Neural Decision Boundaries for Maximal Information Transmission
Author Information
Author(s): Tatyana Sharpee, William Bialek
Primary Institution: The Salk Institute for Biological Studies
Hypothesis
How can multidimensional signals be optimally separated into two categories to maximize information transmission?
Conclusion
The study finds that optimal decision boundaries for non-Gaussian inputs are curved, while Gaussian inputs are best separated by planar boundaries.
Supporting Evidence
- Gaussian inputs are optimally separated by hyper-planes.
- For exponentially distributed inputs, optimal decision contours are curved.
- Closed contours are optimal at extreme probabilities, while extended ones are optimal for spike probabilities near 1/2.
Takeaway
This study looks at how neurons decide when to fire based on different types of signals, showing that the best way to separate these signals can change depending on their characteristics.
Methodology
The authors derived equations for decision boundaries based on the probability distributions of inputs and analyzed specific solutions for Gaussian and exponential distributions.
Limitations
The study focuses on a simplified model and does not account for complex coding strategies involving multiple neurons or temporal patterns.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website