Hello,
Documentation for `tsodyks_synapse` says it is only compatible with `iaf_psc_exp` or `iaf_psc_exp_htum` neuron models. Would it be possible to use it with other `_exp`-type models with postsynaptic currents or conductances with exponential decay (for example, `aeif_cond_exp`)? If not, what could be a work around?
With best regards, Alexander Kozlov, CST EECS KTH.
Hej Alex!
Me and Nikolaos Chrysanthidis use and aeif_cond_exp with Tsodyks-Markram STP all the time with NEST on the supercomputer. Check any of our recent publications for that. Its a custom implementation originally done by Phil Tully, a former PhD student of Anders Lansner. The catch is that it also comes with the BCPNN learning rule for the Hebbian learning component, but ofcourse you could switch that off by setting the BCPNN plasticity time constants( or just the plasticity modulator switch kappa) to zero if you want static weights modulated by TM-based STP only. The TM-mechanisms parameters tau_fac, tau_rec, and U are independent of that, as the TM rule is multiplicative with the underlying weight, or rather the conductance, as its a conductance based model ofcourse. Hope this helps.
Kind regards \Florian
photo *Florian Fiebig, PhD* Researcher Computational Brain Science +46 70-744-7439 tel:+16505426046 | Skype: florianfiebig <#> http://www.numenta.com/ http://us.linkedin.com/in/florian-fiebig-b387b886 Most recent papers:https://rdcu.be/bRLmu https://doi.org/10.1523/JNEUROSCI.1989-16.2016 https://rdcu.be/bRLmu https://www.eneuro.org/content/early/2020/02/28/ENEURO.0374-19.2020
On 2/15/2022 11:49 AM, Alexander Kozlov wrote:
Hello,
Documentation for `tsodyks_synapse` says it is only compatible with `iaf_psc_exp` or `iaf_psc_exp_htum` neuron models. Would it be possible to use it with other `_exp`-type models with postsynaptic currents or conductances with exponential decay (for example, `aeif_cond_exp`)? If not, what could be a work around?
With best regards, Alexander Kozlov, CST EECS KTH. _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
Hello Alex and Florian,
The documentation of the Tsodyks synapse explains (at least attempts to explain) some of the logic behind the interaction between the STP dynamics and the synaptic current dynamics especially with an eye on using the synapse with other neuron models. If the documentation should be unclear, please get in touch.
After a brief look at the original paper and the documentation, my impression is the following
* If you use any model with psc_exp synaptic dynamics and ensure that tau_psc in the synapse matches the corresponding tau_syn_? in the neuron model, synaptic dynamics will be as in the Tsodyks et al paper. Unfortunately, we never found an efficient and generalizable way to automatically check that the time constants in the synapse model and the neuron agree, the user has to take care of this. * If you use the tsodyks synapse with different neuron models, you need to work out exactly which set of equations applies to your combined synaptic potentiation and synaptic current model.
Looking at it a little more, the situation is as follows:
* In the Tsodyks et al paper, variable y(t) describes the post-synaptic input current (equation 2) and drives variable z(t) which gives the fraction of inactive states (bottom of eq 3). * In NEST, the post-synaptic input current is computed in the neuron model, based on the synaptic time course for that neuron model. The value y(t) is computed independently in the tsodyks_synapse for technical reasons. * To match the Tsodyks model, the current computed in NEST and the value y(t) computed in the synapse must be identical, which is the case only if the neuron model has exponential post-synaptic current dynamics with the same time constant as in the tsodyks_synapse. * If you combine the tsodyks_synapse with a neuron model with different synaptic current dynamics, the STD variables x(t), y(t) and z(t) will describe some dynamics of short-term depression not directly coupled to the synaptic time course.
Best, Hans Ekkehard
--
Prof. Dr. Hans Ekkehard Plesser Head, Department of Data Science
Faculty of Science and Technology Norwegian University of Life Sciences PO Box 5003, 1432 Aas, Norway
Phone +47 6723 1560 Email hans.ekkehard.plesser@nmbu.nomailto:hans.ekkehard.plesser@nmbu.no Home http://arken.nmbu.no/~plesser
On 16/02/2022, 13:22, "Florian Fiebig" <fiebig@kth.semailto:fiebig@kth.se> wrote:
Hej Alex!
Me and Nikolaos Chrysanthidis use and aeif_cond_exp with Tsodyks-Markram STP all the time with NEST on the supercomputer. Check any of our recent publications for that. Its a custom implementation originally done by Phil Tully, a former PhD student of Anders Lansner. The catch is that it also comes with the BCPNN learning rule for the Hebbian learning component, but ofcourse you could switch that off by setting the BCPNN plasticity time constants( or just the plasticity modulator switch kappa) to zero if you want static weights modulated by TM-based STP only. The TM-mechanisms parameters tau_fac, tau_rec, and U are independent of that, as the TM rule is multiplicative with the underlying weight, or rather the conductance, as its a conductance based model ofcourse. Hope this helps.
Kind regards \Florian
[photo] Florian Fiebig, PhD Researcher Computational Brain Science +46 70-744-7439tel:+16505426046 | Skype: florianfiebig [cid:part4.03040201.05000302@kth.se]http://www.numenta.com/ [cid:part6.04040203.08020201@kth.se]http://us.linkedin.com/in/florian-fiebig-b387b886 Most recent papers: https://rdcu.be/bRLmu [cid:part10.03040309.00050607@kth.se]https://doi.org/10.1523/JNEUROSCI.1989-16.2016 [cid:part12.02030108.00090204@kth.se]https://rdcu.be/bRLmu [cid:part14.05030304.08040608@kth.se]https://www.eneuro.org/content/early/2020/02/28/ENEURO.0374-19.2020 On 2/15/2022 11:49 AM, Alexander Kozlov wrote:
Hello,
Documentation for `tsodyks_synapse` says it is only compatible with `iaf_psc_exp` or `iaf_psc_exp_htum` neuron models. Would it be possible to use it with other `_exp`-type models with postsynaptic currents or conductances with exponential decay (for example, `aeif_cond_exp`)? If not, what could be a work around?
With best regards,
Alexander Kozlov,
CST EECS KTH.
_______________________________________________
NEST Users mailing list -- users@nest-simulator.orgmailto:users@nest-simulator.org
To unsubscribe send an email to users-leave@nest-simulator.orgmailto:users-leave@nest-simulator.org
Dear Hans,
Thank you for the explanation! If tsodyks_synapse transmits the effective synaptic weight u(t)*x(t)*w to the postsynaptic neuron where it is converted to postsynaptic current or conductance, depending on the type of the model, then the present *_exp models should work with it just fine, as I understand. And taking into account approximate nature of synaptic models, the tsodyks_synapse, for practical purposes, can also be combined with *_alpha or *_beta type neuron models, I think. Although in this case synaptic parameters will have to be adjusted to fit the original behavior.
Thank you for your explanation and sorry for my confusion, I am not a native NEST user.
Best regards, Alex.
Hi Florian,
Nice to hear the answer is near! Thank you for the link, I'll check out your code on github. As far as I can see from Hans' answer and my test run, current adex neuron model in NEST3 runs with `tsodyks_synapse` just fine. This solves part of my issues and others could hopefully be worked around using NESTML which seems to be a recommended way for writing extensions.
Best regards, Alex.
-
Alexander Kozlov -
Florian Fiebig -
Hans Ekkehard Plesser