Izhikevich Neuron Model

I am reading the example of using the Izhikevich neuron model with Nengo.

Close to the end the article states:

Unfortunately, Izhikevich neurons can’t necessarily be used in all of the situations that LIFs are used, due to the more complex dynamics illustrated above.
The way that Nengo encodes and decodes information with neurons is informed by the tuning curves of those neurons. With Izhikevich neurons, the firing rate with a certain input current J changes; the spike rate is initially higher due to the adaptation illustrated above.
We try our best to generate tuning curves for Izhikevich neurons.

My questions are:

  1. Do other neural simulators encode and decode information differently (without using the tuning curves)?
  2. Does using the Izhikevich model causes the same problems in other neural simulators or is its use easier?
  3. Is there an inherent difficulty in using this model in Nengo or its the same in every simulator because the difficulty arises due to its dynamics?

The notion of encoding/decoding is unique to the NEF, so it’s only if you’re using those features in Nengo that this comes up. So at worst this means that if you want to use Izhikevich neurons in Nengo you’re just connecting neurons with standard connection weights (which is what you would do in other neural simulators regardless).

As mentioned above, this quotation is describing how some of the extra, unique features of Nengo (NEF encoding/decoding) are limited in the case of Izhikevich neurons. Other neural simulators don’t have those features at all, so it’s not something they need to worry about.

Same as above, Nengo supports all the standard ways of connecting Izhikevich neurons that other neural simulators would support, it’s just the extra NEF connection style that doesn’t work as well with Izhikevich neurons.


Thanks for the detailed answer!