Physics of neural computation and the critical brain

6 problems

• Do neural avalanche exponents obey critical scaling relations?

  Contested

  Bursts of neural activity, called avalanches, have sizes and durations distributed as power laws. In a true critical system the exponents of these laws must satisfy a fixed relation, which can separate criticality from look-alikes.

• Do real circuits store memories at the theoretical capacity?

  Partially resolved

  Theory predicts how many patterns a network of neurons can store, and that at maximum storage most synapses should be switched off. Testing whether real circuits run at this limit means matching measured synapse statistics to the theory.

• Universality of neural population coarse-graining exponents

  Partially resolved

  Grouping correlated neurons step by step, as physicists do near phase transitions, gives scaling exponents in mouse brain recordings. Whether these numbers are the same across brain regions and animals, as a true fixed point would require, is open.

• Is the brain poised near a critical point?

  Contested

  Power-law statistics in brain activity suggest that neural networks operate at a phase transition. Other explanations, such as shared slowly varying inputs, can produce the same statistics without any tuning.

• Origin of the 1/n spectrum of cortical population responses

  Open

  When thousands of visual-cortex neurons respond to natural images, the variance carried by the n-th strongest activity pattern falls roughly as 1/n. This is close to the slowest decay, meaning the most high-dimensional code, that still keeps the response a smooth function of the image, and which circuit mechanism sets it is unknown.

• How stable memories survive drifting neural representations

  Open

  Recordings over weeks show that the neurons responding to a given place or task keep changing, while the animal's behavior stays the same. How the brain reads a stable meaning from a code that keeps changing is unknown.