Stimulus-dependent maximum entropy models of neural population codes.

Area of Science:

• Computational neuroscience
• Neural coding and information theory

Background:

• Neural populations encode stimuli through collective activity patterns, including spiking and silence.
• Modeling large neural populations requires understanding the conditional probability distribution of neural codewords given sensory input.

Purpose of the Study:

• To investigate the role of inter-neuronal dependencies in neural encoding.
• To introduce and validate a novel model for neural population activity.

Main Methods:

• Developed the stimulus-dependent maximum entropy (SDME) model, extending single-neuron models to pairwise-coupled populations.
• Utilized temporal white-noise stimuli to record activity from 100 salamander retinal ganglion cells.
• Compared the SDME model against uncoupled models for accuracy in predicting neural responses.

Main Results:

• Inter-neuronal dependencies significantly contribute to the encoding role in neural populations.
• The SDME model provides a more accurate account of single-cell responses compared to uncoupled models.
• SDME models significantly outperform uncoupled models in reproducing population codeword distributions.

Conclusions:

• The SDME model effectively captures stimulus-dependent neural population dynamics.
• The SDME model, combined with maximum entropy principles, enables estimation of information-theoretic quantities like information transmission.