Democratic vs Non-Democratic Decision Making in Neural Networks
Author Information
Author(s): Richmond Paul, Buesing Lars, Giugliano Michele, Vasilaki Eleni
Primary Institution: University of Sheffield
Hypothesis
How do different neural network architectures affect learning performance in a navigation task?
Conclusion
Networks using a democratic decision-making process learn better and more robustly than those using a non-democratic process.
Supporting Evidence
- The study demonstrated that strong recurrent connections lead to poorer learning outcomes.
- Using GPU programming allowed for faster simulations and exploration of parameter spaces.
- Democratic decision-making mechanisms showed resilience against noise and crosstalk.
Takeaway
This study shows that when neurons work together without influencing each other, they make better decisions than when they affect each other's choices.
Methodology
The study used GPU simulations to model two-layer networks of spiking neurons and tested their learning performance on a navigation task.
Potential Biases
Potential biases may arise from the specific architectures and parameters chosen for the simulations.
Limitations
The study's findings may not generalize to more complex neural architectures or real biological systems.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website