Sparseness and expansion in sensory representations

Area of Science:

• Computational neuroscience
• Neural coding
• Sensory processing

Background:

• Sensory pathways often feature sparse downstream neuronal populations with more neurons than incoming axons.
• Understanding the computational advantages of this expansion and sparseness for clustered inputs is crucial.

Purpose of the Study:

• To investigate the computational benefits of expansion and sparseness in neural representations for clustered inputs.
• To analyze how variability in sensory stimuli and neuronal noise affects these representations.

Main Methods:

• Analytical calculations and numerical simulations were employed.
• The study modeled clustered inputs with intracluster variability representing noise.
• The impact of feed-forward random synaptic weights and structured synapses was examined.

Main Results:

• Expansion via random synapses amplifies stimulus variability, with noise increasing with representation sparseness.
• Low input dimensionality leads to stimulus representation overlaps, limiting expansion benefits.
• Highly sparse, structured expansive representations reduce variability and overlaps, improving downstream task performance.

Conclusions:

• Sparse, structured neural representations are critical for efficient sensory processing.
• These representations enhance downstream neurons' ability to classify and recognize distinct stimuli.
• Findings have implications for understanding olfactory, cerebellar, and visual processing.