Unconscious Biases in Neural Populations Coding Multiple Stimuli.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Information Theory

Background:

• Neural information is often encoded in distributed activity across neuronal populations.
• Homogeneous population codes for single stimuli allow unbiased readout.
• Overlapping neural representations of multiple stimuli present a challenge for accurate information decoding.

Discussion:

• Overlapping neural representations of multiple stimuli introduce significant readout bias.
• This bias persists even at low noise levels, challenging idealized models.
• The bias originates from the complex interplay of overlapping neural representations.

Key Insights:

• A novel Gaussian process framework accurately models decoder estimate distributions.
• Overlapping stimuli lead to bimodal estimate distributions, explaining the readout bias.
• Competitive encoding and complex decoders can reduce the emergent bias.

Outlook:

• Findings impact understanding of neural coding for complex sensory inputs.
• Implications for designing more robust neural decoding algorithms.
• Informs experimental design in neuroscience, particularly for motion perception studies.