This title appears in the Scientific Report :
2014
Temporal sequence learning via adaptation in biologically plausible spiking neural networks
Temporal sequence learning via adaptation in biologically plausible spiking neural networks
As we navigate the world, multimodal sensory data streams are continuously parsed,in order to isolate and attend to salient and invariant features, upon which higherorder cortical networks will operate, by evaluating the dynamic relations between suchstructural elements [1]. The formation of stable...
Saved in:
Personal Name(s): | Duarte, Renato (Corresponding Author) |
---|---|
Morrison, Abigail / Series, Peggy | |
Contributing Institute: |
Computational and Systems Neuroscience; IAS-6 Computational and Systems Neuroscience; INM-6 |
Published in: | 2014 |
Imprint: |
2014
|
Conference: | 10th Bernstein Conference, Göttingen (Germany), 2014-09-02 - 2014-09-05 |
Document Type: |
Poster |
Research Program: |
W2/W3 Professorinnen Programm der Helmholtzgemeinschaft (Dys-)function and Plasticity ohne Topic |
Publikationsportal JuSER |
As we navigate the world, multimodal sensory data streams are continuously parsed,in order to isolate and attend to salient and invariant features, upon which higherorder cortical networks will operate, by evaluating the dynamic relations between suchstructural elements [1]. The formation of stable representations of spatial/spectralenvironmental features (stimulus selectivity) along with the related ability to discriminatesuch features and their combinations is known to be continuously shaped by synapticplasticity mechanisms, and it has been recently demonstrated that correlation-basedinhibitory plasticity has an important role to play in such computations [2]. However,in order to adequately process information, neural circuits must not only developstable internal representations of perceptual objects, but also reflect and represent thecontinuous unfolding structure of its input, which is poised with intricate temporaldependencies. Much less is currently known about the acquisition of complex temporalrelations between stimuli and the (possibly specialized) role played by different adaptationmechanisms involved in this process. In this work, we study the properties of biologicallyrealistic networks of LIF neurons, with differentially modulated, dynamic excitation andinhibition [2, 3]. Input-specific neuronal assemblies are driven by stimulus sequencesthat contain complex temporal dependencies and signal propagation throughout theseassemblies is gated by transient disruptions of E/I balance, in order to ’prime’ thenetwork to learn the underlying transitional probabilities and input statistics throughtargeted modifications. We explore the representational properties developed by thesenetworks and the impact of the different plasticity rules in shaping the network’s learningabilities while maintaining stable global dynamics. Furthermore, we assess the network’sability to extract complex temporal dependency rules between sequence elements. |