Neural Networks with Nonlinear Couplings
Computing with Synchrony
| dc.contributor.advisor | Timme, Marc Prof. Dr. | |
| dc.contributor.author | Jahnke, Sven | |
| dc.date.accessioned | 2015-05-20T09:07:11Z | |
| dc.date.available | 2015-05-20T09:07:11Z | |
| dc.date.issued | 2015-05-20 | |
| dc.identifier.uri | http://hdl.handle.net/11858/00-1735-0000-0022-5FA1-C | |
| dc.identifier.uri | http://dx.doi.org/10.53846/goediss-5056 | |
| dc.language.iso | eng | de |
| dc.rights.uri | http://creativecommons.org/licenses/by-nc-nd/3.0/ | |
| dc.subject.ddc | 530 | de |
| dc.title | Neural Networks with Nonlinear Couplings | de |
| dc.title.alternative | Computing with Synchrony | de |
| dc.type | doctoralThesis | de |
| dc.contributor.referee | Timme, Marc Prof. Dr. | |
| dc.date.examination | 2014-05-22 | |
| dc.subject.gok | Physik (PPN621336750) | de |
| dc.description.abstracteng | Cortical neural networks generate a ground state of highly irregular spiking activity whose dynamics are sensitive to small perturbations such as missing or additional spikes. A robust, reliable transmission of information in the presence of such perturbations and noise is nonetheless assumed to be essential for neural computation. It has been hypothesized that this might be achieved by propagation of pulses of synchronous spikes (pulse packets) along feed-forward chains. In current models, functionally relevant chains require a dense connectivity between the neuronal layers of the network or strongly enhanced synapses and specifically modified response properties of neurons within the chain. Such highly distinguished large-scale structures however are not observed experimentally. Can less structured networks also guide synchrony? Recently, single neuron experiments have revealed a mechanism that nonlinearly promotes synchronous inputs: The input sites of the neurons, called dendrites, are traditionally considered as a tree of passive cables that conduct the electrical signal from the contact sites with the presynaptic neurons (synapses) to the postsynaptic neurons' cell body. Yet, this view has changed over the last decades: It has been demonstrated that in addition to just conducting signals, dendrites can actively contribute to computational processes by generation of fast dendritic spikes. These spikes are elicited upon sufficiently synchronous and strong dendritic stimulation and induce rapid, strong depolarizations in the cell body that are nonlinearly enhanced compared to depolarizations expected from linear summation of single inputs. Thus dendritic spikes contribute a synchrony detection mechanism to the computing capabilities of single neurons. In this thesis we study the impact of dendritic spikes that induce non-additive coupling, on collective circuit dynamics. As a starting point, we consider synchrony propagation in isolated, strongly diluted, feed-forward networks, i.e., the embedding network is modeled by externally generated random input spike trains. We compare the propagation properties of networks with and without dendritic nonlinearities, derive an analytical description for the propagating (synchronous) pulse-packet, and identify linear and nonlinear propagation as qualitatively different phenomena. We proceed by considering feed-forward networks which are natural part of a recurrent, sparsely connected, random network. We show that dendritic nonlinearities enable robust signal propagation in networks with biologically plausible topology, and synaptic efficiencies in the biologically observed range. Further, we consider the interaction between the embedding network and the embedded substructure in detail. We show that for purely random networks, synchronous activity in the feed-forward subnetworks may either have only a small effect on the activity of the remaining network, or cause pathological activity by inducing global network synchrony. In contrast, in networks with long-tailed degree distribution (that contain some highly connected nodes --- hubs), a propagating signal can induce moderate network oscillations (within the ``hub-network'') without causing pathological activity states, and these oscillations may in turn stabilize signal propagation. This phenomenon of hub-activated signal transmission further relaxes the requirement for a prominent feed-forward anatomy. Motivated by the abundance of cortical oscillations observed in experiments, we study the interaction of (external) oscillations and signal propagation. In particular, we show the existence of resonances between (external) oscillatory input and propagating synchronous signals. Such resonances are absent in linearly coupled networks. Thus the co-action of oscillations and dendritic nonlinearities, additionally to their support of signal transmission in general, can serve as mechanism to selectively activate different pathways in a recurrent network. Further, we identify the hippocampus as a core candidate region for oscillation-induced signal transmission (as described above), since in the hippocampus both high-frequency oscillations and replay of spike patterns are simultaneously observed in experiments: The spiking activity in this cortical region reflects real world features (e.g., position in space) during behavior. In consecutive phases of sleep during intermittent phases of highly increased global spiking activity (Sharp-Waves) these spike patterns are replayed in conjunction with high-frequency oscillations (Ripples). On the basis of the previous chapters, in the final part of the thesis, we develop a unified model for the storage and replay of spatial information in the hippocampus in conjunction with Sharp-Wave/Ripple-like spiking activity. In our model the replay is based on the nonlinear dendritic computation due to dendritic sodium spikes. Moreover, the proposed mechanism explains characteristic features like the overall wave form and the oscillation frequency of Sharp-Wave/Ripple complexes. | de |
| dc.contributor.coReferee | Wörgötter, Florentin Prof. Dr. | |
| dc.contributor.thirdReferee | Diesmann, Markus Prof. Dr. | |
| dc.subject.eng | neural networks | de |
| dc.subject.eng | dendritic spikes | de |
| dc.subject.eng | feed-forward networks | de |
| dc.subject.eng | synfire chains | de |
| dc.subject.eng | hippcampus | de |
| dc.subject.eng | synchrony | de |
| dc.subject.eng | Sharp-Wave/Rippels | de |
| dc.subject.eng | oscillations | de |
| dc.subject.eng | hubs | de |
| dc.subject.eng | recurrent networks | de |
| dc.identifier.urn | urn:nbn:de:gbv:7-11858/00-1735-0000-0022-5FA1-C-3 | |
| dc.affiliation.institute | Fakultät für Physik | de |
| dc.identifier.ppn | 825808774 |