On the Subspace Invariance of Population Responses.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Visual System

Background:

• Neural population responses to complex stimuli are often modeled using principles like subspace invariance.
• Subspace invariance suggests that responses to combined stimuli can be predicted from responses to individual stimuli.

Purpose of the Study:

• To investigate whether subspace invariance holds in the mouse primary visual cortex.
• To determine if contrast invariance, a specific form of subspace invariance, is present in mouse visual cortex.

Main Methods:

• Measuring neural population responses to sinusoidal gratings and plaids (linear combinations of gratings).
• Quantifying the geometric relationship between plaid responses and the plane spanned by individual grating responses.
• Analyzing neuronal responses across varying contrast levels.

Main Results:

• Robust violations of subspace invariance were observed in the mouse primary visual cortex.
• A strong negative correlation was found between neuronal responses to individual gratings and their responses to plaids.
• Contrast invariance also failed, with some neurons showing decreased and others increased responses as contrast rose.

Conclusions:

• Subspace and contrast invariance do not apply to neural population coding in the mouse primary visual cortex.
• These findings exclude specific models of population coding, such as vector averaging and certain normalization models.