neural representations (i.e., *sparse* distributed representations) and restricting learning to only a small subset of neurons, as discussed extensively in the [Comp Cog Neuro textbook](https://CompCogNeuro.org).  It is likely that the combination of positive-only weights and spiking activations, along with inhibitory competition, is *essential* for enabling axon to learn in large, deep networks, where more abstract, unconstrained algorithms like the Boltzmann machine fail to scale (paper TBD).
+The pyramidal cells of the neocortex that are the main target of axon models only send excitatory glutamatergic signals via positive-only discrete spiking communication, and are bidirectionally connected.  With all this excitation, it is essential to have pooled inhibition to balance things out and prevent runaway excitatory feedback loops.  Inhibitory competition provides many computational benefits for reducing the dimensionality of the neural representations (i.e., *sparse* distributed representations) and restricting learning to only a small subset of neurons, as discussed extensively in the [Comp Cog Neuro textbook](https://compcogneuro.org).  It is likely that the combination of positive-only weights and spiking activations, along with inhibitory competition, is *essential* for enabling axon to learn in large, deep networks, where more abstract, unconstrained algorithms like the Boltzmann machine fail to scale (paper TBD).
 
 Inhibition is provided in the neocortex primarily by the fast-spiking parvalbumin positive (PV+) and slower-acting somatostatin positive (SST+) inhibitory interneurons in the cortex ([Cardin, 2018](#references)).  Instead of explicitly simulating these neurons, a key simplification in Leabra that eliminated many difficult-to-tune parameters and made the models much more robust overall was the use of a summary inhibitory function.  This function directly computes a pooled inhibitory conductance `Gi` as a function of the feedforward (FF) excitation coming into a Pool of neurons, along with feedback (FB) from the activity level within the pool.  Fortuitously, this same [FFFB](fffb) function works well with spiking as well as rate code activations, but it has some biologically implausible properties, and also at a computational level requires multiple layer-level `for` loops that interfere with full parallelization of the code.