An efficient coding hypothesis links sparsity and selectivity of neural responses.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Sensory Processing

Background:

• The efficient coding hypothesis posits that neurons minimize spikes for faithful stimulus representation.
• Sparsely firing neurons in higher brain areas often respond more to familiar than unfamiliar stimuli, seemingly contradicting this hypothesis.

Purpose of the Study:

• To reconcile the efficient coding hypothesis with observed neuronal responses to familiar and unfamiliar stimuli.
• To investigate how stimulus-independent firing thresholds and input balance influence neural selectivity.

Main Methods:

• Developed a computational model with a cost function penalizing suprathreshold synaptic currents linearly and subthreshold currents quadratically.
• Trained model neurons on familiar (bird's own song - BOS) and unfamiliar (conspecific songs - CONs) stimuli.
• Analyzed how firing threshold and inhibition affect neuronal response selectivity.

Main Results:

• Model neurons exhibited stimulus selectivity dependent on firing threshold.
• At low thresholds, neurons fired densely and preferred CONs/reversed BOS (REV) over BOS.
• At high thresholds or with hyperpolarization, neurons fired sparsely and preferred BOS over REV and CONs.

Conclusions:

• Preference for familiar stimuli (BOS) correlates with high-threshold or strong-inhibition regimes in efficient coding.
• The brain may possess simple mechanisms to dynamically switch neural response selectivity between familiar and nonfamiliar stimuli.
• Findings support the efficient coding hypothesis and elucidate the relationship between neural sparsity and selectivity.