How multisensory neurons solve causal inference.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Perception

Background:

• Illusory self-motion occurs when conflicting visual and vestibular cues arise, such as when a stationary train occupant perceives motion as another train departs.
• The brain must perform causal inference to determine if motion signals have a single cause (self-motion) or multiple causes (external motion).
• The medial superior temporal area in macaques contains neurons responding to congruent or opposite visual and vestibular motion, but the function of opposite neurons is unclear.

Purpose of the Study:

• To explain the computational role of "opposite" neurons in motion perception.
• To model the brain's causal inference process for estimating self-motion using visual and vestibular cues.
• To elucidate how neural representations contribute to integrating or separating sensory information.

Main Methods:

• Training a feedforward neural network model to perform causal inference for motion estimation.
• Simulating the integration and separation of visual and vestibular motion cues within the model.
• Analyzing the emergent properties of neural units, including "congruent" and "opposite" types.

Main Results:

• The neural network model spontaneously developed both congruent and opposite units, mirroring biological findings.
• Both congruent and opposite units were found to contribute to motion estimation.
• The balance of activity between congruent and opposite units determines whether sensory cues are integrated or separated.

Conclusions:

• The balance between congruent and opposite neural activity is crucial for solving causal inference in motion perception.
• This model provides a computational explanation for the existence and function of "opposite" neurons.
• A simple feedforward network architecture can effectively solve complex causal inference problems in sensory processing.