Die Rolle von Interneuronen in der Plastizität von sensorischen Representationen während Belohnungs-basierten Lernens
Zusammenfassung der Projektergebnisse
It is becoming increasingly clear that inhibitory interneurons play a more important role in cortical computation than was previously assumed. Experimental studies revealed that inhibitory microcircuits in diverse cortical areas are modulated by behaviourally relevant signals, that they are needed for learning, and that inhibition is plastic. The functional roles of inhibitory interneurons in microcircuits is yet unclear and difficult to infer from the experimental data alone. As a result of this project, I presented a hypothesis for a functional role of the plastic inhibitory microcircuit during learning from brief contextual signals. I proposed that learning happens in two stages: 1) unspecific top-down signals rapidly induce an inhibitory connectivity structure without changing the excitatory connectivity. 2) the inhibitory network induces functional changes in neural representations by guiding excitatory plasticity. To test this, I developed biologically constrained spiking (integrate-and-fire) models of layer 2/3 primary visual cortex. The models consisted of excitatory pyramidal cells (PC), and different interneuron populations, corresponding to somatostatin (SST)-positive, parvalbumin (PV)-positive and vasoactive intestinal peptide (VIP)-expressing interneuron types. Excitatory and SST-to-PV connections exhibited Hebbian spike-timing dependent plasticity. We simulated a rewarded phase in which the presentation of one stimulus is paired with a reward signal, which excites VIPs. We then simulated a second refinement phase, where the sensory stimuli without the reward were presented. I showed that the mechanism can be implemented in spiking models of primary visual microcircuits with different interneuron types. In particular, a memory of the rewarded stimulus can indeed be stored in the inhibitory structure. This structure can instruct excitatory plasticity in the absence of reward via a disinhibition mechanism. The PCs then increase their tuning to the rewarded stimulus because they receive strong connections from PCs coding for the rewarded stimulus, regardless of their initial tuning. An unspecific top-down reward signal is sufficient to create a specific circuit structure owing to the temporal coincidence between reward signals and stimulus-evoked activity. Second, the proposed mechanism turns out to also ensure stability of stimulus representations (in excitatory neurons) during behavioural engagement. Third, the models provide precise testable predictions in terms of neural activity. Finally, I suggest that the presented mechanism could provide a biologically plausible solution to weight sharing, a method to achieve better learning performance by generalisation and translation invariance in deep neural networks. The hypothesis challenges the traditional view that does not consider inhibitory neurons as part of memories. It further proposes distinct but complementary roles of inhibitory and excitatory connectivity in learning and memory.
Projektbezogene Publikationen (Auswahl)
- Exploring the role of interneurons in sensory representation during reward learning. BCCN Conference 2017, Göttingen, Germany
K. A. Wilmes and C. Clopath
- Interneuron-mediated top-down plasticity of sensory representations. BCCN Conference 2018, Berlin, Germany
K. A. Wilmes and C. Clopath
- (2019). Inhibitory microcircuits for top-down plasticity of sensory representations. Nature Communications, 10(1):5055
Wilmes, K. A. and Clopath, C.
(Siehe online unter https://doi.org/10.1038/s41467-019-12972-2) - Inhibitory microcircuits for top-down plasticity of sensory representations. Cosyne 2019, Lisbon, Portugal
K. A. Wilmes and C. Clopath
- (2020). Stability by gating plasticity in recurrent neural networks. bioRxiv
Wilmes, K. A. and Clopath, C.
(Siehe online unter https://doi.org/10.1101/2020.09.10.291120)
