Toward a unified theory of efficient, predictive, and sparse coding.

Area of Science:

• Theoretical Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Predicting sensory neuron responses from first principles is a central goal in theoretical neuroscience.
• Efficient coding theories propose neurons maximize information transmission under constraints, but many variants exist (e.g., redundancy reduction, predictive coding, sparse coding).
• The relationship between these coding variants and their combined effects remains unclear.

Purpose of the Study:

• To present a unified framework for efficient coding models in neuroscience.
• To explore how different efficient coding objectives influence neural response properties.
• To explain the diversity of sensory neural responses through multiple functional goals and constraints.

Main Methods:

• Developed a unified mathematical framework encompassing various efficient coding models.
• Analyzed neural responses under different coding objectives (e.g., encoding predictive information, recovering the past, predicting the future).
• Investigated coding of naturalistic visual stimuli, including movies.

Main Results:

• The unified framework integrates previously proposed efficient coding models and extends to novel regimes.
• Optimizing neural responses for predictive information can lead to input correlation or decorrelation, depending on stimulus statistics.
• Efficiently encoding the past, particularly at low noise, consistently predicts decorrelation of neural inputs.
• Coding naturalistic movies predicts distinct visual motion tuning and response sparsity patterns based on whether the objective is past recovery or future prediction.

Conclusions:

• The proposed unified framework offers a way to reconcile diverse efficient coding theories.
• Different functional goals (e.g., past vs. future coding) and constraints can explain the observed variety in sensory neuron response properties.
• This approach provides a basis for understanding how different cell types and circuits fulfill specific coding objectives.