Multiple bumps can enhance robustness to noise in continuous attractor networks

Area of Science:

• Computational neuroscience
• Neural network modeling

Background:

• Continuous attractor networks (CANs) are crucial for path integration and coordinate encoding.
• Brain regions like those with grid cells and head direction cells utilize CANs for spatial and orientational representations.
• The impact of varying the number of activity bumps in these networks on noise resilience is not well understood.

Purpose of the Study:

• To investigate how the number of activity bumps in 1D ring attractor networks affects their robustness to different types of noise.
• To determine if bump number influences the accuracy of encoding linear versus circular coordinates.

Main Methods:

• Constructed 1D ring attractor networks with varying numbers of activity bumps.
• Simulated network responses to three noise types: fluctuating inputs, spiking noise, and connectivity deviations.
• Analyzed the impact of bump number on coordinate encoding accuracy for both linear and circular representations.

Main Results:

• Networks with more bumps exhibited significantly less noise-driven deviation in activity bump motion across all tested noise types.
• Increased bump numbers led to more robust encoding of linear coordinates (e.g., position).
• Bump number had no significant effect on the error in readout for circular coordinates (e.g., orientation).

Conclusions:

• Continuous attractor networks can enhance robustness to noise for linear coordinate encoding by utilizing a higher number of activity bumps.
• This advantage is diminished when encoding circular coordinates, where bump number is less critical for accuracy.
• Findings support the potential role of multiple bumps in mammalian grid cell networks for improved spatial representation.