This title appears in the Scientific Report :
2018
Please use the identifier:
http://hdl.handle.net/2128/19865 in citations.
Encoding symbolic sequences with spiking neural reservoirs
Encoding symbolic sequences with spiking neural reservoirs
Biologically inspired spiking networks are an im-portant tool to study the nature of computation and cognition inneural systems. In this work, we investigate the representationalcapacity of spiking networks engaged in an identity mappingtask. We compare two schemes for encoding symbolic input, onein...
| Personal Name(s): | Duarte, Renato (Corresponding author) |
|---|---|
| Uhlmann, Marvin / van den Broek, Dick / Fitz, Hartmut / Petersson, Karl Magnus / Morrison, Abigail | |
| Contributing Institute: |
Computational and Systems Neuroscience; INM-6 Jara-Institut Brain structure-function relationships; INM-10 Computational and Systems Neuroscience; IAS-6 |
| Imprint: |
2018
|
| Physical Description: |
1-8 |
| Conference: | International Joint Conference on Neural Networks (IJCNN 2018), Rio de Janeiro (Brazil), 2018-07-08 - 2018-07-14 |
| Document Type: |
Contribution to a conference proceedings |
| Research Program: |
W2/W3 Professorinnen Programm der Helmholtzgemeinschaft Theory, modelling and simulation |
| Link: |
OpenAccess OpenAccess |
| Publikationsportal JuSER |
|
Biologically inspired spiking networks are an im-portant tool to study the nature of computation and cognition inneural systems. In this work, we investigate the representationalcapacity of spiking networks engaged in an identity mappingtask. We compare two schemes for encoding symbolic input, onein which input is injected as a direct current and one whereinput is delivered as a spatio-temporal spike pattern. We testthe ability of networks to discriminate their input as a functionof the number of distinct input symbols. We also compareperformance using either membrane potentials or filtered spiketrains as state variable. Furthermore, we investigate how thecircuit behavior depends on the balance between excitationand inhibition, and the degree of synchrony and regularityin its internal dynamics. Finally, we compare different linearmethods of decoding population activity onto desired targetlabels. Overall, our results suggest that even this simple mappingtask is strongly influenced by design choices on input encoding,state-variables, circuit characteristics and decoding methods, andthese factors can interact in complex ways. This work highlightsthe importance of constraining computational network models ofbehavior by available neurobiological evidence. |