Speaker: Jun-nosuke Teramae (Kyoto Univ.)

Date and time: Sep. 9th, 15:30- 

Style: Zoom

Ttile: The cortical critical power law balances energy and information in an optimal fashion

Abstract: How neurons in the brain encode sensory information, such as visual images, has long been a central question in neuroscience. To maintain robustness against noise, neural representations are widely thought to avoid the so-called fractal state, where responses become non-differentiable and highly sensitive to input perturbations. However, by deriving an analytical expression for the Fisher information of population coding, we show that the representation is far more robust than previously thought. Counterintuitively, the population representation does not degrade even in highly sensitive regimes, due to its intrinsically high-dimensional structure. With this result, we show that the trade-off between energy consumption and coding efficiency leads to the critical power law, a recently discovered remarkable feature of population responses, as the optimal neural encoding of sensory information.